THE LIBRARY
OF
THE UNIVERSITY
OF CALIFORNIA
LOS ANGELES
SCIENTIFIC PAPEES
Eonbon: C. J. CLAY AND SONS,
CAMBRIDGE UNIVERSITY PRESS WAREHOUSE,
AVE MARIA LANE.
©laaaoto: 30, WELLINGTON STREET.
F. A. BROCKHAUS.
lorfe: THE MACMILLAN COMPANY.
anto Calcutta: MACMILLAN AND CO., LTD.
[All Rights reserved.]
SCIENTIFIC PAPEBS
BY
JOHN WILLIAM STEUTT,
BARON RAYLEIGH,
O.M., D.Sc., F.R.S.,
HONORARY FELLOW OF TRINITY COLLEGE, CAMBRIDGE,
PROFESSOR OF NATURAL PHILOSOPHY IN THE ROYAL INSTITUTION.
VOL. IV.
1892-1901.
CAMBRIDGE:
AT THE UNIVERSITY PRESS.
1903
dambrtogc:
PRINTED BY J. AND C. F. CLAY,
AT THE UNIVERSITY PRESS.
Engineering
Library
(ZO
PBEFACE.
the present volume the Collection of Papers is brought down to
the end of 1901. The diversity of subjects — many of them, it is to
be feared, treated in a rather fragmentary manner — is as apparent as ever,
and is perhaps intensified by the occurrence of papers recording experi-
mental work on gases. The memoir on Argon (Art. 214) by Sir W. Ramsay
and myself is included by special permission of my colleague.
A Classified Table of Contents and an Index of Names are appended.
The large number of references to the works of Sir George Stokes,
Lord Kelvin and Maxwell, as well as of Helmholtz and some other
investigators abroad, will shew to whom I have been most indebted for
inspiration.
I desire also to record my obligations to the Syndics and Staff of the
University Press for the efficient and ever courteous manner in which they
have carried out my wishes in the republication of this long series of
TERLING PLACE, WITHAM,
December 1902.
782201
The works of the Lord are great,
Sought out of all them that have pleasure therein.
CONTENTS.
ART. PAGE
197. Density of Nitrogen 1
[Nature, XLVI. pp. 512, 513, 1892.]
198. On the Intensity of Light reflected from Water and Mercury
at nearly Perpendicular Incidence ..... 3
Appendix .......... 13
[Philosophical Magazine, xxxiv. pp. 309 — 320, 1892.]
199. On the Interference Bands of Approximately Homogeneous
Light; in a Letter to Prof. A. Michelson . . . . 15
[Philosophical Magazine, xxxiv. pp. 407 — 411, 1892.]
200. On the Influence of Obstacles arranged in Rectangular Order
upon the Properties of a Medium 19
[Philosophical Magazine, xxxiv. pp. 481—502, 1892.]
201. On the Densities of the Principal Gases ..... 39
The Manometer 40
Connexions with Pump and Manometer .... 43
The Weights . . ... . . . . . . 44
The Water Contents of the Globe . . . . . 45
Air . . .. . .... . . . 46
Oxygen . . . .... . . . 47
Nitrogen ... . ... . . . 48
Reduction to Standard Pressure ..... 50
Note A. On the Establishment of Equilibrium of Pressure
in Two Vessels connected by a Constricted Channel . 53
[Proceedings of the Royal Society, LIII. pp. 134—149, 1893.]
202. Interference Bands and their Applications .... 54
[Proc. Roy. Inst. xiv. pp. 72—78, 1893 ; Nature, XLVIII. pp. 212—214, 1893.]
203. On the Theory of Stellar Scintillation . ... . 60
[Philosophical Magazine, xx xvi, pp. 129—142, 1893.]
204. Astronomical Photography . ... ,. «n * ..}?.. 73
[Nature, XLVIII. p. 391, 1893.)
Vlll CONTENTS.
ART. PAGE
205. Grinding and Polishing of Glass Surfaces .... 74
[British Association, Sept. 14, 1893, from a report in Nature, XLVIII. p. 526, 1893.]
206. On the Reflection of Sound or Light from a Corrugated
Surface 75
[British Association Report, pp. 690, 691, 1893.]
207. On a Simple Interference Arrangement 76
[British Association Report, pp. 703, 704, 1893.]
208. On the Flow of Viscous Liquids, especially in Two Dimensions 78
[Philosophical Magazine, xxxvi. pp. 354—372, 1893.]
209. The Scientific Work of Tyndall 94
[Proceedings of the Royal Institution, xiv. pp. 216—224, 1894.]
210. On an Anomaly encountered in Determinations of the Density
of Nitrogen Gas '.""'. .104
[Proceedings of the Royal Society, LV. pp. 340—344, April, 1894.]
211. On the Minimum Current audible in the Telephone . . 109
[Philosophical Magazine, xxxvm. pp. 285 — 295, 1894.]
212. An Attempt at a Quantitative Theory of the Telephone . 119
[Philosophical Magazine, xxxvm. pp. 295 — 301, 1894.]
213. On the Amplitude of Aerial Waves which are but just Audible 125
[Philosophical Magazine, xxxvm. pp. 365 — 370, 1894.]
214. Argon, a New Constituent of the Atmosphere. By LOKJ>
RAYLEIGH, Sec. R.S., and PROFESSOR WILLIAM RAMSAY, F.R.S. 130
1. Density of Nitrogen from Various Sources . . .130
2. Reasons for Suspecting a hitherto Undiscovered Con-
stituent in Air . . . . . . .135
3. Methods of Causing Free Nitrogen to Combine . . 138
4. Early Experiments on sparking Nitrogen with Oxygen
in presence of Alkali . . . . . .141
5. Early Experiments on Withdrawal of Nitrogen from
Air by means of Red-hot Magnesium . . .144
6. Proof of the Presence of Argon in Air, by means
of Atmolysis . . . . . . . . 150
7. Negative Experiments to prove that Argon is not
derived from Nitrogen or from Chemical Sources . 153
8. Separation of Argon on a Large Scale . . .155
9. Density of Argon prepared by means of Oxygen . 105
10. Density of Argon prepared by means of Magnesium . 167
11. Spectrum of Argon . . . . . . 168
12. Solubility of Argon in Water 170
CONTENTS. ix
ART. PAGE
13. Behaviour at Low Temperatures. . . . „ . 173
14. The ratio of the Specific Heats of Argon . . . 174
15. Attempts to induce Chemical Combination . . 176
16. General Conclusions . . . . . . .180
Addendum, March 20 (by PROFESSOR W. RAMSAY) . . 184
Addendum, April 9 . . . . . . . .187
[Phil. Trans. 186 A, pp. 187—241, 1895.]
215. Argon 188
[Royal Institution Proceedings, xiv. pp. 524 — 538, April 1895.]
216. On the Stability or Instability of Certain Fluid Motions. III. 203
Addendum, January 1896 .209
[Proceedings of the London Mathematical Society, xxvu. pp. 5—12, 1895.]
217. On the Propagation of Waves upon the Plane Surface separ-
ating Two Portions of Fluid of Different Vorticities . . 210
[Proceedings of the London Mathematical Society, xxvu. pp. 13 — 18, 1895.]
218. On some Physical Properties of Argon and Helium . . 215
Density of Argon . . . . . . .215
The Refractivity of Argon and Helium .... 218
Viscosity of Argon and Helium ..... 222
Gas from the Bath Springs ...... 223
Buxton Gas 223
Is Helium contained in the Atmosphere ? . . . 224
[Proceedings of the Royal Society, LIX. pp. 198—208, Jan. 1896.]
219. On the Amount of Argon and Helium contained in the Gas
from the Bath Springs . . . . ... . 225
[Proceedings of -the Royal Society, LX. pp. 56, 57, 1896.]
220. The Reproduction of Diffraction Gratings .... 226
[Nature, mv. pp. 332, 333, 1896.]
221. The Electrical Resistance of Alloys 232
[Nature, LIV. pp. 154, 155, 1896.]
222. On the Theory of Optical Images, with Special Reference
to the Microscope ......... 235
[Philosophical Magazine, XLII. pp. 167 — 195, 1896.]
223. Theoretical Considerations respecting the. Separation of Gases
by Diffusion and Similar Processes - . • •'.' . . . 261
[Philosophical Magazine, XLII. pp. 493—498, 1896.]
224. The Theory of Solutions . ..--.• ; . ••" : •."•:•• . ' . . 267
[Nature, T,V. pp. 253, 254, 1897.]
x CONTENTS.
AET. PAGE
225. Observations on the Oxidation of Nitrogen Gas . : . • . 270
[Chemical Society's Journal, 71, pp. 181 — 186, 1897.]
226. On the Passage of Electric Waves through Tubes, or the
Vibrations of Dielectric Cylinders . . . . . 276
General Analytical Investigation . . . . .276
Rectangular Section 279
Circular Section. ........ 280
[Philosophical Magazine, XLIII. pp. 125—132, 1897.]
227. On the Passage of Waves through Apertures in Plane Screens,
and Allied Problems 283
Perforated Screen. — Boundary Condition d<f>/dn = 0 . . 284
Boundary Condition <£ = 0 . . . . ; . 286
Reflecting Plate.— d<j>/dn = 0 288
Reflecting Plate.— 0 = 0 289
Two-dimensional Vibrations ...... 290
Narrow Slit. — Boundary Condition d<j>/dn = 0 . . . 291
Narrow Slit.— Boundary Condition 0=0 . . . .293
Reflecting Blade.— Boundary Condition d<j>/dn = 0 . . 294
Reflecting Blade. — Boundary Condition <f> = 0 . . . 295
Various Applications ........ 295
[Philosophical Magazine, XLIII. pp. 259 — 272, 1897.]
228. The Limits of Audition 297
[Royal Institution Proceedings, xv. pp. 417—418, 1897.]
229. On the Measurement of Alternate Currents by means of
an obliquely situated Galvanometer Needle, with a Method
of Determining the Angle of Lag . . . . . 299
[Philosophical Magazine, XLIII. pp. 343—349, 1897.]
230. On the Incidence of Aerial and Electric Waves upon Small
Obstacles in the Form of Ellipsoids or Elliptic Cylinders,
and on the Passage of Electric Waves through a Circular
Aperture in a Conducting Screen , . . . 305
Obstacle in a Uniform Field ... . . . . 306
In Two Dimensions . . . .... . . 309
Aerial Waves . . ... . . . . . 310
Waves in Two Dimensions . . . . . .314
Electrical Applications . . . . .^ . .317
Electric Waves in Three Dimensions . . . .318
Obstacle in the Form of an Ellipsoid .... 323
Circular Aperture in Conducting Screen .... 324
[Philosophical Magazine, XLIV. pp. 28—52, 1897.]
CONTENTS. xi
ART. PAGE
231. On the Propagation of Electric Waves along Cylindrical
Conductors of any Section . . ... . . 327
[Philosophical Magazine, XLIV. pp. 199 — 204, 1897.]
232. The Electro-Chemical Equivalent of Silver •„, . . . 332
[Nature, LVI. p. 292, 1897.]
233. On an Optical Device for the Intensification of Photographic
Pictures 333
[Philosophical Magazine, XLIV. pp. 282—285, 1897.]
234. On the Viscosity of Hydrogen as affected by Moisture . . 336
[Proceedings of the Royal Society, LXII. pp. 112 — 116, 1897.]
235. On the Propagation of Waves along Connected Systems of
Similar Bodies . . . . . '. . . 340
[Philosophical Magazine, XLIV. pp. 356—362, 1897.]
236. On the Densities of Carbonic Oxide, Carbonic Anhydride,
and Nitrous Oxide . . ...... . . 347
Carbonic Oxide . . . . . . ... .- 347
Carbonic Anhydride 349
Nitrous Oxide ......... 350
[Proceedings of the Royal Society, LXII. pp. 204—209, 1897.]
237. Rontgen Rays and Ordinary Light 353
[Nature, LVII. p. 607, 1898.]
238. Note on the Pressure of Radiation, showing an Apparent
Failure of the Usual Electromagnetic Equations . . . 354
[Philosophical Magazine, XLV. pp. 522—525, 1898.]
239. Some Experiments with the Telephone . . . . .357
[Roy. Inst. Proc. xv. pp. 786—789, 1898 ; Nature, LVIII. pp. 429—430, 1898.]
240. Liquid Air at one Operation 360
[Nature, LVIII. p. 199, 1898.]
241. On the Character of the Impurity found in Nitrogen Gas
Derived from Urea [with an Appendix containing details
of Refractometer] 361
Details of Refractometer 364
[Proceedings of the Royal Society, LXIV. pp. 95 — 100, -1898.]
242. On Iso-periodic Systems . . . . . . . " . 367
[Philosophical Magazine, XLVI. pp. 567—569, 1898.]
243. On James Bernoulli's Theorem in Probabilities . •. . 370
Magazine, XLVII. pp. 246—251, 1899.]
Xll CONTENTS.
ART. PAGE
244. On the Cooling of Air by Radiation and Conduction, and
on the Propagation of Sound . . . . . .376
[Philosophical Magazine, XLVII. pp. 308—314, 1899.]
'245. On the Conduction of Heat in a Spherical Mass of Air
confined by Walls at a Constant Temperature . . . 382
[Philosophical Magazine, XLVII. pp. 314—325, 1899.]
246. Transparency and Opacity -•..-... 392
[Proc. Roy. Inst. xvi. pp. 116—119, 1899; Nature, LX. pp. 64, 65, 1899.]
247. On the Transmission of Light through an Atmosphere con-
taining Small Particles in Suspension, and on the Origin
of the Blue of the Sky . . . . - . . . . 397
[Philosophical Magazine, XLVII. pp. 375—384, 1899.]
248. The Interferometer 406
[Nature, LIX. p. 533, 1899.]
249. On the Calculation of the Frequency of Vibration of a System
in its Gravest Mode, with an Example from Hydrodynamics 407
[Philosophical Magazine, XLVII. pp. 566 — 572, 1899.]
250. The Theory of Anomalous Dispersion ..... 413
[Philosophical Magazine, XLVIII. pp. 151, 152, 1899.]
251. Investigations in Capillarity . . . . . . .415
The Size of Drops 415
The Liberation of Gas from Supersaturated Solutions . 420
Colliding Jets 421
The Tension of Contaminated Water-Surfaces . . .425
A Curious Observation 430
[Philosophical Magazine, XLVIII. pp. 321—337, 1899.]
252. The Mutual Induction of Coaxial Helices . . . .431
[British Association Report, pp. 241, 242, 1899.]
253. The Law of Partition of Kinetic Energy -. w . . 433
[Philosophical Magazine, XLIX. pp. 98—118, 1900.]
254. On the Ariscosity of Argon as affected by Temperature . . 452
[Proceedings of the Royal Society, LXVI. pp. 68—74, 1900.]
255. On the Passage of Argon through Thin Films of Indiarubber 459
[Philosophical Magazine, XLIX. pp. 220, 221, 1900.]
256. On the Weight of Hydrogen desiccated by Liquid Air . . 461
[Proceedings of the Royal Society, LXVI. p. 344, 1900.]
CONTENTS. xiii
ART. PAGE
257. The Mechanical Principles of Flight . . ^ '. • :. 462
[Manchester Memoirs, XLIV. pp. 1 — 26, 1900.]
258. On the Law of Reciprocity in Diffuse Reflexion . . . 480
[Philosophical Magazine, XLIX. pp. 324, 325, 1900.]
259. On the Viscosity of Gases as Affected by Temperature . 481
[Proceedings of the Royal Society, LXVII. pp. 137—139, 1900.]
260. Remarks upon the Law of Complete Radiation . . . 483
[Philosophical Magazine, XLIX. pp. 539, 540, 1900.]
261. On Approximately Simple Waves ...... 486
[Philosophical Magazine, L. pp. 135 — 139, 1900.]
262. On a Theorem analogous to the Virial Theorem . . . 491
[Philosophical Magazine, L. pp. 210—213, 1900.]
263. On Balfour Stewart's Theory of the Connexion between
Radiation and Absorption ....... 494
[Philosophical Magazine, i. pp. 98—100, 1901.]
264. Spectroscopic Notes concerning the Gases of the Atmosphere 496
On the Visibility of Hydrogen in Air .... 496
Demonstration at Atmospheric Pressure of Argon from
very small quantities of Air. ..... 499
Concentration of Helium from the Atmosphere . . 500
[Philosophical Magazine, I. pp. 100—105, 1901.]
265. On the Stresses in Solid Bodies due to Unequal Heating,
and on the Double Refraction resulting therefrom . . 502
[Philosophical Magazine, I. pp. 169—178, 1901.]
266. On a New Manometer, and on the Law of the Pressure of
Gases between 1'5 and O'Ol Millimetres of Mercury . . 511
Introduction . . . . . . . . .511
Improved Apparatus for Measuring very small Pressures. 514
Experiments to determine the Relation of Pressure and
Volume at given Temperature 519
[Philosophical Transactions, cxcvi A. pp. 205 — 223, 1901.]
267. On a Problem relating to the Propagation of Sound between
Parallel Walls 532
[Philosophical Magazine, I. pp. 301 — 311, 1901.]
268. Polish . 542
[Proceedings of the Royal Institution, xvi. pp. 563 — 570, 1901 ;
Nature, LXIV. pp. 385—388, 1901.]
XIV CONTENTS.
ART. PAGE
269. Does Chemical Transformation influence Weight t . . . 549
[Nature, LXIV. p. 181, June, 1901.]
270. Acoustical Notes. VI. . . ... . . . 550
Forced Vibrations . 550
Vibrations of Strings . . . . . . .551
Beats of Sounds led to the Two Ears separately . . 553
Loudness of Double Sounds . . . .. . . 554
[Philosophical Magazine, u. pp. 280—285, 1901.]
271. On the Magnetic Rotation of Light and the Second Law
of Thermodynamics . ... . . . . 555
[Nature, LXIV. pp. 577, 578, 1901.]
272. On the Induction-Coil . . . . ... . . 557
[Philosophical Magazine, u. pp. 581—594, 1901.]
CLASSIFIED TABLE OF CONTENTS 569
INDEX OF NAMES .... 599
ILLUSTRATIONS.
Portrait of LORD RAYLEIGH . . . . Frontispiece
Plate I (Figs. 1 and 2) . . . . . To face p. 545
Plate II (Figs. 3 and 4) ... . . 548
197.
DENSITY OF NITROGEN.
[Nature, XLVI. pp. 512, 513, 1892.]
I AM much puzzled by some recent results as to the density of nitrogen,
and shall be obliged if any of your chemical readers can offer suggestions as
to the cause. According to two methods of preparation I obtain quite distinct
values. The relative difference, amounting to about 1/1000 part, is small in
itself; but it lies entirely outside the errors of experiment, and can only be
attributed to a variation in the character of the gas.
In the first method the oxygen of atmospheric air is removed in the
ordinary way by metallic copper, itself reduced by hydrogen from the oxide.
The air, freed from C02 by potash, gives up its oxygen to copper heated in
hard glass over a large Bunsen, and then passes over about a foot of red-hot
copper in a furnace. This tube was used merely as an indicator, and the
copper in it remained bright throughout. The gas then passed through a
wash-bottle containing sulphuric acid, thence again through the furnace
over copper oxide, and finally over sulphuric acid, potash and phosphoric
anhydride.
In the second method of preparation, suggested to me by Prof. Ramsay,
everything remained unchanged, except that the first tube of hot copper was
replaced by a wash-bottle containing liquid ammonia, through which air was
allowed to bubble. The ammonia method is very convenient, but the nitrogen
obtained by means of it was 1/1000 part lighter than the nitrogen of the first
method. The question is, to what is the discrepancy due ?
The first nitrogen would be too heavy, if it contained residual oxygen.
But on this hypothesis, something like 1 per cent, would be required. I
could detect none whatever by means of alkaline pyrogallate. It may be
remarked that the density of the nitrogen agrees closely with that recently
obtained by Leduc using the same method of preparation.
6 7 R. iv. 1
2 DENSITY OF NITROGEN. [197
On the other hand, can the ammonia-made nitrogen be too light from the
presence of impurity ? There are not many gases lighter than nitrogen, and
the absence of hydrogen, ammonia, and water seems to be fully secured. On
the whole it seemed the more probable supposition that the impurity was
hydrogen, which in this degree of dilution escaped the action of the copper
oxide. But a special experiment seems to exclude this explanation.
Into nitrogen prepared by the first method, but before its passage into
the furnace tubes, one or two thousandths by volumes of hydrogen were
introduced. To effect this in a uniform manner the gas was made to bubble
through a small hydrogen generator, which would be set in action under its
own electro-motive force by closing an external contact. The rate of hydrogen
production was determined by a suitable galvanometer enclosed in the
circuit. But the introduction of hydrogen had not the smallest effect upon
the density, showing that the copper oxide was capable of performing the
part desired of it.
Is it possible that the difference is independent of impurity, the nitrogen
itself being to some extent in a different (dissociated) state ?
I ought to have mentioned that during the fillings of the globe, the rate
of passage of gas was very uniform, and about 2/3 litre per hour.
198.
ON THE INTENSITY OF LIGHT REFLECTED FROM WATER
AND MERCURY AT NEARLY PERPENDICULAR INCIDENCE.
[Philosophical Magazine, xxxiv. pp. 309 — 320, 1892.]
IN a former paper* I gave an account of some experiments upon the
reflexion from glass surfaces tending to show that "recently polished glass
surfaces have a reflecting-power differing not more than 1 or 2 per cent, from
that given by Fresnel's formula; but that after some months or years the
reflexion may fall off from 10 to 30 per cent., and that without any apparent
tarnish." Results in the main confirmatory have been published by Sir John
Conroyf.
The accurate comparison of Fresnel's formula with observation is a matter
of great interest from the point of view of optical theory, but it seems scarcely
possible to advance the matter much further in the case of solids. Apart
from contamination with foreign bodies of a greasy nature, and disintegration
under atmospheric influences, we can never be sure that the results are
unaffected by the polishing-powder which it is necessary to employ. For
these reasons I have long thought it desirable to institute experiments
with liquids, of which the surfaces are easily renewed ; and the more since
I succeeded in proving that (in the case of water at any rate) the deviation
from Fresnel's formula found by Jamin in the neighbourhood of the polarizing
angle is due to greasy contamination. The very close verification of the
theoretical formula in this critical case seemed to render its applicability to
perpendicular incidence in a high degree probable. I was thus induced to
attack the somewhat troublesome problem of designing a photometric method
capable of dealing with the reflexion from a horizontal surface. The details
of the apparatus and of the measures will be given presently; but in the
meantime it may be well to consider rather closely what is to be expected
upon the supposition that Fresnel's formulas are really applicable. Fresnel's
formulas are spoken of, because although at strictly perpendicular incidence
we should have to do only with Young's expression (JJL — 1)2/G* -f I)2, in
* Proc. Roy. Soc. November, 1886. [Vol. n. p. 522.]
t Phil. Trans, 1889 A, p. 245.
1—2
4 ON THE INTENSITY OF LIGHT REFLECTED FROM WATER [198
practice we are forced to work at finite angles of incidence. It is thus
important to examine the march of Fresnel's expressions, when the angle of
incidence (0) is small.
Writing
sin (0- ft) tan (0- ft)
sin (0 + ft) ' tan (0 + ft) '
where
sin ft = sin 0//i,
we find
Thus S2 and T2 differ from the value appropriate to 0 = 0 in opposite
directions and by quantities of the order 0*. But on addition we get
differing from the value appropriate to 0 = 0 by a quantity of the fourth order
only in 0. When therefore the circumstances are such that it is unnecessary
to distinguish the two polarized components, the intensity of reflexion at
small incidences is in a high degree independent of the precise angle. If fi is
nearly equal to unity, we have
simply. Again, if p, = |,
(5)
A few calculations from the original expressions will serve to indicate the
field of these approximations.
^ = f, 0 = 10°, ft = 7° 29',
S»--jLx 1-0467, r* = ^x
S* + T2 = 2 x ~ x 1-0004.
4y
From (5) we get as the last factor 1-00050.
/* = f, 0 = 20°, ft = 1
S2 = -!gX 1-2021, T2 =
8* + T2 = 2 x ^ x 1-0090.
By (5) the last factor is T0080.
1892] AND MERCURY AT NEARLY PERPENDICULAR INCIDENCE.
Again,
e = 30°,
= JLX 1-5189,
0! = 22° l'-4,
=4 x -5866,
' = 2x^x 1-0527.
According to (5) the last factor is here 1*0405.
Fig. l.
It appears that in the case of water the aggregate reflexion scarcely
begins to vary sensibly from its value for 0 = 0 until 6 = 20°, a property of
some importance for our present purpose, as it absolves us from the necessity
of striving after very small angles of incidence.
I will now describe the actual arrangement adopted for the experi-
ments. The source of light at A (Fig. 1) is a small incandescent lamp, the
6 ON THE INTENSITY OF LIGHT REFLECTED FROM WATER [198
current through which is controlled with the aid of a galvanometer. It is
so mounted that its equatorial plane coincides with the (vertical) plane of
the diagram. Underneath, upon the floor, is placed the liquid (B) whose
reflecting power is to be examined. At C, just under the roof, the direct
ray AC and the reflected ray BC are turned into the same horizontal
direction by two mirrors silvered in front and meeting one another at C
under a small angle. The eye situated opposite to the edge C and looking
into the double mirror thus sees the direct and reflected images superposed,
so far as the different apparent magnitudes allow. D represents a diaphragm
and E a photographic portrait-lens of about 3 inches aperture which forms
an image of A and A' on or near the plane F. At F is placed a screen
perforated with a hole sufficiently large to make sure of including all the
rays from A, A' which pass D. To determine this point an eye-piece is
focused upon F, so that the images of A, A' are seen nearly in focus. Some
margin is necessary because the images of A, A' cannot (both) be accurately
in focus at F.
These adjustments being made, an eye placed behind F and focused
upon C sees the upper mirror illuminated by the direct light (from A), and
the lower illuminated by the reflected light (from A). And if the aperture
at F is less than that of the pupil of the eye, the apparent brightnesses
of the two parts of the field are in the same proportion as would be the
illuminations on a diffusing screen at C due to the two sources. The
advantage of the present arrangement, as compared for example with the
double-shadow method, lies in the immense saving of light. In the case
of water there is a great disproportion (of about 50 to 1) in the illuminations
as seen from F. In order to reduce the direct light to at least approximate
equality with the reflected, Talbot's device* of a revolving disk was employed.
This is shown in section at 7, and in plan at /'. The angular opening may be
chosen so as to allow for the loss in reflexion, and for the further disadvantage
under which the reflected light acts in respect of distance. The disk finally
employed was of zinc, stiffened with wood, and covered on both faces with
black velvet.
It was at first proposed to work as above described by eye estimations ;
but the necessity for a ready adjustment capable of introducing small relative
changes of brightness leads to further complications. Moreover, the large
disk which it is advisable to use for the sake of accurate measurement of the
angular opening, cannot well be rotated at the necessary speed of 20 or 25
revolutions per second. For this reason, and also for the sake of obtaining
a record capable of being examined at leisure, it was decided to work by
photography. This involves no change of principle. The photographic
plate H simply takes the place of the retina of the eye. But now the
* Phil. Mag. Vol. v. p. 327 (1834).
1892] AND MERCURY AT NEARLY PERPENDICULAR INCIDENCE. 7
integration of the effect over a somewhat prolonged exposure (of several
minutes) dispenses with the necessity for a rapid rotation of the Talbot disk,
and allows us to obtain at will a fine adjustment by screening one or the
other light from the plate for a measured interval of time. In practice the
direct light was thus partially cut off, a mechanically held screen being
advanced a little above the plane of the revolving disk. The reader will not
fail to observe that the incomplete coincidence of the times of exposure has
the disadvantage of rendering the calculation dependent upon the assumption
that the light is uniform over the duration of an experiment. Error that
might otherwise enter is, however, in great degree obviated by the precaution
of choosing the middle of the total period of exposure as the time for
screening.
The above is a sufficient explanation of the general scheme, but there
are many points of importance still to be described. With respect to the
source of light, it was at first supposed that even if the radiation upwards
and downwards could not be assumed to be equal, at any rate a reversal by
rotation of the lamp through 180° in the plane of the diagram would suffice
to eliminate error. On examination, however, it appeared that owing to
veins in the glass bulb the radiation in various directions was very irregular,
so much so that it was feared that mere reversal might prove an insufficient
precaution. The difficulty thus arising was met by covering the bulb, or at
least an equatorial belt of sufficient width, with thin tissue-paper, by which
anything like sudden variations of radiation with direction would be prevented,
and by causing the lamp to revolve slowly about its axis during the whole
time of exposure. The diameter of the bulb was about 1J inch, and the
illuminating-power rather less than that of one candle.
Another point of great importance is to secure that the light regularly
reflected from the upper surface of the liquid, which we wish to measure,
shall be free from admixture. It must be remembered that by far the greater
part of the light incident upon the liquid penetrates into the interior, and
must be annulled or at any rate diverted into a harmless direction. To this
end it is necessary that the liquid be free from turbidity and that proper
provision be made for the disposal of the light after its passage. It is not
sufficient merely to blacken the bottom of the dish in which the water is
contained. But the desired object is attained by the insertion into the water
of a piece of opaque glass, held at such a slight inclination to the horizon
that the light from the lamp regularly reflected at its upper surface is thrown
to one side. As additional precautions the disk and its mountings were
blackened, as were also the walls and ceiling of the room in which the
experiments were made.
The surface of water must be large enough to avoid curvature due to
capillarity. Shortly before an experiment it is cleansed with the aid of
8
ON THE INTENSITY OF LIGHT REFLECTED FROM WATER
[198
a hoop of thin sheet-brass about 2 inches wide. The hoop is deposited upon
the water so doubled up that it includes but an insensible area, and is then
opened out into a circle. In this way not only is the greasy film usually
present upon the surface greatly attenuated, but also dust is swept away.
The avoidance of dust, especially of a fibrous character, is important. Other-
wise the resulting deformation of the surface causes the field of the reflected
light to become patchy and irregular.
We come now to the silvered glass reflectors, which are assumed to reflect
the direct and reflected lights equally well. It seems safe to suppose that no
appreciable error can enter depending upon the slightly differing angles at
which the reflexion takes place in the two cases. But the mirrors are liable
to tarnish, and, indeed, in the earlier experiments soon showed signs of being
affected. The influence of this tarnish would be much greater in photographs
done upon ordinary plates, sensitive principally to blue light, than in the
estimation of the eye ; and it was thought desirable to eliminate once for all
any question of the effect of differential tarnishing by interchanging the
mirrors in the middle of each exposure. For this purpose a somewhat
elaborate mounting had to be contrived. It was executed by Mr Gordon
and answered its purpose extremely well.
The mirrors are carried by a brass tube B (Fig. 2), which revolves in an
Fig. 2.
external tube A A rigidly attached to the stand of the apparatus. A lateral
arm G, some inches in length, projects from B, and near its extremity bears
against one or other of two screw-stops D. The lower end of B carries
1892]
AND MERCURY AT NEARLY PERPENDICULAR INCIDENCE.
perpendicular to itself a brass plate EE (Fig. 3). The mirrors GG are of
plate-glass and are fixed by cement to two brass plates FF. The latter
plates are attached by friction only to EE, being on the one hand pushed
away by adjusting-screws HH, and on the other held up by four steel
springs /. The edges of the reflecting surfaces meet accurately in a line
passing through the axis of rotation, and the stops D are so adjusted that
the transition from the one bearing to the other corresponds to a rotation
through precisely 180°, so that on reversal the common edge of the reflectors
recovers its position. The two mirrors were originally silvered in one piece,
and the common edge corresponds to the division made by a diamond-cut at
the back. These arrangements were so successful that in spite of the reversal
between the two parts of the exposure the division-line appears sharp in the
photographs and exhibits no appearance of duplicity.
When not in use the reflecting-surfaces are protected by a sort of cap of
tin-plate, which fits loosely over them. The improvement thus obtained was
very remarkable, the mirrors not suffering so much in a month as they
formerly did in a day before the protection was provided.
The following are the measures of distances required for the calculation.
From the division-line C to the axis of rotation of the lamp A (Fig. 1),
AC =82-21 inches;
45=11-28, 50=93-15,
so that
45 + 50=104-43.
The factor expressing the ratio of the squares of the distances is thus
1-6137.
The angle of incidence is best obtained from a measurement of the
horizontal distance between G and A. This proved to be 11£ inches; so
that
Sin* = ^3 = 'n' and e = W'
This applies to all the experiments referred to in the present paper.
10 ON THE INTENSITY OF LIGHT REFLECTED FROM WATER [198
The estimation of the angular opening in the disk used for the water
experiments depended upon measurements of corresponding chord and
diameter. The chord, measured by means of the screw of a travelling-
microscope, was '7574 inch. The radius, expressed in terms of the same
unit, was found to be 7'79. Hence, if a be the angular opening,
•7574
or £or = 2° 47' =167'.
The ratio in which the direct light is reduced is thus
167 167
180 x 60 10800
= -01546.
It will now be necessary to give some details with respect to the actual
matches as determined photographically. At first the intention was to
employ ordinary plates (Ilford), which worked very satisfactorily. But when
the attempt was made to compare the result with theory, the comparison
was found to be embarrassed by uncertainty as to the effective wave-length
of the light in operation. Moreover, as these plates are scarcely sensitive
to yellow and green light, the effective wave-length is liable to considerable
variation with the current used to ignite the lamp. Photographs were in-
deed taken of the spectrum of the lamp as actually employed, but the
unsymmetrical character of the falling off at the two ends made it difficult
to fix upon the centre of activity. Recourse was then had to Edwards'
" isochromatic " plates. The spectrum of the lamp, as photographed upon
these plates after passing through a pale yellow glass, was very well defined,
lying with almost perfect symmetry between the sodium and the thallium
lines. It was, therefore, determined to use these plates and the same yellow
glass in the actual experiments, so that
X = | (5892 + 5349) = 5620
could be taken as the representative wave-length.
The only disadvantage arising from this change was in the necessary
prolongation of the exposure, which became somewhat tedious. Although
no dense image is required or indeed desirable, the exposure should be such
that the development does not need to be forced. Two photographs, with
different times of screening, were usually taken upon the same plate, the
object being to obtain a reversal of relative intensity, so that in one image
the semicircle representing the direct light should be more intense and in
the other image the semicircle representing the reflected light. The best
way of examining the pictures depended somewhat upon circumstances.
When the exposure and development had been suitable, the most effective
view for the detection of a feeble difference was obtained by placing the dry
picture, film downwards, upon a piece of opal glass. The light returned to
1892] AND MERCURY AT NEARLY PERPENDICULAR INCIDENCE. 11
the eye had then for the most part traversed the film twice, with the effect
of doubling any feeble difference which would occur on simple transmission.
Under favourable circumstances it was possible to detect a reversal between
the two images when the difference amounted to 3| per cent. A few such
experiments might therefore be expected to give the required result accurate
to less than one per cent.
With the Edwards' plates an exposure of 12 minutes was found to be
necessary. This was divided into two parts of 6 minutes each, with an
interval of one minute during which the mirrors were reversed. About the
middle of each period of 6 minutes the direct light was screened off for a time
which varied from picture to picture. For example, on June 6, the time of
screening for one picture was 71 seconds, and for the second picture 48 seconds.
This means that while in both pictures the exposure for the reflected light
was 12 minutes or 720 seconds, the exposures for the direct light were
respectively 720 - 2 x 71 = 578 seconds, and 720 - 2 x 48 = 624 seconds. The
water was distilled, and its temperature was 17°'7 C. The examination of
the finished pictures showed that the contrast was reversed, so that the
total exposure (to the direct light) required for a balance was intermediate
between 578 and 624, and, further, that the first mentioned was the nearer
to the mark.
The general conclusion derived from a large number of photographs was
that the balance corresponded to a total screening of 121 seconds, viz., to an
exposure of 720 - 121 = 599 seconds. This is for the direct light, the exposure
to the reflected light being always 720 seconds. The ratio of exposures
required for a balance is thus
599 : 720;
and this may be considered to correspond to a temperature of 18° C.
We can now calculate the observed reflexion for 6|° incidence, reckoned as
a fraction of the incident light. We have
599 167 /104-4ay
720' 10800 \82-2l) ~
The above relates to the impression upon Edwards' plates after the light
had been transmitted through a yellow glass. When Ilford plates were
substituted and the yellow glass omitted, the reflexion appeared decidedly
more powerful, and the ratio of exposures necessary for a balance was about
425 : 480, or 637 : 720. It appears, therefore, that the reflexion of the light
operative in this case is some 6 per cent, more than before, or about '0220 of
the incident light. As to a large increase of reflexion there was no doubt ;
but, owing perhaps to variations in the quality of the light, the agreement
between individual results was not so good as before.
12 ON THE INTENSITY OF LIGHT REFLECTED FROM WATER [198
It now remains to calculate the reflexion as given by Fresnel's formulae ;
and it appears from the discussion at the commencement of this paper that
we may ignore the small angle of incidence (6^°) and take the formula in the
simple form given by Young, viz. : —
As to the value of /A for water, Wiillner* gives
^ = 1-326067 - -000099 1 + '30531 \~2,
t denoting the temperature in Centigrade degrees. Applied to 18° and to
\ = 5620, this gives
p = 1-333951,
whence
The reflexion actually found is accordingly about 1|- per cent, greater than
that given by Fresnel's formulae.
In order to estimate the effect, according to the formula, of a change in
index, we may use
SR_ 4fy,
R ~"j#^I'
or, in the case of water,
8R /R = 5Sfj, nearly.
To cause a variation of 1^ per cent, in the reflexion, £//, would have to be
•003, and to cause 6 per cent. S/z would have to be "012. The latter exceeds
the variation of /j, in passing between the lines D and H.
The agreement with Fresnel's formula is thus pretty good, but the
question arises whether it ought not to be better. Apart from a priori
ideas as to the result to be expected, I should have estimated the errors
of experiment as not likely to exceed one-half per cent., and certainly no
straining of judgment in respect of the photometric pictures would bring
about agreement. On the other hand, it must be remembered that one per
cent, is not a large error in photometry, and that in the present case a
one per cent, error in the reflexion is but one in 5000 reckoned as a fraction
of the incident light. While, therefore, the disagreement may be real, it is
too small a foundation upon which to build with any confidence.
It only remains to record the results of some observations upon the
reflexion from mercury. In these experiments the revolving disk was dis-
pensed with, and the photographs were taken upon Edwards' plates through
yellow glass. The angle of incidence and all the other arrangements remained
as before. In order to obtain a balance it appeared that the direct light
* Fogg. Ann. Bd. cxxxiu.
1892] AND MERCURY AT NEARLY PERPENDICULAR INCIDENCE. 13
required to be screened for 64 seconds out of 120 seconds. The reflexion is
accordingly
56 /104-43y
The mercury was of good quality, and was filtered into a glass vessel just
before use. The level was adjusted to be the same as that adopted for the
observations upon water. A surface thus obtained would not be free from
a greasy layer, but it is not probable that this would sensibly influence the
reflexion.
APPENDIX.
The calculation of the reflexion depends upon the assumption that the
reflecting surface is plane ; and a very moderate concavity would suffice to
explain the small excess in the observed number for water over that calculated
from Fresnel's formula?. It is thus of importance to assure ourselves that
the concavity due to capillarity is really small enough to be neglected. For
this purpose an estimate founded upon the capillary surface applicable in two
dimensions will suffice.
If 6 be the inclination to the horizon at any point, x the horizontal and y
the vertical coordinate, the equations to the surface are : —
where
a*
At a great distance from the edge,
0 = 0, y = 0, x = oo .
At the vertical edge of a wetted vessel, 0 = ^tr.
The origin of x corresponds to
0 = TT, y=2a.
In the case of water T = 74, p = l, and g = 981 C.G.s. ; so that
a = '274 centim.
In the experiments upon reflexion the part of the surface in action was
about 11 centim. away from the boundary, so that x\ a = 40, and 6 is very
small.
14 INTENSITY OF LIGHT REFLECTED FROM WATER AND MERCURY. [198
For the curvature
or for our present purpose
To find 0 we have approximately,
cot £ 0 = e 3a, or 0 = 4e-38.
Accordingly
p '274 xe38'
This may be multiplied by 4 to represent the increase of effect in the actual
circumstances as compared with what is supposed in the two-dimensional
problem ; but it remains absolutely insensible in comparison with the other
curvatures involved.
199.
ON THE INTERFERENCE BANDS OF APPROXIMATELY HOMO-
GENEOUS LIGHT; IN A LETTER TO PROF. A. MICHELSON.
[Philosophical Magazine, xxxiv. pp. 407—411, 1892.]
WHEN we were discussing together the results of your interesting work
upon high interference, you asked my opinion upon one or two questions
connected therewith. I have delayed answering until I had the opportunity
of seeing your paper in print (Phil. Mag. Sept. 1892), but now I may as well
send you what I have to say.
First, as to the definiteness with which the character of the spectral line
(f>(x) can be deduced from the " visibility-curve." By Fourier's theorem,
i r°° f f+o° r+o0 . )
<£(#)= — I dulcosuxl cosuv<f>(v)dv + sm ux I sin uv<b(o)dv\ ;
TTJo ( J-oo J -oo J
or in your notation, if we identify u with 27T.D,
= - I du I G cos ux + S sin ux
Hence, if C and S are both given as functions of u, <f>(x) is absolutely, and
uniquely, determined. However, the visibility-curve by itself gives, not both
C and S, but only C* + S*; so that we must conclude that in general an
indefinite variety of structures is consistent with a visibility-curve given in
all its parts.
But if we may assume that the structure is symmetrical, S = 0 ; and <f> is
then determined by means of (7. And, since F2= (72/P2, the visibility-curve
determines C, or at least C2. In practice, considerations of continuity would
always fix the choice of the square root. Thus, in the case of a spectral band
of uniform brightness, where
we are to take
and not
16 ON THE INTERFERENCE BANDS [199
In order to determine both C and S, observations would have to be made
not only upon the visibility, but also upon the situation of the bands. You
remark that " it is theoretically possible by this means to determine, in case
of an unequal double, or a line unsymmetrically broadened, whether the
brighter side is towards the blue or the red end of the spectrum." But I
suppose that a complete determination of both C and S, though theoretically
possible, would be an extremely difficult task.
If the spectral line has a given total width, the " visibility " begins to fall
away from the maximum (unity) most rapidly when the brightness of the line
is all concentrated at the edges, so as to constitute a double line.
It is interesting to note that in several simple cases the bands seen with
ever increasing retardation represent the character of the luminous vibration
itself. In the case of a mathematical spectral line, the waves are regular to
infinity, and the bands are formed without limit and with maximum visibility
throughout. Again, in the case of a double line (with equal components) the
waves divide themselves into groups with intermediate evanescences, and
this is also the character of the interference bands. Thirdly, if the spectral
line be a band of uniform brightness, and if the waves at the origin be
supposed to be all in one phase, the actual compound vibration will be
accurately represented by the corresponding interference bands. But this
law is not general for the reason that in one case we have to deal with
amplitudes and in the other with intensities. The accuracy of correspondence
thus requires that the finite amplitudes involved be all of one magnitude. A
partial exception to this statement occurs in the case of a spectral line in
which the distribution of brightness is exponential.
Another question related to the effect of the gradual loss of energy, from
communication to the ether, upon the homogeneity of the light radiated from
freely vibrating molecules. In illustration of this we may consider the
analysis by Fourier's theorem of a vibration in which the amplitude follows
the exponential law, rising from zero to a maximum, and afterwards falling
again to zero. It is easily proved that
du COS UX (
in which the second member expresses an aggregate of trains of waves, each
individual train being absolutely homogeneous. If a be small in comparison
with r, as will happen when the amplitude on the left varies but slowly,
e-<M+r)'/4a« may ke neglected, and e~(u~r^^ is sensible only when u is very
nearly equal to r.
As an example in which the departure from regularity consists only in an
abrupt change of phase, let us suppose that
1892] OF APPROXIMATELY HOMOGENEOUS LIGHT. 17
the sign being reversed at every interval of ml, so that the positive sign
applies from 0 to ml, 2 ml to 3 ml, 4 ml to 5 ml, &c., and the negative sign
from ml to 2 ml, 3 ml to 4 ml, &c. As the analysis into simple waves we find
,
^'
the summation extending to odd values 1, 3, 5, ... of n. The fundamental
component cos(27nc/2raZ) and every odd harmonic occur, but not to the same
extent. When n is nearly equal to 2m, the terms rise to great relative
magnitude. The most important are thus
27raj/1 1 \ 27nc/n 2 \
cos— r (l + a— , cos -j- 1 1 -I- ^- , &c.;
I \ ~ 2m/ I \ ~ 2mJ
and it is especially to be remarked that what might at first sight be regarded
as the principal, if not the solitary, wave-length, viz. I, does not occur at all.
Besides communication of energy to the ether, and disturbance during
encounters with neighbours, the motion of the molecule itself has to be con-
sidered as hostile to homogeneity of radiation. The effect, according to
Doppler's principle, of motion in the line of sight was calculated by me on a
former occasion and is fully regarded in your paper. But there is another,
and perhaps more important, consequence of molecular motion, which does
not appear to have been remarked. Besides the motion of translation there
is the motion of rotation to be reckoned with. The effect of the latter will
depend upon the law of radiation in various directions from a stationary
molecule. As to this we do not know much, but enough to exclude the case
of radiation alike in all directions, as from an ideal source of sound. Such a
symmetry is indeed inconsistent with the law of transverse vibrations. The
simplest supposition is that the radiation is like that generated in an elastic
solid, at one point of which there acts a periodic force in a given direction.
In this case the amplitude in any direction varies as the sine of the angle
between the ray and the force, and the direction of (transverse) vibration lies
in the plane containing these two lines. A complete investigation of the
radiation from such molecules vibrating and rotating about all possible axes
would be rather complicated, but from one or two particular cases it is easy
to recognize the general character of the effect produced. Suppose, for
example, that the axis of rotation is perpendicular to the axis of vibration,
and consider the radiation in a direction perpendicular to the former axis.
If o> be the angular velocity, the amplitude varies as costot, and the vibration
may be represented by
2 cos cat . cos nt = cos (n + o>) t + cos (n — &>) t.
The spectrum would thus show a double line, whose components are separated
by a distance proportional to o>.
R. iv. 2
18 INTERFERENCE BANDS OF APPROXIMATELY HOMOGENEOUS LIGHT. [199
Again, if the ray be parallel to the axis of rotation, the amplitude is
indeed constant in magnitude, but its direction rotates. The plane-polarized
rays into which the vibration may be resolved are represented as before by
cos ait . cos nt. There is of course one case in which these complications fail to
occur, i.e. when the axis of rotation coincides with the axis of vibration ;
but with axes distributed at random we must expect vibrations (n ± <u) to be
almost as important as the vibration n. The law of distribution of brightness
in the spectral line would probably be exponential, as when the widening is
due to motion of molecules as wholes in the line of sight.
It will be of interest to compare the magnitudes of the two effects. If v
be the linear velocity of a molecule and V that of light, the comparison is
between a> and nv/ V, or between «o and v/\. If r be the radius of a molecule,
the circumferential velocity of rotation is o>r, and we may compare o>r with
vr/\. Now, according to Boltzmann's theorem, ra> would be of the same order
of magnitude as v, so that the importance of the rotatory and linear effects
would be somewhat as X : r. There is every reason to suppose that X is much
greater than r, and thus (if Boltzmann's relation held good) to expect that
the disturbance of homogeneity due to rotation would largely outweigh that
due to translation.
Your results seem already to interpose serious obstacles in the way of
accepting such a conclusion; and the fact that light may thus be thrown
upon a much controverted question in molecular physics is only another proof
of the importance of the research upon which you are engaged.
200.
ON THE INFLUENCE OF OBSTACLES ARRANGED IN
RECTANGULAR ORDER UPON THE PROPERTIES OF
A MEDIUM.
[Philosophical Magazine, xxxiv. pp. 431 — 502, 1892.]
THE remarkable formula, arrived at almost simultaneously by L. Lorenz*
and H. A. Lorentzf, and expressing the relation between refractive index
and density, is well known ; but the demonstrations are rather difficult to
follow, and the limits of application are far from obvious. Indeed, in some
discussions the necessity for any limitation at all is ignored. I have thought
that it might be worth while to consider the problem in the more definite
form which it assumes when the obstacles are supposed to be arranged in
rectangular or square order, and to show how the approximation may be
pursued when the dimensions of the obstacles are no longer very small in
comparison with the distances between them.
Taking, first, the case of two dimensions, let us investigate the con-
ductivity for heat, or electricity, of an otherwise uniform medium interrupted
by cylindrical obstacles which are arranged in rectangular order. The sides
of the rectangle will be denoted by a, /3, and the radius of the cylinders by a.
The simplest cases would be obtained by supposing the material composing
the cylinders to be either non-conducting or perfectly conducting; but it
will be sufficient to suppose that it has a definite conductivity different from
that of the remainder of the medium.
By the principle of superposition the conductivity of the interrupted
medium for a current in any direction can be deduced from its conductivities
in the three principal directions. Since conduction parallel to the axes of
the cylinders presents nothing special for our consideration, we may limit
* Wied. Ann. xi. p. 70 (1880).
t Wied. Ann. ix. p. 641 (1880).
2—2
20
ON THE INFLUENCE OF OBSTACLES
[200
our attention to conduction parallel to one of the sides (a) of the rectangular
structure. In this case lines parallel to a, symmetrically situated between
Fig- l.
o.
B
0
a
o
0
o
A
O
o
o
0
the cylinders, such as AD, BC, are lines of flow, and the perpendicular lines
AB, CD are equipotential.
If we take the centre of one of the cylinders P as origin of polar co-
ordinates, the potential external to the cylinder may be expanded in the
series
V= A0 + (^V + ^r-1) cos 6 + (A3r> + B3r~3) cos 30 + ... , (1)
and at points within the cylinder in the series
F ' = <70 + <7j r cos 0 + <73 r3 cos 3 0 + . . . , ( 2 )
0 being measured from the direction of a. The sines of 0 and its multiples
are excluded by the symmetry with respect to 0=0, and the cosines of the
even multiples by the symmetry with respect to 0 = |TT. At the bounding
surface, where r = a, we have the conditions
F=F', vdV'/dr = dV/dr,
v denoting the conductivity of the material composing the cylinders in terms
of that of the remainder reckoned as unity. The application of these con-
ditions to the term in cosn0 gives
7? — ^ 2w A (^\
In the case where the cylinders are perfectly conducting, v = x . If they
are non-conducting, v = 0.
The values of the coefficients .4 1} .B^ A3,B3... are necessarily the same
for all the cylinders, and each may be regarded as a similar multiple source
of potential. The first term A0, however, varies from cylinder to cylinder, as
we pass up or down the stream.
Let us now apply Green's theorem,
an
an
.(4)
1892] IN RECTANGULAR ORDER UPON A MEDIUM. 21
to the contour of the region between the rectangle A BCD and the cylinder P.
Within this region V satisfies Laplace's equation, as also will U, if we
assume
U = x = r cos 6 .................................. (5)
Over the sides BC, AD, dU/dn, dV/dn both vanish. On CD, $dV/dn.ds
represents the total current across the rectangle, which we may denote by C.
The value of this part of the integral over CD, AB is thus aC. The value
of the remainder of the integral over the same lines is — F,$, where V^
is the fall in potential corresponding to one rectangle, as between CD
and AB.
On the circular part of the contour,
and thus the only terms in (1) which will contribute to the result are those
in cos 0. Thus we may write
dV/dn = - (A1 -
so that this part of the integral is 2-rrB^ The final result from the application
of (4) is thus
«C-/9F1 + 2irB1 = 0 ............................ (6)
If #1 = 0, we fall back upon the uninterrupted medium of which the con-
ductivity is unity. For the case of the actual medium we require a further
relation between Bl and Vl.
The potential V at any point may be regarded as due to external sources
at infinity (by which the flow is caused) and to multiple sources situated
on the axes of the cylinders. The first part may be denoted by Hx. In
considering the second it will conduce to clearness if we imagine the (infinite)
region occupied by the cylinders to have a rectangular boundary parallel to
a and /3. Even then the manner in which the infinite system of sources
is to be taken into account will depend upon the shape of the rectangle.
The simplest case, which suffices for our purpose, is when we suppose the
rectangular boundary to be infinitely more extended parallel to a than parallel
to /3. It is then evident that the periodic difference Vl may be reckoned
as due entirely to Hx, and equated to Ha. For the difference due to the
sources upon the axes will be equivalent to the addition of one extra column
at + QO , and the removal of one at — oo , and in the case supposed such
a transference is immaterial*. Thus
V, = Ha .................................... (7)
simply, and it remains to connect H with BI.
* It would be otherwise if the infinite rectangle were supposed to be of another shape, e.g. to
be square.
22 j. ON THE INFLUENCE OF OBSTACLES [200
This we may do by equating two forms of the expression for the potential
at a point x, y near P. The part of the potential due to Hx and to the
multiple sources Q (P not included) is
or, if we subtract Hx, we may say that the potential at x, y due to the
multiple sources at Q is the real part of
A0 + (A1-H)(x + iy) + A3(x + iy)* + As(x + iy)* + .......... (8)
But if x', y' are the coordinates of the same point when referred to the centre
of one of the Q's, the same potential may be expressed by
2{B1(x' + iy')-> + B3(x' + iy')-*+...}, .................. (9)
the summation being extended over all the Q's. If £, 17 be the coordinates
of a Q referred to P,
x' = x-£, y' = y-r);
so that
Bn(x' + iy')~n = Bn(x + iy-£ -117)-*.
Since (8) is the expansion of (9) in rising powers of (x + iy), we obtain,
equating term to term,
-1.2.3^3=1.2.35^4 + 3.4.55326 + ... ...(10)
- 1 . 2 . 3 . 4 . 5 4 6 = 1 . 2 . 3 . 4 . 5 #! 26 + 3 . 4 . 5 . 6 . 7 55 28 + . . . J
and so on, where
2,n = 2(f + t17)-», .............................. (11)
the summation extending over all the Q's.
By (3) each B can be expressed in terms of the corresponding A. For
brevity, we will write
An = v'a-mBn, .... .......................... (12)
where
*/=(! + *)/(!-*) ............................. (13)
We are now prepared to find the approximate value of the conductivity.
From (6) the conductivity of the rectangle is
so that the specific conductivity of the actual medium for currents parallel
to a is
and the ratio of H to Bl is given approximately by (10) and (12).
In the first approximation we neglect 24, 26 ..., so that AS,AS... B3, Bs ...
vanish. In this case
(15)
1892] IN RECTANGULAR ORDER UPON A MEDIUM. 23
and the conductivity is
2-Tra2
"££(,/ + a%) ............................ (16)
The second approximation gives
^W + a^ — a'S,', ........................ (17)
and the series may be continued as far as desired.
The problem is thus reduced to the evaluation of the quantities 2a, 24,....
We will consider first the important particular case which arises when the
cylinders are in square order, that is when f3 = a. £ and 77 in (11) are then
both multiples of a, and we may write
2n=a-nSn, ................................. (18)
where
Sn=2(m' + im)-n; ........................... (19)
the summation being extended to all integral values of m, m', positive or
negative, except the pair m = 0, m' = 0. The quantities S are thus purely
numerical, and real.
The next thing to be remarked is that, since m, m' are as much positive
as negative, Sn vanishes for every odd value of n. This holds even when
a and ft are unequal.
Again,
Sm = 2 (TO' + tw)-*1 = i-™ 2 (- im'
- m
Whenever n is odd, S^ = — S^, or $m vanishes. Thus for square order,
S. = 8U = -SU= ...... = 0 ......................... (20)
This argument does not, without reservation, apply to 8,. In that case
the sum is not convergent ; and the symmetry between m and m', essential
to the proof of evanescence, only holds under the restriction that the infinite
region over which the summation takes place is symmetrical with respect
to the two directions a and ft — is, for example, square or circular. On the
contrary, we have supposed, and must of course continue to suppose, that the
region in question is infinitely elongated in the direction of a.
The question of convergency may be tested by replacing the parts of
the sum relating to a great distance by the corresponding integral. This is
[f dxdy _ [[cos2n0rdrd0 .
])(x + iyr~M »*»
and herein
fr-^+1dr = r-^+2l(- 2w + 2) ;
so that if 7i > 1 there is convergency, but if n — 1 the integral contains an
infinite logarithm.
24 ON THE INFLUENCE OF OBSTACLES [200
We have now to investigate the value of S2 appropriate to our purpose ;
that is, when the summation extends over the region bounded by x — ± u,
y = ±v, where u and v are both infinite, but so that v/u = Q. If we suppose
that the region of summation is that bounded by &• = + v, y—±v, the sum
vanishes by symmetry. We may therefore regard the summation as ex-
tending over the region bounded externally by x = + 00 , y = ±v, and internally
Fig. 2.
by ac = ± v (Fig. 2). When v is very great, the sum may be replaced by the
corresponding integral. Hence
the limits for y being ± v, and those for x being v and oo . Ultimately v is to
be made infinite.
We have
dy j i 2V
/*+*
J _„ (a;
= =
+ iyf x + iv x — iv x* + vz '
and
Accordingly
S.-T (22)
In the case of square order, equations (10), (12) give
Ha? 3 _ 7 .
= ,,' + ^._±!l^4»_l/ _£82- ...; (23)
and by (14)
Conductivity =1--^ . ^ (24)
If p denote the proportional space occupied by the cylinders,
P — TTO^IO?; (25)
and
Conductivity = 1 ^-^ - (26)
1892] IN RECTANGULAR ORDER UPON A MEDIUM. 25
Of the double summation indicated in (19) one part can be effected
without difficulty. Consider the roots of
sin (f — irmr) = 0.
They are all included in the form
' innr,
where m is any integer, positive, negative, or zero. Hence we see that
sin (f — irmr) may be written in the form
-^ . ...
irmr/ \ irmr + TrJ\ irmr — irj \ irmr + 2-rrJ
in which
A = — sin irmr.
Thus
log fcos £ — cot ivmr sin £] = log ( 1 — ^— ] + log ( 1 — -: — - — ) + . .
s\ irmrj ^ \ irmr + Tr/
If we change the sign of m, and add the two equations, we get
whence, on expansion of the logarithms,
sm2 sin4£
I __ 2 __ I __ »
'
_ __ __ __ __
sin2im7r 2snrH'w7r 3 sin6 i rmr
' * """ '
(irmr)2 (irmr 4- 7r)2 (irmr — 7r)2
+ W \ ,_.__. x, + (t-m7r + ^4 + ^-m ^ _ ^4 + •••
+ ^a{J_ 1 1 ] ,
^? ((irmr)6 (irmr + TT)" (imw - vr)6 j
By expanding the sines on the left and equating the corresponding powers
of |, we find
1 1 1 1 7T2
(tm)2 (im + I)2 (im-1)2 (im + 2)2 ...... ~ son" tin w
l I ^^ *
(im)4 (im + 1)4 3sin2im7T si"'-'
__ __ __
8 ' 6 15sin2tm7r s4 8'
26
ON THE INFLUENCE OF OBSTACLES
[200
In the summation with respect to m, required in (19), we are to take
all positive and negative integral values. But in the case of m = 0 we are
to leave out the first term, corresponding to m' = 0. When m = 0,
sm2ira7r (im)2 3 '
which, as is well known, is the value of
1111
Hence
and in like manner
»l = oo
2 2 SUrtWTT + J7T2;
.(30)
?4 = ^ + 27r4 2 {- 1 snrtW -f sin-Hm-Tr}, (31)
27r8 4.9 e"!00
7i 7^~~ ~T ATT £*
(32)
We have seen already that 86 = 0, and that S2 = TT. The comparison of the
latter with (30) gives
"-" ' -"' 1-J. ...(33)
We will now apply (31) to the numerical calculation of S4. We
find:
m
— sin~2 irmr
sin"4 imx
I
•00749767
•0000562150
2
v 1395
2
3
3
Sum
•00751165
•0000562152
so that
S4=7r4x. 03235020 (34)
In the same way we may verify (33), and that (32) = 0.
If we introduce this value into (26), taking for example the case where
the cylinders are non-conductive (y'= 1), we get
1-
.(35)
From the above example it appears that in the summation with respect
to m there is a high degree of convergency. The reason for this will appear
more clearly if we consider the nature of the first summation (with respect
1892] IN RECTANGULAR ORDER UPON A MEDIUM. 27
to m). In (19) we have to deal with the sum of (x + iy)~n, where y is for
the moment regarded as constant, while x receives the values x = m. If
instead of being concentrated at equidistant points, the values of x were
uniformly distributed, the sum would become
dx
Now, n being greater than 1, the value of this integral is zero. We see,
then, that the finite value of the sum depends entirely upon the discontinuity
of its formation, and thus a high degree of convergency when y increases may
be expected.
The same mode of calculation may be applied without difficulty to any
particular case of a rectangular arrangement. For example, in (11)
22 = 2 (ma. + tra/3)-2 = cr22(w' +im/3/a)-'.
If m be given, the summation with respect to mf leads, as before, to
l ,
a.)
and thus
(36)
The numerical calculation would now proceed as before, and the final
approximate result for the conductivity is given by (16). Since (36) is not
symmetrical with respect to a and ft, the conductivity of the medium is
different in the two principal directions.
When /3 = a, we know that a~222 = TT. And since this does not differ
much from |7r2, it follows that the series on the right of (36) contributes
but little to the total. The same will be true, even though ft be not equal
to a, provided the ratio of the two quantities be moderate. We may then
identify a~2S2 with TT, or with ^7r2, if we are content with a very rough
approximation.
The question of the values of the sums denoted by JLm is intimately
connected with the theory of the 0- functions *, inasmuch as the roots of H(u),
or O^TTu/ZK), are of the form
2m K + 2m'iK'.
The analytical question is accordingly that of the expansion of log #j(#)
in ascending powers of x. Now, Jacobif has himself investigated the ex-
pansion in powers of x of
#!(#) = 2 fa174 sm#-g9/4 sin Stf + g26''4 sin 5#- ...}, ............ (37)
* Cayley's Elliptic Functions, p. 300. The notation is that of Jacobi.
t Crelle, Bd. LIV. p. 82.
28 ON THE INFLUENCE OF OBSTACLES [200
where q = e-«K'tK. ................................. (38)
So far as the cube of x the result is
D being a constant which it is not necessary further to specify. K and E
are the elliptic functions of k usually so denoted. By what has been stated
above the roots of Ox are of the form
(40)
so that
Z-k*)K'}, ............ (41)
the summation on the left being extended to all integral values of m and m,
except m = 0, m' = 0.
This is the general solution for 22. If K' = K, k? = %, and
2 {m + im'}-* = 2 {2KE - K*} = TT,
since in general*,
EK' + E'K-KK' = £TT.
In proceeding further it is convenient to use the form in which an
exponential factor is removed from the series. This is
.. .,... (42)
-3
5! 7!
in which
(43)
7T 7T 7T
the law of formation of s being
sm+l = 2m (2m + 1) ps^ + aj3dsm/d/3 - 8^dsm/da, ...... (44)
while
a = #»-*», f3 = ^(kk') ............................ (45)
I have thought it worth while to quote these expressions, as they do
not seem to be easily accessible ; but I propose to apply them only to the
case of square order, K' = K, k'2 = k?=%. Thus
(46)
and
* Cayley's Elliptic Functions, p. 49.
1892] IN RECTANGULAR ORDER UPON A MEDIUM.
Hence
, e^x) a? AW Aao?
l0^ 5^ =-2^-275-!- i^35X!-
If +XD ±X.j, ... are the roots of 0l(as)/x = Q, we have
-
*•* /v — f\ > •*•' 'v — K i ? •** /v — v, — HT (r i K i •
2-7T 5! 7.5.4.5!
Now by (40) the roots in question are TT (m, + ira'), and thus
™ 7T4 . rt 7TSA
in which
Leaving the two-dimensional problem, I will now pass on to the case
of a medium interrupted by spherical obstacles arranged in rectangular order.
As before, we may suppose that the side of the rectangle in the direction
of flow is a, the two others being /3 and 7. The radius of the sphere is a.
The course of the investigation runs so nearly parallel to that already
given, that it will suffice to indicate some of the steps with brevity. In place
of (1) and (2) we have the expansions
*)Yn+..., ............ (50)
V'=C0+C1Y1r+...+CnYnrn+..., ............ (51)
Yn denoting the spherical surface harmonic of order n. And from the surface
conditions
V=V,
we find
We must now consider the limitations to be imposed upon Yn. In
general,
Yn = ^ ®n <*> (Hs cos s<f> + Ks sin s<f>), .................. (53)
* = 0
where
©„<"> = sin*0 (cosw-*0 - ^-sKrc-*-1) cos«-«-20 + ...... \ . . .(54)
0 being supposed to be measured from the axis of a; (parallel to a), and <f>
from the plane of xz. In the present application symmetry requires that
s should be even, and that Yn (except when n = 0) should be reversed when
30 ON THE INFLUENCE OF OBSTACLES [200
(tr—d) is written for 0. Hence even values of n are to be excluded altogether.
Further, no sines of s<f> are admissible. Thus we may take
(55)
F3 = cos30-f cos 0+H2sm*0 cos 0 cos 2<f>, ..................... (56)
F5 = cos8 6 - J£ cos30 + ^ cos (9
sin20 (cos30 - £ cos 0) cos 20
40cos0cos4<£ ............................... (57)
In the case where ft = y symmetry further requires that
#2 = 0, Z2=0 ............................... (58)
In applying Green's theorem (4) the only difference is that we must now
understand by s the area of the surface bounding the region of integration.
If C denote the total current flowing across the faces /3<y, V1 the periodic
difference of potential, the analogue of (6) is
aC-ftjV, + 4-^ = 0 ............................ (59)
We suppose, as before, that the system of obstacles, extended without
limit in every direction, is yet infinitely more extended in the direction
of a than in the directions of ft and 7. Then, if Hx be the potential due to
the sources at infinity other than the spheres, Vl = Ha, and
so that the specific conductivity of the compound medium parallel to a is
We will now show how the ratio B1/H is to be calculated approximately,
limiting ourselves, however, for the sake of simplicity to the case of cubic
order, where a = j3 = y. The potential round P, viz.
may be regarded as due to Hx and to the other spheres Q acting as sources
of potential. Thus, if we revert to rectangular coordinates and denote the
coordinates of a point relatively to P by a, y, z, and relatively to one of the
Q's by a/, y", z', we have
in which
a' = x - £, y' = y-i1, z' = z-%,
^ %y *?> £ be the coordinates of Q referred to P. The left side of (61) is thus
the expansion of the right in ascending powers of x, y, z. Accordingly,
1892] IN RECTANGULAR ORDER UPON A MEDIUM. 31
Al—H is found by taking dfdx of the right-hand member and then making
x, y, z vanish. In like manner 6 A3 will be found from the third differential
coefficient. Now, at the origin,
_ = ___= =
dx r'3 ~ d% r'3 d% p* p*
in which
p* = ? + i)* + ?.
It will be observed that we start with a harmonic of order 1 and that
the differentiation raises the order to 2. The law that each differentiation
raises the order by unity is general ; and, so far as we shall proceed, the
harmonics are all zonal, and may be expressed in the usual way as functions
Pn(fji) of p where /* = £/?. Thus
In like manner,
and
The comparison of terms in (61) thus gives
^3= -45J So-5 P\ + 5 I <62)
... = j
In each of the quantities, such as 2p~3P2> the summation is to be ex-
tended to all the points whose coordinates are of the form la, mx, not, where
I, m, n are any set of integers, positive or negative, except 0, 0, 0. If we
take a = 1, and denote the corresponding sums by $2, S4, ... , these quantities
will be purely numerical, and
V"-1^ = «-*-'£„ (63)
From (52), (62) we now obtain
which with (60) gives the desired result for the conductivity of the medium.
We now proceed to the calculation of $2. We have
By the symmetry of a cubical arrangement, it follows that
32 ON THE INFLUENCE OF OBSTACLES [200
so that if $ were calculated for a space bounded by a cube, it would
necessarily vanish. But for our purpose $2 is to be calculated over the space
bounded by f = ± oc , i) = ±v, £=±v, where v is finally to be made infinite ;
and, as we have just seen, we may exclude the space bounded by
so that £$2 will be obtained from the space bounded by
^ — v, f=oo, ?)= + v, £= ±v.
Now when p is sufficiently great, the summation may be replaced by an
integration; thus
In this,
("
J,
and finally
1 -v (*>2 + r9(20* + f1)* = Jo V(2 + tan20) = Jo >/(2 - 52) = 3 '
Thus
c ^7r /«K\
^2 = -g- V65)
If we neglect a10/a10, and write p for the ratio of volumes, viz.
we have by (60) for the conductivity
--
'
or in the particular case of non-conducting obstacles (v = 0)
In order to carry on the approximation we must calculate S4 &c. Not
seeing any general analytical method, such as was available in the former
problem, I have calculated an approximate value of $4 by direct summation
from the formula
Compare Maxwell's Electricity, § 314.
1892]
IN RECTANGULAR ORDER UPON A MEDIUM.
33
We may limit ourselves to the consideration of positive and zero values of
f , i], £ Every term for which (-, 17, £ are finite is repeated in each octant,
that is 8 times. If one of the three coordinates vanish, the repetition is
fourfold, and if two vanish, twofold.
The following table contains the result for all points which lie within
p* = 18. This repetition in the case, for example, of p2 = 9 represents two
kinds of composition. In the first
and in the second
= 32 + O2 + O2 = 9.
The success of the approximation is favoured by the fact that P vanishes
when integrated over the complete sphere, so that the sum required is only
a kind of residue depending upon the discontinuity of the summation.
The result is
.(69)
P2
P2
0, 0, 1
1
+ 3-5000
0, 0, 3
9
+ -0144
0, 1, 1
2
- -3094
0, 1, 3
10
+ -0243
1, 1, 1
3
- -1996
1, 1, 3
11
+ -0075
0, 0, 2
4
+ -1094
2, 2, 2
12
- -0062
0, 1, 2
5
+ -0501
0, 2, 3
13
- -0015
1, 1, 2
6
- -0397
1, 2, 3
14
- -0095
0, 2, 2
8
- -0097
0, 0, 4
16
+ -0034
1, 2, 2
9
- -0277
2, 2, 3
17
- -0061
0, 1, 4
17
+ '0085
1
The results of our investigation have been expressed for the sake of
simplicity in electrical language as the conductivity of a compound medium,
but they may now be applied to certain problems of vibration. The simplest
of these is the problem of wave-motion in a gaseous medium obstructed by
rigid and fixed cylinders or spheres. It is assumed that the wave-length
is very great in comparison with the period (a, ft, 7) of the structure. Under
these circumstances the flow of gas round the obstacles follows the same
law as that of electricity, and the kinetic energy of the motion is at once
given by the expressions already obtained. In fact the kinetic energy
corresponding to a given total flow is increased by the obstacles in the same
proportion as the electrical resistances of the original problem, so that the
influence of the obstacles is taken into account if we suppose that the
B. IV. 3
34 ON THE INFLUENCE OF OBSTACLES [200
density of the gas is increased in the above ratio of resistances. In the
case of cylinders in square order (35), the ratio is approximately
and in the case of spheres in cubic order by (68) it is approximately
(1
But this is not the only effect of the obstacles which we must take
into account in considering the velocity of propagation. The potential
energy also undergoes a change. The space available for compression
or rarefaction is now (1 - p) only instead of 1; and in this proportion
is increased the potential energy corresponding to a given accumulation of
gas*. For cylindrical obstruction the square of the velocity of propagation is
thus altered in the ratio
so that if fj, denote the refractive index, referred to that of the unobstructed
medium as unity, we find
(p?-l)jp = constant, ........................ (70)
which shows that a medium thus constituted would follow Newton's law
as to the relation between refraction and density of obstructing matter. The
same law (70) obtains also in the case of spherical obstacles ; but reckoned
absolutely the effect of spheres is only that of cylinders of halved density.
It must be remembered, however, that while the velocity in the last case
is the same in all directions, in the case of cylinders it is otherwise. For
waves propagated parallel to the cylinders the velocity is uninfluenced by
their presence. The medium containing the cylinders has therefore some
of the properties which we are accustomed to associate with double refraction,
although here the refraction is necessarily single. To this point we shall
presently return, but in the meantime it may be well to apply the formulae
to the more general case where the obstacles have the properties of fluid,
with finite density and compressibility.
To deduce the formula for the kinetic energy we have only to bear in
mind that density corresponds to electrical resistance. Hence, by (26), if
a denote the density of the cylindrical obstacle, that of the remainder of
the medium being unity, the kinetic energy is altered by the obstacles in the
approximate ratio
(<r + l)/(<r-l)+ff
(<r + l)/(cr-l)-p-
* Theory of Sound, § 303.
1892] IN RECTANGULAR ORDER UPON A MEDIUM. 35
The effect of this is the same as if the density of the whole medium were
increased in the like ratio.
The change in the potential energy depends upon the " compressibility "
of the obstacles. If the material composing them resists compression m times
as much as the remainder of the medium, the volume^? counts only as p/m,
and the whole space available may be reckoned asl— p +p/m instead of 1.
In this proportion is the potential energy of a given accumulation reduced.
Accordingly, if p be the refractive index as altered by the obstacles,
^ = (n)x(l-p+plm) (72)
The compressibilities of all actual gases are nearly the same, so that if we
suppose ourselves to be thus limited, we may set m = l, and
( }
or, as it may also be written,
At2- 1 1
^-y - = constant ............................ (74)
In the case of spherical obstacles of density a- we obtain in like manner
--
or
^Ji = constant ............................ (76)
In the general case, where m is arbitrary, the equation expressing p in
terms of p? is a quadratic, and there are no simple formulae analogous to
(74) and (76).
It must not be forgotten that the application of these formulae is limited
to moderately small values of p. If it be desired to push the application
as far as possible, we must employ closer approximations to (26), &c. It
may be remarked that however far we may go in this direction, the final
formula will always give p? explicitly as a function of p. For example, in the
case of rigid cylindrical obstacles, we have from (35)
It will be evident that results such as these afford no foundation for
a theory by which the refractive properties of a mixture are to be deduced
by addition from the corresponding properties of the components. Such
theories require formulae in which p occurs in the first power only, as
in (76).
3—2
36 ON THE INFLUENCE OF OBSTACLES [200
If the obstacles are themselves elongated, or even, though their form
be spherical, if they are disposed in a rectangular order which is not cubic,
the velocity of wave-propagation becomes a function of the direction of the
wave-normal. As in Optics, we may regard the character of the refraction as
determined by the form of the wave-surface.
The seolotropy of the structure will not introduce any corresponding
property into the potential energy, which depends only upon the volumes
and compressibilities concerned. The present question, therefore, reduces
itself to the consideration of the kinetic energy as influenced by the direction
of wave-propagation. And this, as we have seen, is a matter of the electrical
resistance of certain compound conductors, on the supposition, which we
continue to make, that the wave-length is very large in comparison with the
periods of the structure. The theory of electrical conduction in general
has been treated by Maxwell (Electricity, § 297). A parallel treatment of
the present question shows that in all cases it is possible to assign a system
of principal axes, having the property that if the wave-normal coincide with
any one of them the direction of flow will also lie in the same direction,
whereas in general there would be a divergence. To each principal axis
corresponds an efficient " density," and the equations of motion, applicable to
the medium in the gross, take the form
d°-% dB d*-n dS d*£ dS
<rx -r- = ml j- , a-v jTr- = Wa -3- , <rz -3-2. = m
dtz dx y dt* dy dt2 dz
where £, 17, £ are the displacements parallel to the axes, ml is the compressi-
bility, and
d£ dq d£
o — -j — (- -j— -|- -j— .
dx dy dz
If X, fi, v are the direction-cosines of the displacement, I, m, n of the
wave-normal, we may take
£=X0, 7? = /i0, £=v0,
where
0 _ gi(lx+my+nz - Vt)
Thus
d8/da; = -W(l\ + mfji + nv), &c.
and the equations become
<rx\Vz = mj(l\ + mfi + nv),
a-yfiV2 = TT^m^X + w/i + nv),
crzvV* = mjn (l\ + m/j. + nv),
from which, on elimination of X : /* : v, we get
V* =mi(l- + m\ ^] = an'~+ Km? + c%2, . . .(78)
-
if a, b, c denote the velocities in the principal directions x, y, z.
1892] IN RECTANGULAR ORDER UPON A MEDIUM 37
The wave-surface after unit time is accordingly the ellipsoid whose axes
are a, b, c.
As an example, if the medium, otherwise uniform, be obstructed by rigid
cylinders occupying a moderate fraction (p) of the whole space, the velocity
in the direction z, parallel to the cylinders, is unaltered ; so that
In the application of our results to the electric theory of light we con-
template a medium interrupted by spherical, or cylindrical, obstacles, whose
inductive capacity is different from that of the undisturbed medium. On
the other hand, the magnetic constant is supposed to retain its value un-
broken. This being so, the kinetic energy of the electric currents for the
same total flux is the same as if there were no obstacles, at least if we regard
the wave-length as infinitely great*. And the potential energy of electric
displacement is subject to the same mathematical laws as the resistance of
our compound electrical conductor, specific inductive capacity in the one
question corresponding to electrical conductivity in the other.
Accordingly, if v denote the inductive capacity of the material composing
the spherical obstacles, that of the undisturbed medium being unity, then
the approximate value of p? is given at once by (67). The equation may
also be written in the form given by Lorentz,
(79)
and, indeed, it appears to have been by the above argument that (79) was
originally discovered.
The above formula applies in strictness only when the spheres are
arranged in cubic order f, and, further, when p is moderate. The next
approximation is
(80)
If the obstacles be cylindrical, and arranged in square order, the compound
medium is doubly refracting, as in the usual electric theory of light, in which
the medium is supposed to have an inductive capacity variable with the
direction of displacement, independently of any discontinuity in its structure.
The double refraction is of course of the uniaxal kind, and the wave-surface is
the sphere and ellipsoid of Huygens.
* See Prof. Willard Gibbs's " Comparison of the Elastic and Electric Theories of Light,"
Am. Journ. Sci. xxxv. (1888).
t An irregular isotropic arrangement would, doubtless, give the same result.
38 INFLUENCE OF OBSTACLES IN RECTANGULAR ORDER UPON A MEDIUM. [200
For displacements parallel to the cylinders the resultant inductive capacity
(analogous to conductivity in the conduction problem) is clearly 1 —p + vp:
so that the value of p? for the principal extraordinary index is
l*=I + (»-l)p, - (81)
giving Newton's law for the relation between index and density.
For the ordinary index we have
^=(26),
in which v = (1 + z>)/(l — v), while $4, S8 ... have the values given by (49).
If we omit p*, &c. we get
(82)
v+p'
-11 1 v-l
/A2 + 1 P V V+l'
.(83)
The general conclusion as regards the optical application is that, even
if we may neglect dispersion, we must not expect such formulae as (79)
to be more than approximately correct in the case of dense fluid and solid
bodies.
201.
ON THE DENSITIES OF THE PRINCIPAL GASES.
[Proceedings of the Royal Society, LIII. pp. 134 — 149, 1893.]
IN former communications * I have described the arrangements by which
I determined the ratio of densities of oxygen and hydrogen (1 5*882). For
the purpose of that work it was not necessary to know with precision the
actual volume of gas weighed, nor even the pressure at which the containing
vessel was filled. But I was desirous, before leaving the subject, of ascertain-
ing not merely the relative, but also the absolute, densities of the more
important gases, that is, of comparing their weights with that of an equal
volume of water. To effect this it was necessary to weigh the globe, used to
contain the gases, when charged with water, an operation not quite so simple
as at first sight it appears. And, further, in the corresponding work upon the
gases, a precise absolute specification is required of the temperature and
pressure at which a filling takes place. To render the former weighings
available for this purpose, it would be necessary to determine the errors of
the barometers then employed. There would, perhaps, be no great difficulty
in doing this ; but I was of opinion that it would be an improvement to use a
manometer in direct connexion with the globe, without the intervention of
the atmosphere. In the latter manner of working, there is a doubt as to
the time required for full establishment of equilibrium of pressure, especially
when the passages through the taps are partially obstructed by grease.
When the directly connected manometer is employed, there is no temptation
to hurry from fear of the entrance of air by diffusion, and, moreover (Note A),
the time actually required for the establishment of equilibrium is greatly
diminished. With respect to temperature, also, it was thought better to
avoid all further questions by surrounding the globe with ice, as in Regnault's
original determinations. It is true that this procedure involves a subsequent
cleaning and wiping of the globe, by which the errors of weighing are con-
siderably augmented ; but, as it was not proposed to experiment further with
hydrogen, the objection was of less force. In the^case of the heavier gases,
unsystematic errors of weighing are less to be feared than doubts as to the
actual temperature.
* Roy. Soc. Proc. February, 1888 [Vol. in. p. 37] ; February, 1892 [Vol. in. p. 524].
40 ON THE DENSITIES OF THE PRINCIPAL GASES. [201
In order to secure the unsystematic character of these errors, it is
necessary to wash and wipe the working globe after an exhaustion in the
same manner as after a filling. The dummy globe (of equal external volume,
as required in Regnault's method of weighing gases) need not be wiped
merely to secure symmetry, but it was thought desirable to do so before each
weighing. In this way there would be no tendency to a progressive change.
In wiping the globes the utmost care is required to avoid removing any
loosely attached grease in the neighbourhood of the tap. The results to be
given later will show that, whether the working globe be full or empty, the
relative weights of the two globes can usually be recovered to an accuracy of
about 0'3 milligramme. As in the former papers, the results were usually
calculated by comparison of each " full " weight with the mean of the
immediately preceding and following empty weights. The balance and the
arrangements for weighing remained as already described.
The Manometer.
The arrangements adopted for the measurement of pressure must be
described in some detail, as they offer several points of novelty. The apparatus
actually used would, indeed, be more accurately spoken of as a manometric
gauge, but it would be easy so to modify it as to fit it for measurements
extending over a small range.
The object in view was to avoid certain defects to which ordinary
barometers are liable, when applied to absolute measurements. Of these
three especially may be formulated : —
a. It is difficult to be sure that the vacuum at the top of the mercury is
suitable for the purpose.
6. No measurements of a length can be regarded as satisfactory in which
different methods of reading are used for the two extremities.
c. There is necessarily some uncertainty due to irregular refraction by
the walls of the tube. The apparent level of the mercury may
deviate from the real position.
d. To the above may be added that the accurate observation of the
barometer, as used by Regnault and most of his successors, requires
the use of a cathetometer, an expensive and not always satisfactory
instrument.
The guiding idea of the present apparatus is the actual application of a
measuring rod to the upper and lower mercury surfaces, arranged so as to be
vertically superposed. The rod A A, Fig. 1, is of iron (7 mm. in diameter),
pointed below, B. At the upper end, C, it divides at the level of the mercury
into a sort of fork, and terminates in a point similar to that at B, and, like it,
directed downwards. The coincidence of these points with their images
1893] ON THE DENSITIES OF THE PRINCIPAL GASES. 41
reflected in the mercury surfaces, is observed with the aid of lenses of about
30 mm. focus, held in position upon the wooden framework of the apparatus.
It is, of course, independent of any irregular refraction which the tube may
exercise. The vertically of the line joining the points is tested without
difficulty by a plumb-line.
Fig. l.
The upper and lower chambers C, B are formed from tubing of the same
diameter (about 21 mm. internal). The upper communicates through a tap,
D, with the Toppler, by means of which a suitable vacuum can at any time
be established and tested. In ordinary use, D stands permanently open, but
its introduction was found useful in the preliminary arrangements and in
testing for leaks. The connexion between the lower chamber B and the
vessel in which the pressure is to be verified takes place through a side
tube, E.
The greater part of the column of mercury to which the pressure is due is
contained in the connecting tube FF, of about 3 mm. internal diameter. The
temperature is taken by a thermometer whose bulb is situated near the
42 ON THE DENSITIES OF THE PRINCIPAL GASES. [201
middle of FF. Towards the close of operations the more sensitive parts are
protected by a packing of tow or cotton-wool, held in position between two
wooden boards. The anterior board is provided with a suitable glass window,
through which the thermometer may be read.
It is an essential requirement of a manometer on the present plan that
the measuring rod pass air-tight from the upper and lower chambers into the
atmosphere. To effect this the glass tubing is drawn out until its internal
diameter is not much greater than that of the rod. The joints are then made
by short lengths of thick-walled india-rubber H, G, wired on and drowned
externally in mercury. The vessels for holding the mercury are shown at /,
K. There is usually no difficulty at all in making perfectly tight joints
between glass tubes in this manner ; but in the present case some trouble
was experienced in consequence apparently of imperfect approximation be-
tween the iron and the mercury. At one time it was found necessary to
supplement the mercury with vaseline. When tightness is once obtained,
there seems to be no tendency to deterioration, and the condition of things is
under constant observation by means of the Toppler.
The distance between the points of the rod is determined under
microscopes by comparison with a standard scale, before the apparatus is put
together. As the rod is held only by the rubber connexions, there is no fear
of its length being altered by stress.
The adjustment of the mercury (distilled in a vacuum) to the right level
is effected by means of the tube of black rubber LM, terminating in the
reservoir N. When the supply of mercury to the manometer is a little short
of what is needed, the connexion with the reservoir is cut off by a pinch -cock
at 0, and the fine adjustment is continued by squeezing the tube at P
between a pair of hinged boards, gradually approximated by a screw. This
plan, though apparently rough, worked perfectly, leaving nothing to be
desired.
It remains to explain the object of the vessel shown at Q. In the early
trials, when the rubber tube was connected directly to R, the gradual fouling
of the mercury surface, which it seems impossible to avoid, threatened to
interfere with the setting at B. By means of Q, the mercury can be discharged
from the measuring chambers, and a fresh surface constituted at B as well
as at (7.
The manometer above described was constructed by my assistant,
Mr Gordon, at a nominal cost for materials ; and it is thought that the same
principle may be applied with advantage in other investigations. In cases
where a certain latitude in respect of pressure is necessary, the measuring rod
might be constructed in two portions, sliding upon one another. Probably a
range of a few millimetres could be obtained without interfering with the
india-rubber connexions.
1893]
ON THE DENSITIES OF THE PRINCIPAL GASES.
43
The length of the iron rod was obtained by comparison under microscopes
with a standard bar R divided into millimetres. In terms of R the length
at 15° C. is 762*248 mm. It remains to reduce to standard millimetres.
Mr Chaney has been good enough to make a comparison between R and the
iridio-platmum standard metre, 1890, of the Board of Trade. From this it
appears that the metre bar R is at 15° C. 0'3454 mm. too long; so that the
true distance between the measuring points of the iron rod is at 15° C.
762-248 x 1-0003454 = 762-511 mm.
Connexions with Pump and Manometer.
Some of the details of the process of filling the globe with gas under
standard conditions will be best described later under the head of the
particular gas; but the general arrangement and the connexions with the
pump and the manometer are common to all. They are sketched in Fig. 2, in
Fig. 2.
which S represents the globe, T the inverted bell-glass employed to contain
the enveloping ice. The connexion with the rest of the apparatus is by a
short tube U of thick rubber, carefully wired on. The tightness of these
joints was always tested with the aid of the Toppler X, the tap V leading to
the gas-generating apparatus being closed. The side tube at D leads to the
vacuum chamber of the manometer, while that at E leads to the pressure
chamber B. The wash-out of the tubes, and in some cases of the generator,
was aided by the Toppler. When this operation was judged to be complete,
V was again closed, and a good vacuum made in the parts still connected to
the pump. W would then be closed, and the actual filling commenced by
opening V, and finally the tap of the globe. The lower chamber of the
manometer was now in connexion with the globe, and through a regulating
44 ON THE DENSITIES OF THE PRINCIPAL GASES. [201
tap (not shown) with the gas-generating apparatus. By means of the Toppler
the vacuum in the manometer could be carried to any desired point. But
with respect to this a remark must be made. It is a feature of the method
employed* that the exhaustions of the globe are carried to such a point that
the weight of the residual gas may be neglected, thus eliminating errors due
to a second manometer reading. There is no difficulty in attaining this result,
but the delicacy of the Toppler employed as a gauge is so great that the
residual gas still admits of tolerably accurate measurement. Now in exhaust-
ing the head of the manometer it would be easy to carry the process to a
point much in excess of what is necessary in the case of the globe, but there
is evidently no advantage in so doing. The best results will be obtained by
carrying both exhaustions to the same degree of perfection.
At the close of the filling the pressure has to be adjusted to an exact
value, and it might appear that the double adjustment required (of pressure
and of mercury) would be troublesome. Such was not found to be the case.
After a little practice the manometer could be set satisfactorily without too
great a delay. When the pressure was nearly sufficient, the regulating tap
was closed, and equilibrium allowed to establish itself. If more gas was then
required, the tap could be opened momentarily. The later adjustments were
effected by the application of heat or cold to parts of the connecting tubes.
At the close, advantage was taken of the gradual rise in the temperature
which was usually met with. The pressure being just short of what was
required, and V being closed, it was only necessary to wait until the point
was reached. In no case was a reading considered satisfactory when the
pressure was changing at other than a very slow rate. It is believed that the
comparison between the state of things at the top and at the bottom of
the manometer could be effected with very great accuracy, and this is all that
the method requires. At the moment when the pressure was judged to be
right, the tap of the globe was turned, and the temperature of the manometer
was read. The vacuum was then verified by the Toppler.
The Weights.
The object of the investigation being to ascertain the ratio of densities of
water and of certain gases under given conditions, the absolute values of
the weights employed is evidently a matter of indifference. This is a point
which I think it desirable to emphasise, because v. Jolly, in his, in many
respects, excellent work upon this subject -f-, attributes a discrepancy between
his final result for oxygen and that of Regnault to a possible variation in the
standard of weight. On the same ground we may omit to allow for the
buoyancy of the weights as used in air, since only the variations of buoyancy,
* Due to von Jolly.
t Munich Acad. Trans. Vol. xm. Part n. p. 49, 1880.
1893] ON THE DENSITIES OF THE PRINCIPAL GASES. 45
due, for example, to changing barometer, could enter; and these affect the
result so little that they may safely be neglected*.
But, while the absolute values of the weights are of no consequence, their
relative values must be known with great precision. The investigation of
these over the large range required (from a kilogramme to a centigramme) is
a laborious matter, but it presents nothing special for remark. The weights
quoted in this paper are, in all cases, corrected, so as to give the results as
they would have been obtained from a perfectly adjusted system.
The Water Contents of the Globe.
The globe, packed in finely-divided ice, was filled with boiled distilled
water up to the level of the top of the channel through the plug of
the tap, that is, being itself at 0°, was filled with water also at 0°. Thus
charged the globe had now to be weighed ; but this was a matter of some
difficulty, owing to the very small capacity available above the tap. At about
9° there would be a risk of overflow. Of course the water could be retained
by the addition of extra tubing, but this was a complication that it was
desired to avoid. In February, 1892, during a frost, an opportunity was
found to effect the weighing in a cold cellar at a temperature ranging from 4°
to 7°. The weights required (on the same side of the balance as the globe
and its supports) amounted to O1822 gram. On the other side were other
weights whose values did not require to be known so long as they remained
unmoved during the whole series of operations. Barometer (corrected)
758-9 mm.; temperature 6'3°.
A few days later the globe was discharged, dried, and replaced in the
balance with tap open. 1834'! 701 grams had now to be associated with it in
order to obtain equilibrium. The difference,
1834170 - 0182 = 1833-988,
represents the weight of the water less that of the air displaced by it. The
difference of atmospheric conditions was sufficiently small to allow the neglect
of the variation in the buoyancy of the glass globe and of the brass counter-
poises.
It remains to estimate the actual weight of the air displaced by the water
under the above mentioned atmospheric conditions. It appears that, on this
account, we are to add 2'314, thus obtaining
1836-30
as the weight of the water at 0° which fills the globe at 0°.
* In v. Jolly's calculations the buoyancy of the weights seems to be allowed for in dealing
with the water, and neglected in dealing with the gases. If this be so, the result would be affected
with a slight error, which, however, far exceeds any that could arise from neglecting buoyancy
altogether.
46
ON THE DENSITIES OF THE PRINCIPAL GASES.
[201
A further small correction is required to take account of the fact that the
usual standard density is that of water at 4° and not at 0°. According to
Broch (Everett's C. G. S. System of Units), the factor required is 0-99988, so
that we have
as the weight of water at 4° which would fill the globe at 0°.
Air.
Air drawn from outside (in the country) was passed through a solution of
potash. On leaving the regulating tap it traversed tubes filled with frag-
ments of potash, and a long length of phosphoric anhydride, followed by a
filter of glass wool. The arrangements beyond the regulating tap were the
same for all the gases experimented upon. At the close of the filling it was
necessary to use a condensing syringe in order to force the pressure up to the
required point, but the air thus introduced would not reach the globe. It
may be well to give the results for air in some detail, so as to enable the
reader to form a judgment as to the degree of accuracy attained in the
manipulations.
Date
Globe
empty
Globe
full
Temp, of
manometer
Correction
to 15°
Corrected
to 15°
1892
September 24
2-90941
27
28
29
2-90867
0-53327
0-53271
17-8
15-7
-0-00112
-0-00028
0-53219
0-53243
October 1
2-90923
» 3
0-53151
12-7
+ 0-00093
0-53244
„ 4. . . .
2-90872
„ 7
8
2-91036
Tap regi
0-53296
•eased
12'4
+ 0-00105
0-53401
10
2-91056
11
0-53251
11-8
+ 0-00129
0-53380
» 12
„ 13
2-91039
0-53201
11-0
+ 0-00161
0-53362
„ 14
2-91043
15
0-53219
10*6
+ 0-00177
0-53396
The column headed "globe empty" gives the (corrected) weights, on the
side of the working globe, required for balance. The third column gives the
corresponding weights when the globe was full of air, having been charged at
0° and up to the pressure required to bring the mercury in the manometer
into contact with the two points of the measuring rod.
1893] ON THE DENSITIES OF THE PRINCIPAL GASES. 47
This pressure was not quite the same on different occasions, being subject
to a temperature correction for the density of mercury and for the expansion
of the iron rod. The correction is given in the fifth column, and the weights
that would have been required, had the temperature been 15°, in the sixth.
The numbers in the second and sixth columns should agree, but they are
liable to a discontinuity when the tap is regreased.
In deducing the weight of the gas we compare each weighing " full " with
the mean of the preceding and following weights " empty," except in the case
of October 15, when there was no subsequent weighing empty. The results
are
September 27 2'37686
29 2-37651
October 3 2'37653
8 2-37646
11 2-37668
13 2-37679
15 ., . 2-37647
Mean 2-37661
There is here no evidence of the variation in the density of air suspected
by Regnault and v. Jolly. Even if we include the result for September 27th,
obviously affected by irregularity in the weights of the globe empty, the
extreme difference is only 0'4 milligram, or about l/6000th part.
To allow for the contraction of the globe (No. 14) when weighed empty,
discussed in my former papers, we are to add 0-00056 to the apparent weight,
so that the result for air becomes
2-37717.
This is the weight of the contents at 0° and under the pressure defined by
the manometer gauge at 15° of the thermometer. The reduction to standard
conditions is, for the present, postponed.
Oxygen.
This gas has been prepared by three distinct methods: (a) from chlorates,
(6) from permanganate of potash, (c) by electrolysis.
In the first method mixed chlorates of potash and soda were employed,
as recommended by Shenstone, the advantage lying in the readier fusibility.
The fused mass was contained in a Florence flask, afid during the wash-out
was allowed slowly to liberate gas into a vacuum. After all air had been
expelled, the regulating tap was closed, and the pressure allowed gradually
to rise to that of the atmosphere. The temperature could then be pushed
without fear of distorting the glass, and the gas was drawn off through the
48 ON THE DENSITIES OF THE PRINCIPAL GASES. [201
regulating tap. A very close watch over the temperature was necessary to
prevent the evolution of gas from becoming too rapid. In case of excess, the
superfluous gas was caused to blow off into the atmosphere, rather than risk
imperfect action of the potash and phosphoric anhydride. Two sets of five
fillings were effected with this oxygen. In the first set (May, 1892) the
highest result was 2'6272, and the lowest 2'6266, mean 2'62691. In the
second set (June, July, 1892) the highest result was 2*6273 and the lowest
2-6267, mean 2'62693.
The second method (6) proved very convenient, the evolution of gas being
under much better control than in the case of chlorates. The recrystallised
salt was heated in a Florence flask, the wash-out, in this case also, being
facilitated by a vacuum. Three fillings gave satisfactory results, the highest
being 2'6273, the lowest 2'6270, and the mean 2*62714. The gas was quite
free from smell.
By the third method I have not as many results as I could have wished,
operations having been interrupted by the breakage of the electrolytic
generator. This was, however, of less importance, as I had evidence from
former work that there is no material difference between the oxygen from
chlorates and that obtained by electrolysis. The gas was passed over hot
copper [oxide], as detailed in previous papers. The result of one filling,
with the apparatus as here described, was 2'6271. To this may be added the
result of two fillings obtained at an earlier stage of the work, when the head
of the manometer was exhausted by an independent Sprengel pump, instead
of by the Toppler. The value then obtained was 2'6272. The results stand
thus :—
Electrolysis (2), May, 1892 2'6272
(1) „ 2-6271
Chlorates (5), May, 1892 2'6269
(5), June, 1892 2'6269
Permanganate (3), January, 1893 ... 2'6271
Mean 2-62704
Correction for contraction . . . 0*00056
2-62760
It will be seen that the agreement between the different methods is very
good, the differences, such as they are, having all the appearance of being
accidental. Oxygen prepared by electrolysis is perhaps most in danger of
being light (from contamination with hydrogen), and that from chlorates of
being abnormally heavy.
Nitrogen.
This gas was prepared, in the usual manner, from air by removal of oxygen
with heated copper. Precautions are required, in the first place, to secure a
1893] ON THE DENSITIES OF THE PRINCIPAL GASES. 49
sufficient action of the reduced copper, and, secondly, as was shown by v. Jolly,
and later by Leduc, to avoid contamination with hydrogen which may be
liberated from the copper. I have followed the plan, recommended by v. Jolly,
of causing the gas to pass finally over a length of unreduced copper. The
arrangements were as follows : —
Air drawn through solution of potash was deprived of its oxygen by
reduced copper, contained in a tube of hard glass heated by a large flame. It
then traversed a U-tube, in which was deposited most of the water of combus-
tion. The gas, practically free, as the event proved, from oxygen, was passed,
as a further precaution, over a length of copper heated in a combustion
furnace, then through strong sulphuric acid*, and afterwards back through
the furnace over a length of oxide of copper. It then passed on to the regu-
lating tap, and thence through the remainder of the apparatus, as already
described. In no case did the copper in the furnace, even at the end where
the gas entered, show any sign of losing its metallic appearance.
Three results, obtained in August, 1892, were —
August 8 2-31035
10 2-31026
15 2-31024
Mean 2'31028
To these may be added the results of two special experiments made to
test the removal of hydrogen by the copper oxide. For this purpose a small
hydrogen generator, which could be set in action by closing an external
contact, was included between the two tubes of reduced copper, the gas
being caused to bubble through the electrolytic liquid. The quantity of
hydrogen liberated was calculated from the deflection of a galvanometer
included in the circuit, and was sufficient, if retained, to alter the density
very materially. Care was taken that the small stream of hydrogen should
be uniform during the whole time (about 2| hours) occupied by the filling,
but, as will be seen, the impurity was effectually removed by the copper
oxide f . Two experiments gave —
September 17 2-31012
20 2-31027
Mean 2-31020
We may take as the number for nitrogen —
2-31026
Correction for contraction... 56
2-31082
* There was no need for this, but the acid was in position for another purpose,
t Much larger quantities of hydrogen, sufficient to reduce the oxide over several centimetres,
have been introduced without appreciably altering the weight of the gas.
50 ON THE DENSITIES OF THE PRINCIPAL GASES. [201
Although the subject is not yet ripe for discussion, I cannot omit to
notice here that nitrogen prepared from ammonia, and expected to be pure,
turned out to be decidedly lighter than the above. When the oxygen of air
is burned by excess of ammonia, the deficiency is about I/ 1000th part*.
When oxygen is substituted for air, so that all (instead of about one-seventh
part) of the nitrogen is derived from ammonia, the deficiency of weight may
amount to ^ per cent. It seems certain that the abnormal lightness cannot
be explained by contamination with hydrogen, or with ammonia, or with
water, and everything suggests that the explanation is to be sought in a
dissociated state of the nitrogen itself. Until the questions arising out of
these observations are thoroughly cleared up, the above number for nitrogen
must be received with a certain reserve. But it has not been thought
necessary, on this account, to delay the presentation of the present paper,
more especially as the method employed in preparing the nitrogen for which
the results are recorded is that used by previous experimenters.
Reduction to Standard Pressure.
The pressure to which the numbers so far given relate is that due to
762-511 mm. of mercury at a temperature of 14-85°f, and under the gravity
operative in my laboratory in latitude 51° 47'. In order to compare the
results with those of other experimenters, it will be convenient to reduce
them not only to 760 mm. of mercury pressure at 0°, but also to the value of
gravity at Paris. The corrective factor for length is 760/762'511. In order
to correct for temperature, we will employ the formula J
1 + 0-0001818 1+ 0-00000000017 1*
for the volume of mercury at t°. The factor of correction for temperature is
thus 1-002700. For gravity we may employ the formula
g = 980-6056 - 2-5028 cos 2X,
\ being the latitude. Thus, for my laboratory—
# = 981193,
and for Paris —
g = 980-939,
the difference of elevation being negligible. The factor of correction is thus
0-99974.
The product of the three factors, corrective for length, for temperature,
and for gravity, is accordingly 0'99914. Thus multiplied, the numbers are as
follows : —
Air Oxygen Nitrogen
2-37512 2-62534 2-30883
* Nature, Vol. XLVI. p. 512. [Vol. iv. p. 1.]
t The thermometer employed with the manometer read 0-15° too high.
t Everett, p. 142.
1893]
ON THE DENSITIES OF THE PRINCIPAL GASES.
51
and these may now be compared with the water contents of the globe,
viz., 1836-52.
The densities of the various gases under standard conditions, referred to
that of distilled water at 4°, are thus : —
Air
0-00129327
Oxygen
0-00142952
Nitrogen
0-00125718
With regard to hydrogen, we may calculate its density by means of the
ratio of densities of oxygen and hydrogen formerly given by me, viz., 15'882.
Hence
Hydrogen
0-000090009
The following table shows the results arrived at by various experimenters.
Von Jolly did not examine hydrogen. The numbers are multiplied by 1000
so as to exhibit the weights in grams per litre : —
Air
Oxygen
Nitrogen
Hydrogen
Re°iiault 1847
1-29319
1 '42980
1-25617
0'08958
Corrected by Crafts. . . .
1-29349
1-43011
1-25647
0-08988
Von Jolly, 1880
1-29351
1-42939
1-25787
Ditto corrected
1-29383
1-42971
1-25819
Leduc, 1891* . .
1-29330
1-42910
1-25709
0-08985
Eayleigh, 1893
1-29327
1-42952
1-25718
0-09001
The correction of Regnault by Crafts f represents allowance for the con-
traction of Regnault's globe when exhausted, but the data were not obtained
from the identical globe used by Regnault. In the fourth row I have
introduced a similar correction to the results of von Jolly. This is merely an
estimate founded upon the probability that the proportional contraction
would be about the same as in my own case and in that of M. Leduc.
In taking a mean we may omit the uncorrected numbers, and also that
obtained by Regnault for nitrogen, as there is reason to suppose that his gas
was contaminated with hydrogen. Thus
Mean Numbers.
Air
1-29347
Oxygen
1-42961
Nitrogen
1-25749
Hydrogen
0-08991
The evaluation of the densities as compared with water is exposed to
many sources of error which do not affect the comparison of one gas with
* Bulletin des Seances de la Societe de Physique.
t Comptes Rendug, Vol. cvi. p. 1664.
4—2
52
ON THE DENSITIES OF THE PRINCIPAL GASES.
[201
another. It may therefore be instructive to exhibit the results of various
workers referred to air as unity*.
Oxygen
Nitrogen
Hydrogen
Regnault (corrected)
1-10562
0-97138
G'06949
1-10502
0-97245
Leduc
1-1050
0-9720
0-06947
Eayleigh
1-10535
0-97209
0-06960
1-10525
0-97218
0*06952
As usually happens in such cases, the concordance of the numbers
obtained by various experimenters is not so good as might be expected from
the work of each taken separately. The most serious discrepancy is in the
difficult case of hydrogen. M. Leduc suggests f that my number is too high
on account of penetration of air through the blow-off tube (used to establish
equilibrium of pressure with the atmosphere), which he reckons at 1 m. long
and 1 cm. in diameter. In reality the length was about double, and the
diameter one-half of these estimates ; and the explanation is difficult to
maintain, in view of the fact, recorded in my paper, that a prolongation of
the time of contact from 4m to 30m had no appreciable ill effect. It must be
admitted, however, that there is a certain presumption in favour of a lower
number, unless it can be explained as due to an insufficient estimate of the
correction for contraction. On account of the doubt as to the appropriate
value of this correction, no great weight can be assigned to Regnault's
number for hydrogen. If the atomic weight of oxygen be indeed 15'88, and
the ratio of densities of oxygen and hydrogen be 15'90, as M. Leduc makes
them, we should have to accept a much higher number for the ratio of
volumes than that (2'0002) resulting from the very elaborate measurements
of Morley. But while I write the information reaches me that Mr A. Scott's
recent work upon the volume ratio leads him to just such a higher ratio,
viz., 2-00245, a number a priori more probable than 2'0002. Under the
circumstances both the volume ratio and the density of hydrogen must be
regarded as still uncertain to the I/ 1000th part.
* [1902. Cooke's value for hydrogen, viz. -06958, of date 1889, should have been included in
the above.]
t Comptes Rendw, July, 1892.
1893] ON THE DENSITIES OF THE PRINCIPAL GASES. 53
NOTE A.
On the Establishment of Equilibrium of Pressure in Two Vessels connected by
a Constricted Channel.
It may be worth while to give explicitly the theory of this process, sup-
posing that the difference of pressures is small throughout, and that the
capacity of the channel may be neglected. If vlt p^ denote the volume and
pressure of the gas in the first vessel at time t\ v2, p2 the corresponding
quantities for the second vessel, we have
vldp1/dt + c(pl — p2) = 0,
where c is a constant which we may regard as the conductivity of the channel.
In these equations inertia is neglected, only resistances of a viscous nature
being regarded, as amply suffices for the practical problem. From the above
we may at once deduce
showing that (PI — p2} varies as e~qt, where
c c I
q=~+- =- ,
V1 V2 T
if T be the time in which the difference of pressures is reduced in the ratio
of e : 1.
Let us now apply this result (a) to the case where the globe of volume
v^ communicates with the atmosphere, (6) to the case where the globe is con-
nected with a manometer of relatively small volume vz. For (a) we have
I/T = CK
and for (6) l/r = c/v2;
so that r/r'=vl/v2.
For such a manometer as is described in the text, the ratio v1/v2 is at least
as high as 30 ; and in this proportion is diminished the time required for the
establishment of equilibrium up to any standard of perfection that may be
fixed upon.
[1902. The question of the weight of nitrogen is further treated in
Arts. 210, 214. It will be understood that the results given in the present
paper relate to the atmospheric mixture of nitrogen and argon.]
202.
Fig. 1.
INTERFERENCE BANDS AND THEIR APPLICATIONS.
[Proceedings of the Royal Institution, xiv. pp. 72—78, 1893 ;
Nature, XLVIII. pp. 212—214, 1893.]
THE formation of the interference bands, known as Newton's Rings,
when two slightly curved glass plates are pressed into contact, was illustrated
by an acoustical analogue. A high-pressure flame B (Fig. 1) is sensitive to
sounds which reach it in the direction EB, but is insensitive to similar
sounds which reach it in the nearly perpendicular direction AB. A is
a " bird-call," giving a pure sound (inaudible) of wave-
length (A.) equal to about 1 cm. ; C and D are reflectors
of perforated zinc. If C acts alone, the flame is visibly
excited by the waves reflected from it, though by far the
greater part of the energy is transmitted. If D, held
parallel to G, be then brought into action, the result
depends upon the interval between the two partial re-
flectors. The reflected sounds may co-operate, in which
case the flame flares vigorously; or they may interfere, so
that the flame recovers, and behaves as if no sound at all
were falling upon it. The first effect occurs when the
reflectors are close together, or are separated by any
multiple of ^ V2 • ^ j the second when the interval is
midway between those of the above-mentioned series, that
is, when it coincides with an odd multiple of i\/2.X.
depends upon the obliquity of the reflection.
The coloured rings, as usually formed between glass plates, lose a good
deal of their richness by contamination with white light reflected from the
exterior surfaces. The reflection from the hindermost surface is easily got
rid of by employing an opaque glass, but the reflection from the first surface
is less easy to deal with. One plan, used in the lecture, depends upon the
use of slightly wedge-shaped glasses (2°) so combined that the exterior
surfaces are parallel to one another, but inclined to the interior operative
surfaces. In this arrangement the false light is thrown somewhat to one
The factor V2
1893] INTERFERENCE BANDS AND THEIR APPLICATIONS. 55
side, and can be stopped by a screen suitably held at the place where the
image of the electric arc is formed.
The formation of colour and the ultimate disappearance of the bands
as the interval between the surfaces increases, depends upon the mixed
character of white light. For each colour the bands are upon a scale
proportional to the wave-length for that colour. If we wish to observe
the bands when the interval is considerable — bands of high interference
as they are called — the most natural course is to employ approximately
homogeneous light, such as that afforded by a soda flame. Unfortunately,
this light is hardly bright enough for projection upon a large scale.
A partial escape from this difficulty is afforded by Newton's observations
as to what occurs when a ring system is regarded through a prism. In this
case the bands upon one side may become approximately achromatic, and are
thus visible to a tolerably high order, in spite of the whiteness of the light.
Under these circumstances there is, of course, no difficulty in obtaining
sufficient illumination; and bands formed in this way were projected upon
the screen*.
The bands seen when light from a soda flame falls upon nearly parallel
surfaces have often been employed as a test of flatness. Two flat surfaces
can be made to fit, and then the bands are few and broad, if not entirely
absent; and, however the surfaces may be presented to one another, the
bands should be straight, parallel, and equidistant. If this condition be
violated, one or other of the surfaces deviates from flatness. In Fig. 2,
A and B represent the glasses to be tested, and C is a lens of 2 or 3 feet
focal length. Rays diverging from a soda flame at E are rendered parallel by
the lens, and after reflection from the surfaces are recombined by the lens at
E. To make an observation, the coincidence of the radiant point and its
image must be somewhat disturbed, the one being displaced to a position
a little beyond, and the other to a position a little in front of, the diagram.
The eye, protected from the flame by a suitable screen, is placed at the
image, and being focused upon AB, sees the field traversed by bands. The
reflector D is introduced as a matter of convenience to make the line of
vision horizontal.
These bands may be photographed. The lens of the camera takes the
place of the eye, and should be as close to the flame as possible. With
suitable plates, sensitised by cyanin, the exposure required may vary from
ten minutes to an hour. To get the best results, the hinder surface of A
should be blackened, and the front surface of B should be thrown out of
action by the superposition of a wedge-shaped plate of glass, the intervening
space being filled with oil of turpentine or other fluid having nearly the same
* The theory is given in a paper upon " Achromatic Interference Bands," Phil. Mag. Aug.
1889. [Vol. in. p. 288.]
56
INTERFERENCE BANDS AND THEIR APPLICATIONS.
[202
Fig. 2.
refraction as glass. Moreover, the light should be purified from blue rays by
a trough containing solution of bichromate of potash. With these pre-
cautions the dark parts of the bands are very black, and the exposure may
be prolonged much beyond what would otherwise be admissible.
The lantern slides exhibited showed the elliptical rings indicative of
a curvature of the same sign in
both directions, the hyperbolic
bands corresponding to a saddle-
shaped surface, and the approxi-
mately parallel system due to the
juxtaposition of two telescopic
"flats," kindly lent by Mr Common.
On other plates were seen grooves
due to rubbing with rouge along
a defined track, and depressions,
some of considerable regularity,
obtained by the action of diluted
hydrofluoric acid, which was al-
lowed to stand for some minutes as a drop upon the surface of the glass.
By this method it is easy to compare one flat with another, and thus, if
the first be known to be free from error, to determine the errors of the
second. But how are we to obtain and verify a standard ? The plan
usually followed is to bring three surfaces into comparison. The fact that
two surfaces can be made to fit another in all azimuths proves that they are
spherical and of equal curvatures, but one convex and the other concave, the
case of perfect flatness not being excluded. If A and B fit one another, and
also A and C, it follows that B and C must be similar. Hence, if B and G
also fit one another, all three surfaces must be flat. By an extension of this
process the errors of three surfaces which are not flat can be found from
a consideration of the interference bands which they present when combined
in three pairs.
But although the method just referred to is theoretically complete, its
application in practice is extremely tedious, especially when the surfaces are
not of revolution. A very simple solution of the difficulty has been found in
the use of a free surface of water, which, when protected from tremors and
motes, is as flat as can be desired *. In order to avoid all trace of capillary
curvature it is desirable to allow a margin of about 1£ inch. The surface to
be tested is supported horizontally at a short distance (^ or ^ inch) below
that of the water, and the whole is carried upon a large and massive levelling
stand. By the aid of screws the glass surface is brought into approximate
* The diameter would need to be 4 feet in order that the depression at the circumference,
due to the general curvature of the earth, should amount to ^ X.
1893]
INTERFEKENCE BANDS AND THEIR APPLICATIONS.
57
parallelism with the water. In practice the principal trouble is in the
avoidance of tremors and motes. When the apparatus is set up on the
floor of a cellar in the country, the tremors are sufficiently excluded, but
care must be taken to protect the surface from the slightest draught. To
this end the space over the water must be enclosed almost air-tight. In
towns, during the hours of traffic, it would probably require great precaution
to avoid the disturbing effects of tremors. In this respect it is advantageous
to diminish the thickness of the layer of water; but if the thinning be
carried too far, the subsidence of the water surface to equilibrium becomes
surprisingly slow, and a doubt may be felt whether after all there may not
remain some deviation from flatness due to irregularities of temperature.
Fig. 3.
With the aid of the levelling screws the bands may be made as broad as
the nature of the surface admits; but it is usually better so to adjust the
level that the field is traversed by five or six approximately parallel bands.
Fig. 3 represents bands actually observed from the face of a prism. That
these are not straight, parallel, and equidistant is a proof that the surface
deviates from flatness. The question next arising is to determine the
direction of the deviation. This may be effected by observing the dis-
placement of the bands due to a known motion of the levelling screws ;
but a simpler process is open to us. It is evident that if the surface under
test were to be moved downwards parallel to itself, so as to increase the
thickness of the layer of water, every band would move in a certain direction,
viz. towards the side where the layer is thinnest. What amounts to the
same, the retardation may be increased, without touching the apparatus, by
so moving the eye as to diminish the obliquity of the reflection. Suppose,
for example, in Fig. 3, that the movement in question causes the bands to
travel downwards, as indicated by the arrow. The inference is that the
surface is concave. More glass must be removed at the ends of the bands
than in the middle in order to straighten them. If the object be to
correct the errors by local polishing operations upon the surface, the rule
is that the bands, or any parts of them, may be rubbed in the direction of
the arrow.
A good many surfaces have thus been operated upon ; and although a fair
amount of success has been attained, further experiment is required in order
to determine the best procedure. There is a tendency to leave the marginal
58 INTERFERENCE BANDS AND THEIR APPLICATIONS. [202
parts behind; so that the bands, though straight over the greater part of
their length, remain curved at their extremities. In some cases hydro-
fluoric acid has been resorted to, but it appears to be rather difficult to
control.
The delicacy of the test is sufficient for every optical purpose.
A deviation from straightness amounting to T'^ of a band interval could
hardly escape the eye, even on simple inspection. This corresponds to
a departure from flatness of ^ of a wave-length in water, or about 3^ of
the wave-length in air. Probably a deviation of -^X could be made
apparent.
For practical purposes a layer of moderate thickness, adjusted so that
the two systems of bands corresponding to the duplicity of the soda line do
not interfere, is the most suitable. But if we wish to observe bands of high
interference, not only must the thickness be increased, but certain pre-
cautions become necessary. For instance, the influence of obliquity must '
be considered. If this element were absolutely constant, it would entail no
ill effect. But in consequence of the finite diameter of the pupil of the eye,
various obliquities are mixed up together, even if attention be confined to
one part of the field. When the thickness of the layer is increased, it
becomes necessary to reduce the obliquity to a minimum, and further to
diminish the aperture of the eye by the interposition of a suitable slit. The
effect of obliquity is shown by the formula
[2^t cos & = n\].
The necessary parallelism of the operative surfaces may be obtained, as in
the above described apparatus, by the aid of levelling. But a much simpler
device may be employed, by which the experimental difficulties are greatly
reduced. If we superpose a layer of water upon a surface of mercury, the
flatness and parallelism of the surfaces take care of themselves. The
objection that the two surfaces would reflect very unequally may be obviated
by the addition of so much dissolved colouring matter, e.g. soluble aniline
blue, to the water as shall equalise the intensities of the two reflected lights.
If the adjustments are properly made, the whole field, with the exception of
a margin near the sides of the containing vessel, may be brought to one
degree of brightness, being in fact all included within a fraction of a band.
The width of the margin, within which rings appear, is about one inch, in
agreement with calculation founded upon the known values of the capillary
constants. During the establishment of equilibrium after a disturbance,
bands are seen due to variable thickness, and when the layer is thin, they
persist for a considerable time.
When the thickness of the layer is increased beyond a certain point, the
difficulty above discussed, depending upon obliquity, becomes excessive, and
it is advisable to change the manner of observation to that adopted by
1893] INTERFERENCE BANDS AND THEIR APPLICATIONS. 59
Michelson*. In this case the eye is focused, not, as before, upon the
operative surfaces, but upon the flame, or rather upon its image at E
(Fig. 2). For this purpose it is only necessary to introduce an eye-piece
of low power, which with the lens C (in its second operation) may be
regarded as a telescope. The bands now seen depend entirely upon obliquity
according to the formula above written, and therefore take the form of
circular arcs. Since the thickness of the layer is absolutely constant, there
is nothing to interfere with the perfection of the bands except want of
homogeneity in the light.
But, as Fizeau found many years ago, the latter difficulty soon becomes
serious. At a very moderate thickness it becomes necessary to reduce the
supply of soda, and even with a very feeble flame a limit is soon reached.
When the thickness was pushed as far as possible, the retardation, calculated
from the volume of liquid and the diameter of the vessel, was found to be
50,000 wave-lengths, almost exactly the limit fixed by Fizeau.
To carry the experiment further requires still more homogeneous sources
of light. It is well known that Michelson has recently observed interference
with retardations previously unheard of, and with the aid of an instrument of
ingenious construction has obtained most interesting information with respect
to the structure of various spectral lines.
A curious observation respecting the action of hydrofluoric acid upon
polished glass surfaces was mentioned in conclusion. After the operation of
the acid the surfaces appear to be covered with fine scratches, in a manner
which at first suggested the idea that the glass had been left in a specially
tender condition, and had become scratched during the subsequent wiping.
But it soon appeared that the effect was a development of scratches previously
existent in a latent state. Thus parallel lines ruled with a knife-edge, at first
invisible even in a favourable light, became conspicuous after treatment with
acid. Perhaps the simplest way of regarding the matter is to consider the
case of a furrow with perpendicular sides and a flat bottom. If the acid may
be supposed to eat in equally in all directions, the effect will be to broaden
the furrow, while the depth remains unaltered. It is possible that this
method might be employed with advantage to intensify (if a photographic
term may be permitted) gratings ruled upon glass for the formation of
spectra.
* [1902. The influence of the diameter of the pupil of the eye in lessening the visibility of
fringes dependent primarily upon variable thickness, seems to have been first pointed out by
Lummer (Wied. Ann. xxm. p. 4'J, 1884), who also emphasised the advantages attending the use of
a plate of uniform thickness and of rings dependent solely upon obliquity, whether the object
be the investigation of high interference itself, or the examination for uniformity of plates
intended to be plane-parallel.
The circular ring system dependent upon obliquity was first observed by Haidinger (Pogg.
Ann. LXXVII. p. 219, 1849) and explained by Mascart (Ann. de Chim. xxm. p. 116, 1871).]
203.
ON THE THEORY OF STELLAR SCINTILLATION.
[Philosophical Magazine, xxxvi. pp. 129—142, 1893.]
ARAGO'S theory of this phenomenon is still perhaps the most familiar,
although I believe it may be regarded as abandoned by the best authorities.
According to it the momentary disappearance of the light of the star is due
to accidental interference between the rays which pass the two halves of the
pupil of the eye or the object-glass of the telescope. When the relative
retardation amounts to an odd multiple of the half wave-length of any kind
of light, such light, it is argued, vanishes from the spectrum of the star.
, But this theory is based upon a complete misconception. " It is as far as
possible from being true that a body emitting homogeneous light would
disappear on merely covering half the aperture of vision with a half wave
plate. Such a conclusion would be in the face of the principle of energy,
which teaches plainly that the retardation in question would leave the
aggregate brightness unaltered*." It follows indeed from the principle of
interference that there will be darkness at the precise point which before the
introduction of the half wave plate formed the centre of the image, but the
light missing there is to be found in a slightly displaced position f.
* Enc. Brit., " Wave Theory," p. 441. [Vol. ra. p. 123.]
t Since the remarks in the text were written I have read the version of Arago's theory given
by Mascart (Traite d'Optique, t. in. p. 348). From this some of the most objectionable features
have been eliminated. But there can be no doubt as to Arago's meaning. " Supposons que les
rayons qui tombent a gauche du centre de 1'ohjectif aient rencontre, depuis les limites superieures
de I'atmosphere, des couches qui, a cause de leur densite, de leur temperature, ou de leur etat
hygrometrique, etaient douees d'une refringence differente de celle que possedaient les conches
traversees par les rayons de droite ; il pourra arriver, qu'a raison de cette difference de refringence,
les rayons rouges de droite detrnisent en totalite les rayons rouges de gauche, et que le foyer
passe du blanc, son etat normal, an vert Voila done le resnltat theorique par fai lenient
d'accord avec les observations ; voila le phenomena de la scintillation dans nne lunette rat t ache
d'une maniere intime a la doctrine des interferences" (I'Annuaire du Bureau des Longitude* pour
1852, pp. 423, 425).
That the difference between Arago's theory and that followed in the present paper is funda-
mental will be recognized when it is noticed that, according to the former, the colour effects of
scintillation would be nearly independent of atmospheric dupersion. Arago gives an interesting
summary of the views held by early writers.
1893] ON THE THEORY OF STELLAR SCINTILLATION. 61
The older view that scintillation is due to the actual diversion of light
from the aperture of vision by atmospheric irregularities was powerfully
supported by Montigny*, to whom we owe also a leading feature of the true
theory, that is, the explanation of the chromatic effects by reference to the
different paths pursued by rays of different colours in virtue of regular
atmospheric dispersion. The path of the violet ray lies higher than that of
the red ray which reaches the eye of the observer from the same star, and the
separation may be sufficient to allow the one to escape the influence of an
atmospheric irregularity which operates upon the other. In Montigny 's view
the diversion of the light is caused by total reflexion at strata of varying
density.
But the most important work upon this subject is undoubtedly that of
Respighi-f-, who, following in the steps of Montigny and Wolf, applied the
spectroscope to the investigation of stellar scintillation. The results of these
observations are summed up under thirteen heads, which it will be convenient
to give almost at full length.
(I.) In spectra of stars near the horizon we may observe dark or bright
bands, transversal or perpendicular to the length of the spectrum, which more
or less quickly travel from the red to the violet or from the violet to the red,
or oscillate from one to the other colour ; and this however the spectrum may
be directed from the horizontal to the vertical.
(II.) In normal atmospheric conditions the motion of the bands proceeds
regularly from red to violet for stars in the west, and from violet to red for
stars in the east ; while in the neighbourhood of the meridian the movement
is usually oscillatory, or even limited to one part of the spectrum.
(III.) In observing the horizontal spectra of stars more and more elevated
above the horizon, the bands are seen sensibly parallel to one another, but
more or less inclined to the axis of the spectrum, passing from red to violet or
reversely according as the star is in the west or the east.
(IV.) The inclination of the bands, or the angle formed by them with
the axis (? transversal) of the spectrum, depends upon the height of the star ;
it reduces to 0° at the horizon and increases rapidly with the altitude so as to
reach 90° at an elevation of 30° or 40°, so that at this elevation the bands
become longitudinal.
(V.) The inclination of the bands, reckoned downwards, is towards the
more refrangible end of the spectrum.
(VI.) The bands are most marked and distinct when the altitude of the
star is least. At an altitude of more than 40° the longitudinal bands are
reduced to mere shaded streaks, and often can only be observed upon the
spectrum as slight general variations of brightness.
* Mem. de VAcad. d. Bruxelles, i. xxvin. (1856).
t Roma, Atti Nuovi Lincei, xxi. (1868) ; Assoc. Fran$aise, Compt. Rend. i. (1872), p. 169.
62 ON THE THEORY OF STELLAK SCINTILLATION. [203
(VII.) As the altitude increases, the movement of the bands becomes
quicker and less regular.
(VIII.) As the prism is turned so as to bring the spectrum from the
horizontal to the vertical position, the inclination of the bands to the
transversal of the spectrum continually diminishes until it becomes zero when
the spectrum is nearly vertical; but the bands then become less marked,
retaining, however, the movement in the direction indicated above (III.)-
(IX.) Luminous bands are less frequent and less regular than dark
bands, and occur well marked only in the spectra of stars near the horizon.
(X.) In the midst of this general and violent movement of bright and
dark masses in the spectra of stars, the black spectral lines proper to the
light of each star remain sensibly quiescent or undergo very slight oscil-
lations.
(XI.) Under abnormal atmospheric conditions the bands are fainter and
less regular in shape and movement.
(XII.) When strong winds prevail the bands are usually rather faint
and ill defined, and then the spectrum exhibits mere changes of brightness,
even in the case of stars near the horizon.
(XIII.) Good definition and regular movement of the bands seems to be
a sign of the probable continuance of fine weather, and, on the other hand,
irregularity in these phenomena indicates probable change.
These results show plainly that the changes of intensity and colour in the
images of stars are produced by a momentary real diversion of the luminous
rays from the object-glass of the telescope ; that in the neighbourhood of the
horizon rays of different colours are affected separately and successively, and
that all the rays of a given colour are momentarily withdrawn from the whole
of the object-glass.
Most of his conclusions from observation were readily explained by
Respighi as due to irregular refractions, not necessarily or usually amounting
(as Montigny supposed) to total reflexions, taking place at a sufficient distance
from the observer. The progress of the bands in one direction along the
spectrum (II.) is attributed to the diurnal motion. In the case of a setting
star, for instance, the blue rays by which it is seen, pursuing a higher course
through the atmosphere, encounter an obstacle somewhat later than do the
red rays. Hence the band travels towards the violet end of the spectrum.
In the neighbourhood of the meridian this cause of a progressive movement
ceases to operate.
The observations recorded in (III.) are of special interest as establishing a
connexion between the rates with which various parts of the object-glass and
of the spectrum are affected. Since the spectrum is horizontal, various parts
1893] ON THE THEORY OF STELLAR SCINTILLATION. 63
of its width correspond to various horizontal sections of the objective, and
the existence of bands at a definite inclination shows that at the moment
when the shadow of the obstacle thrown by blue rays reaches the bottom
of the glass the shadow at the top is that thrown by green, yellow, or red
rays of less refrangibility. When the altitude of the star reaches 30° or
40°, the difference of path due to atmospheric dispersion is insufficient to
differentiate the various parts of the spectrum. The bands then appear
longitudinal.
The definite obliquity of the bands at moderate altitudes, reported by
Respighi, leads to a conclusion of some interest, which does not appear to
have been noticed. In the case of a given star, observed at a given altitude,
the linear separation at the telescope of the shadows of the same obstacle
thrown by rays of various colours will of necessity depend upon the distance
of the obstacle. But the definiteness of the obliquity of the bands requires
that this separation shall not vary, and therefore that the obstacles to which
the effects are due are sensibly at one distance only. It would seem to follow
from this that, under " normal atmospheric conditions," scintillation depends
upon irregularities limited to a comparatively narrow horizontal stratum
situated overhead. A further consequence will be that the distance of the
obstacles increases as the altitude of the star diminishes, and this according
to a definite law.
The principal object of the present communication is to exhibit some
of the consequences of the theory of scintillation in a definite mathematical
form. The investigation may be conducted by simple methods, if, as suffices
for most purposes, we regard the whole refraction as small, and neglect the
influence of the earth's curvature. When the object is to calculate with
accuracy the refraction itself, further approximations are necessary, but even
in this case the required result can be obtained with more ease than is
generally supposed.
The foundation upon which it is most convenient to build is the idea of
James Thomson*, which establishes instantaneously the connexion between
the curvature of a ray travelling in a medium of varying optical constitution
and the rate at which the index changes at the point in question. The
following is from Everett's memoir: —
" Draw normal planes to a ray at two consecutive points of its path.
Then the distance of their intersection from either point will be p, the radius
of curvature. But these normal planes are tangential to the wave-front in
its two consecutive positions. Hence it is easily shown by similar triangles
that a very short line dN drawn from either of the points towards the centre
of curvature is to the whole length p, of which it forms part, as do the
* Brit. Assoc. Eep. 1872. Everett, Phil. Mag. March 1873.
64 ON THE THEORY OF STELLAR SCINTILLATION. [203
difference of the velocities of light at its two ends is to v the velocity at
either end. That is
dN/p = - dv/v,
the negative sign being used because the velocity diminishes in approaching
the centre of curvature. But, since v varies inversely as //,, we have
— dv/v = dpi p.
Hence the curvature 1 /p is given by any of the four following expressions : —
1 1 dv _ d log v _ 1 dfj, _d log //,
~P^~vdN dN~=~jj,dN = ~dN~' '
" The curvatures of different rays at the same point are directly as the
rates of increase of JJL in travelling along their respective normals." If 6
denote the angle which the ray makes with the direction of most rapid
increase of index, the curvatures will be directly as the values of sin 6. In
fact, if dfi/dr denote the rate at which //, increases in a direction normal to
the surfaces of equal index, we have
da dfj, . a
-j^r = -r- sin 6,
dN dr
and therefore
1 1 da . n d log a . f. ,_,
- = - --sm#= 8"sm0 ...................... (2)
,
~ p p. r dr
Everett shows how the well-known equation
ftp = const ..................................... (3)
can be deduced from (2), p being the perpendicular upon the ray from the
centre of spherical surfaces of equal index. In general,
I I dp a p
- = -/-, sin 6 = *- ,
p r dr r
and thus
1 dp _ p d log fj,
r dr r dr
giving (3) on integration.
At a first application of (2) we may find by means of it a first ap-
proximation to the law of atmospheric refraction, on the supposition that
the whole refraction is small and that the curvature of the earth may be
neglected. Under these limitations B in (2) may be treated as constant
along the whole path of the ray ; and if dty be the angle through which
the ray turns in describing the element of arc ds, we have
d-*fr = °M sin 6 ds = tan 6 . d log p.
1893] ON THE THEORY OF STELLAR SCINTILLATION. 65
If we integrate this along the whole course of the ray through the
atmosphere, that is from p, = 1 to /A = /i0, we get, as the whole refraction,
^ = log /*„ tan 0 = 0*0-1) tan 0, ..................... (4)
for to the order of approximation in question log /*„ may be identified with
0"o-l).
If Si/r denote the chromatic variation of ^ corresponding to 8jj,0, we have
from (4)
-l) ............................ (5)
According to Mascart* the value of the right-hand member of (5) in
the case of air and of the lines B and H is
^0/0*0-1) = -024 ............................... (6)
We will now take a step further and calculate the linear deviation of
a ray from a straight course, still upon the supposition that the whole
refraction is small. If rj denote the linear deviation (reckoned perpen-
dicularly) at any point defined by the length s measured along the ray 0,
we have
,
ds
so that
~ = ltan0c?log/,t = tan0(yLt— 1) + a,
a being a constant of integration. A second integration now gives
17 = tan 0j(fi-l)ds + as + {3, ..................... (7)
which determines the path of the ray. If y be the height of any point
above the surface of the earth, ds = dy sec 0; so that (7) may also be written
The origin of s is arbitrary, but we may conveniently take it at the point
(A ) where the ray strikes the earth's surface.
We will now consider also a second ray, of another colour, deviating
from the line 0 by the distance 77 + 877, and corresponding to a change of /*
to /i + 8fji. The distance between the two rays at any point y is
(9)
In this equation &@ denotes the separation of the rays at A , where y = 0,
* Everett's C, G. S. System of Units.
QQ ON THE THEORY OF STELLAR SCINTILLATION. [203
s = 0. And 8a denotes the angle between the rays when outside the atmo-
sphere.
Equation (9) may be applied at once to Montigny's problem, that is to
determine the separation of two rays of different colours, both coming from
the same star, and both arriving at the same point A. The first condition
gives Sot. = 0, and the second gives S/3 = 0. Accordingly,
is the solution of the question.
The integral in (10) may be otherwise expressed by means of the principle
that (/z — 1) and Sp are proportional to the density. Thus, if I denote the
" height of the homogeneous atmosphere," and h the elevation in such an
atmosphere determined by the condition that there shall be as much air
below it as actually exists below y,
(11)
S/AO being the value of S/A at the surface of the earth. Equation (10) thus
becomes
At the limits of the atmosphere and beyond, h = I, and the separation there is
cos'0 ...............................
These results are applicable to all altitudes higher than about 10°.
The formulae given by Montigny (loc. cit.) are quite different from the
above. That corresponding to (13) is
By = S/i0asin 0, .............................. (14)
a being the radius of the earth ! The substitution of a for I increases the
calculated result some 800 times. But this is in a large measure compen-
sated by the factor sec2 6 in (13), for at low altitudes sec 0 is large. According
to Montigny the separation at moderately low altitudes would be nearly in-
dependent of the altitude, a conclusion entirely wide of the truth.
The value of (/*»-!) for air at 0° and 760 millim. at Paris is "0002927,
so that S/*0 (for the lines E and H) is "000007025. The height of the
homogeneous atmosphere is 7'990 x 105 centim., and thus &? reckoned in
centim. is
1893] ON THE THEORY OF STELLAR SCINTILLATION.
The following are a few corresponding values of 6 and sin 0/cos20
67
e
sin 0/cos2 6
e
sin 0/cos2 0
0
0
o-ooo
60°
3-46
20
0-387
70
8-03
40
1-095
80
32-66
Thus at the limit of the atmosphere the separation of rays which reach
the observer at an apparent altitude of 10° is 185 centim. Nearer the
horizon the separation would be still greater, but its value cannot well be
found from (15). Although these estimates are considerably less than those
of Montigny, the separation near the horizon seems to be sufficient to
explain the vertical position of the bands in the spectrum, recorded by
Respighi (I.). The fact that the margin is not very great suggests that the
obstacles to which scintillation is due may often be situated at a considerable
elevation.
We have now to consider the effect of an obstacle situated at a given
point B at level y on the course of the ray. And the first desideratum will
be the estimation of the separation at A, the object-glass of the telescope,
of rays of various colours corning from the same star, which all pass through
the given point B. It will appear at once that no fresh question is raised.
For, since the rays come from the same star at the same time, 8a — 0, and
thus by (9) 8-rjA = S/3. The value of £/3 is given at once by the condition
that &i)B = 0. Thus
as before. The discussion, already given of (15), is thus immediately ap-
plicable.
Equation (16) solves the problem of determining the inclination of the
bands seen in the spectra of stars not very low (III.). It is only necessary
to equate — 8rjA to the aperture of the telescope. fy*0 then gives the range
of refrangibility covered by the bands as inclined. In practice h would not
be known beforehand; but from the observed inclination of the bands it
would be possible to determine it.
In a given state of the atmosphere h, so far as it is definite, must be
constant and then B/JLO must be proportional to cos2 6 1 sin 6. This gives the
relation between the altitude of the star and the inclination of the bands.
When 6 is small, fyi0 is large ; that is, the bands become longitudinal.
5—2
68 ON THE THEORY OF STELLAR SCINTILLATION. [203
As a numerical example, let us suppose that the aperture of the telescope
is 10 centim., and that at an altitude of 10° the obliquity of the bands is
such that the vertical diameter of the object-glass corresponds to the entire
range from B to H. In this case (15) gives
, 10 1
indicating that the obstacles to which the bands are due are situated at such
a level that about -£$ of the whole mass of the atmosphere is below them.
The next question to which (9) may be applied is to find the angle Set
outside the atmosphere between two rays of different colours which pass
through the two points A and B. Here &r]A = 0, and thus 8/3 = 0. And
further, since BrjB = 0, we get
sin 0 [u ~ j &/j,0h tan 6
If the height of the obstacle above the ground be so small that the
density of the air below it is sensibly uniform, then h = y, and
- Sa = S/-iotan0 ............................... (18)
In this case the angle is the same as that of the spectrum of the star
observed at A, as appears from (4) and (5). In general, y is greater than h,
so that So. is somewhat less than the value given by (18).
The interest of (18) lies in the application of it to find the time occupied
by a band in traversing the spectrum in virtue of the diurnal motion, ac-
cording to Respighi's observation (II.). The time required is that necessary
for the star to rise or fall through the angle of its dispersion-spectrum at
the altitude in question. At an altitude of 10°, this angle will be 8", being
always about ^ of the whole refraction. The rate at which a star rises or
falls depends of course upon the declination of the star and upon the latitude
of the observer, and may vary from zero to 15° per hour. At the latter
maximum rate the star would describe 8" in about one half of a second,
which would therefore be the time occupied by a band in crossing the
spectrum under the circumstances supposed. In the case of a star quite
close to the horizon, the progress of the band would be a good deal slower.
The fact that the larger planets scintillate but little, even under favour-
able conditions, is readily explained by their sensible apparent magnitude.
The separation of rays of one colour thus arising during their passage through
the atmosphere is usually far greater than the already calculated separation,
due to chromatic dispersion ; so that if a fixed star of no apparent magnitude
scintillates in colours, the different parts of the area of a planet must a
fortiori scintillate independently. But under these circumstances the eye
perceives only an average effect, and there is no scintillation visible.
1893] ON THE THEORY OF STELLAR SCINTILLATION. 69
The non-scintillation of small stars situated near the horizon may be
referred to the failure of the eye to appreciate colour when the light is faint.
In the case of stars higher up, the whole spectrum is affected simul-
taneously. A momentary accession of illumination, due to the passage of
an atmospheric irregularity, may thus render visible a star which on account
of its faintness could not be steadily seen through an undisturbed atmo-
sphere *.
In the preceding discussion the refracting obstacles have for the sake of
brevity been spoken of as throwing sharp shadows. This of course cannot
happen, if only in consequence of diffraction ; and it is of some interest to
inquire into the magnitude of the necessary diffusion. The theory of
diffraction shows that even in the case of an opaque screen with a definite
straight boundary, the transition of illumination at the edge of the shadow
occupies a space such as ^/(b\), where X is the wave-length of the light, and
b is the distance across which the shadow is thrown. We may take X at
6 x 10~5 centim., and if 6 be reckoned in kilometres, we have as the space
of transition, \'(6b). Thus if b were 4 kilometres, the space of transition
would amount to about 5 centim. The inference is that the various parts of
the aperture of a small telescope cannot be very differently affected unless
the obstacles to which the scintillation is due are at a less distance than
4 kilometres.
One of the principal outstanding difficulties in the theory of scintillation
is to see how the transition from one index to another in an atmospheric
irregularity can be sufficiently sudden. The fact that the various parts of a
not too small object-glass are diversely affected seems to prove that the
transitions in question do not occupy many centimetres. Now, whether the
irregularity be due to temperature or to moisture, we should expect that a
transition, however abrupt at first, would after a few minutes or hours be
eased off to a greater degree than would accord with the above estimate.
Perhaps the abruptness of transition is, as it were, continually renewed by
the coming into contact of fresh portions of light and dense air as the
ascending and descending streams proceed in their courses. The speculations
and experiments of Jevons on the Cirrus form of Cloud f may find some
application here. A preliminary question requiring attention is as to the
origin of the irregularities which cause scintillation. Is it always at the
ground, and mainly under the influence of sunshine ? Or may irregular
absorption of solar heat in the atmosphere, due to varying proportions of
moisture, give rise to transitions of the necessary abruptness ? Again, we
may ask how many obstacles are to be supposed operative upon the same
* The theory of Arago leads him to a directly opposite conclusion (loc. cit. p. 381).
t Phil. Mag. xiv. p. 22, 1857. For a mathematical investigation, by the author, see Math.
Soc. Proc. xiv. April 1883. [Vol. n. p. 200.]
70
ON THE THEORY OF STELLAR SCINTILLATION.
[203
ray ? Is the ultimate effect only a small residue from many causes in the
main neutralizing one another? It does not appear that in the present
state of meteorological science satisfactory answers can be given to these
questions.
A complete investigation of atmospheric refraction can only be made
upon the basis of some hypothesis as to the distribution of temperature ; but,
as has already been hinted, a second approximation to the value of the
refraction can be obtained independently of such knowledge and without
difficulty. In Laplace's elaborate investigation it is very insufficiently recog-
nized, if indeed it be recognized at all, that the whole difficulty of the
problem depends upon the curvature of the earth. If this be neglected,
that is if the strata are supposed to be plane, the desired result follows at
once from the law of refraction, without the necessity of knowing anything
more than the condition of affairs at the surface. For in virtue of the law
of refraction,
fju sin 0 = constant ;
so that if 6 be the apparent zenith distance of a star seen at the earth's
surface, and 80 the refraction, we have at once
/*o sin 6 = sin (6 + 86), (19)
from which the refraction can be rigorously calculated. If an expansion be
desired,
B6 = sin Be = tan 0 (>0 - cos 80)
= (/*„- 1) tan 0{l+£O0-l)tan20} (20)
is the second approximation.
When the curvature of the earth is retained, so that the atmospheric
strata are supposed to be spheres described round 0 the centre of the earth,
the appropriate form of the law of refraction is
ftp = constant.
Thus, if A be the point of observation at the earth's surface where the
1893] ON THE THEORY OF STELLAR SCINTILLATION. 71
apparent zenith distance is 0, and if the original direction of the ray outside
the atmosphere meet the vertical OA at the point Q,
1^.0 A. sin 6 = OQ . sin (6 + BO) ;
or if OJ. = a, AQ = c,
fjt^a sin 6 = (a + c) sin (0 + 86) ...................... (21)
If c be neglected altogether, we fall back upon the former equations (19),
(20). For the purposes of a second approximation c, though it cannot be
neglected, may be calculated as if the refraction were small, and the curvature
of the strata negligible. If 77 be the whole linear deviation of the ray due
to the refraction,
c = f)fsm0, ................................. (22)
and, as in (16),
(23)
so that c = ° ~ " . ....(24)
cos2 6
By equations (21), (24) the value of 80 may be calculated from the trigono-
metrical tables without further approximation.
To obtain an expansion, we have
+cja
O/o - 1) tan 0 l - - + i G-O - 1) tan'
tatf0 (25)
To this order of approximation the refraction can be expressed in terms of
the condition of things at the earth's surface, and (25) is equivalent to an
expression deduced at great length by Laplace.
From the value of I already quoted, and a = 6'3709 x 108 centim., we get
I/a = -0012541 (26)
If further we take as the value under standard conditions for the line D
^0-1 =-0002927, ~ (27)
we find as the refraction expressed in seconds of arc
86 = 60"-29 tan 0 - 0"'06688 tan80 (28)
In (28) 6 is the apparent zenith distance, and it should be understood
that the application of the formula must not be pushed too close to the
72
ON THE THEORY OF STELLAR SCINTILLATION.
[203
horizon. If the density of the air at the surface of the earth differ from
the standard density (0° and 760 millim.) the numbers in (28) must be
altered proportionally. It will be observed that the result has been deduced
entirely a priori on the basis of data obtained in laboratory experiments.
It may be convenient for reference to give a few values calculated from
(28) of the refraction, and of the dispersion, reckoned at ^ of the refraction.
Apparent zenith
distance
Refraction
Dispersion
(B to H)
6
o-o
o'-b
20
21-9
•5
40
50-5
1-3
45
i 0-2
1-5
60
1 40-1
2-5
70
2 44-2
4-1
75
3 41 5
5-5
80
5 29-7
8-2
85
9 49-2
14-7
The results of the formula (28) agree with the best tables up to a zenith
distance of 75°, at which point the value of the second term is 3"*5. For
85° the number usually given is about 10' 16", and for 90° about 36'; but
at these low altitudes the refraction is necessarily uncertain on account of
irregularities such as those concerned in the production of mirage.
[1902. When the above memoir was written I was unacquainted with
the excellent work of K. Exner upon this subject (Wien. Akad. Sitzber. 84,
p. 1038, 1881; Uber die Scintillation. Eine Monographic. Wien, 1891).
Exner's general conclusions are very similar to those at which I arrived
independently, and his work includes also an interesting historical discussion.]
204.
ASTRONOMICAL PHOTOGRAPHY.
[Nature, XLVIII. p. 391, 1893.]
THE announcement (Nature, August 10) that it is in contemplation to
raise a sum exceeding £2000 for the establishment of a special photographic
telescope at the Cambridge Observatory, leads me to ask whether astronomers
have duly considered the facilities afforded by modern photography. At the
time of my early experience of the art, thirty-five years ago, it would have
been thought a great feat to photograph the Fraunhofer lines in the yellow
or red regions of the spectrum, although even then the statement so
commonly made that chemical activity was limited to the blue and ultra-
blue rays was quite unwarranted. With the earlier photographic processes
the distinction was necessary between telescopes to be used with the eye or
for photography. In the former case the focal length had to be a minimum
for the yellow rays, in the latter for the blue rays of the spectrum.
But the situation is entirely changed. There is now no difficulty in
preparing plates sensitive to all parts of the spectrum, witness the beautiful
photographs of Rowland and Higgs. I have myself used " Orthochromatic "
plates in experiments where it was desirable to work with the same rays as
most influence the eye. The interference bands of sodium light may be
photographed with the utmost facility on plates sensitised in a bath con-
taining cyanin.
The question I wish to ask is whether the time has not come to
accommodate the photographic plates to the telescopes rather than the
telescopes to the plates. It is possible that plates already in the market
may not exactly meet the requirements of the case, but I feel sure that
a tithe of the sums lavished upon instruments would put us in possession of
plates suitable for object-glasses that have been designed for visual purposes.
There would be no difficulty even in studying the requirements of a particular
instrument, over or under corrected as the case might be.
A doubt may arise whether plates so adjusted would be as sensitive as
those now in use. Probably Captain Abney, or some other authority, could
give the required information. For some astronomical purposes a moderate
loss of sensitiveness could hardly be of much consequence; for others
doubtless it would be a serious matter.
205.
GRINDING AND POLISHING OF GLASS SURFACES.
[British Association, Sept. 14, 1893, from a report in
Nature, XLVIII. p. 526, 1893.]
LORD RAYLEIGH stated that he had been investigating the nature of these
processes, and gave a most interesting description of the results. He first
pointed out that the process of grinding with emery is not, as is commonly
supposed, a scratching process. The normal effect is the production of isolated
detached pits — not scratches. The glass gives way under the emery ; at the
same time the emery gives way under the glass and suffers abrasion. An
image seen through glass which has been finely ground (but not yet polished)
has perfect definition. And so when the sun is viewed through a cloud the
image is sharp as long as there is an image ; even when the cloud thickens,
the edge appears to be sharp until we lose the image altogether. A glass
lens finely ground gives very good definition, but there is great loss of light
by irregular reflection. To obviate this, the lens is polished, and examination
under the microscope shows that in the process of polishing with pitch and
rouge the polishing goes on entirely on the surface or plateau, the bottom
of each pit being left untouched until the adjoining surface is entirely
worked down to it. It appeared interesting to investigate the amount of
glass removed during the process of polishing. This was done both by
weighing and interference methods, and the amount removed was found to
be surprisingly small. A sufficiently good polish was obtained when a
thickness corresponding to 2^- wave-lengths of sodium light was removed,
and the polishing was complete when a thickness corresponding to 4 wave-
lengths was removed. Lord Rayleigh is of opinion that the process of
polishing is not continuous with that of grinding, but that it consists of a
removal of molecular layers of the surface of the glass. Grinding is easy
and rapid, whereas polishing is tedious and difficult. The action of hydro-
fluoric acid in dissolving glass was also investigated and was found to be
much more regular than it has generally been assumed to be by chemists. It
was found to be easy to remove a layer corresponding in thickness to half a
wave-length of sodium light ; and with due precautions as little as one-tenth
of a wave-length. [1902. For a further discussion of this subject see Nature,
LXIV. p. 385, 1901.]
206.
ON THE REFLECTION OF SOUND OR LIGHT FROM A
CORRUGATED SURFACE.
[British Association Report, pp. 690, 691, 1893.]
THE angle of incidence is supposed to be zero, and the amplitude of the
incident wave to be unity. If then
£=ccosjtw? ................................. (1)
be the equation of the surface, the problem of reflection is readily solved
so long as p in (1) is small relatively to k or 2?r/X; that is so long as the
wave-length of the corrugation is large in comparison with that of the
vibrations. The solution assumes a specially simple form when the second
medium is impenetrable, so that the whole energy is thrown back either in
the perpendicularly reflected wave or in the lateral spectra. Of this two
cases are notable (a) when — in the application to sound — the second medium
is gaseous and devoid of inertia, as in the theory of the 'open ends' of
organ pipes. The amplitude A0 of the perpendicularly reflected wave, so
far as the fourth power of p/k inclusive, is then given by
- A0 = o . .
in which there is no limitation upon the value of kc, so that the corrugation
may be as deep as we please in relation to X. If p be very small, the result,
viz. — Jr0(2/cc), is the same as would be obtained by the methods usual in
Optics; and it appears that these methods cease to be available when p
cannot be neglected.
The second case (/3) arises when sound is reflected from a rigid and fixed
wall. We find, as far as p*/k?,
If p, instead of being relatively small, exceeds k in magnitude, there are no
lateral spectra in the reflected vibrations; and if the second medium is
impenetrable, the regular reflection is necessarily total. It thus appears
that an extremely rough wall reflects sounds of medium pitch as well as
if it were mathematically smooth.
The question arises whether, when the second medium is not impenetrable,
the regular reflection from a rough wall (p > k) is the same as if c = 0.
Reasons are given for concluding that the answer should be in the negative.
207.
ON A SIMPLE INTERFERENCE ARRANGEMENT.
[British Association Report, pp. 703, 704, 1893.]
IF a point, or line, of light be regarded through a telescope, the aperture
of which is limited to two narrow parallel slits, interference bands are seen,
of which the theory is given in treatises on Optics. The width of the bands
is inversely proportional to the distance between the centres of the slits,
and the width of the field, upon which the bands are seen, is inversely
proportional to the width of the individual slits. If the latter element be
given, it will usually be advantageous to approximate the slits until only a
small number of bands are included. In this way not only are the bands
rendered larger, but illumination may be gained by the then admissible
widening of the original source.
Supposing, then, the proportions of the double slit to be given, we may
inquire as to the effect of an alteration in scale. A diminution in ratio m
will have the effect of magnifying m times the field and the bands (fixed in
number) visible upon it. Since the total aperture is diminished m times, it
might appear that the illumination would be diminished ?n,2 times, but the
admissible widening of the original source m times reduces the loss, so that
it stands at m times, instead of m2 times.
It remains, and this is more particularly the object of the present note,
to point out the effect of the telescope upon the angular magnitude and
illumination of the bands. If the magnifying power of the telescope exceed
the ratio of aperture of object-glass and pupil, its introduction is prejudicial.
And even if the above limit be not exceeded, the use of the telescope is
without advantage. The relation between the greatest brightness and the
apparent magnitude of the bands is the same whether a telescope be used
or not, the loss by reflections and absorptions being neglected. The function
of the telescope is merely to magnify the linear dimensions of the slit system.
1893] ON A SIMPLE INTERFERENCE ARRANGEMENT. 77
This magnification is sometimes important, especially when it is desirable
to operate separately upon the interfering pencils. But when the object is
merely to see the bands, the telescope may be abolished without loss. The
only difficulty is to construct the very diminutive slit system then required.
In the arrangement now exhibited the slits are very fine lines formed by
ruling with a knife upon a silver film supported upon glass. This double
slit is mounted at one end of a tube and at the other is placed a parallel
slit. It then suffices to look through the tube at a candle or gas flame in
order to see interference bands in a high degree of perfection.
It is suggested that this simple apparatus could be turned out very
cheaply, and that its introduction into the market would tend to popularise
acquaintance with interference phenomena.
208.
ON THE FLOW OF VISCOUS LIQUIDS, ESPECIALLY IN TWO
DIMENSIONS.
[Philosophical Magazine, xxxvi. pp. 354—372, 1893.]
THE problems in fluid motion of which solutions have hitherto been given
relate for the most part to two extreme conditions. In the first class the
viscosity is supposed to be sensible, but the motion is assumed to be so slow
that the terms involving the squares of the velocities may be omitted ; in
the second class the motion is not limited, but viscosity is supposed to be
absent or negligible.
Special problems of the first class have been solved by Stokes and other
mathematicians ; and general theorems of importance have been established
by v. Helmholtz* and by Kortewegf, relating to the laws of steady motion.
Thus in the steady motion (M0) of an incompressible fluid moving with velo-
cities given at the boundary, less energy is dissipated than in the case of
any other motion (M) consistent with the same conditions. And if the
motion M be in progress, the rate of dissipation will constantly decrease until
it reaches the minimum corresponding to M0. It follows that the motion M0
is always stable.
It is not necessary for our purpose to repeat the investigation of
Korteweg; but it may be well to call attention to the fact that problems
in viscous motion in which the squares of the velocities are neglected, fall
under the general method of Lagrange, at least when this is extended by
the introduction of a dissipation function J. In the present application there
is no potential energy to be considered, and everything depends upon the
expressions for the kinetic energy T and the dissipation function F. The
conditions to be satisfied may be expressed by ascribing given constant
* Collected Works, i. p. 223.
t Phil. Mag. xvi. p. 112, 1883.
t Theory of Sound, § 81. [See Vol. I. of the present collection, p. 176.]
1893] ON THE FLOW OF VISCOUS LIQUIDS. 79
values to some of the generalized velocities; but it is unnecessary to in-
troduce more than one into the argument, inasmuch as any others may be
eliminated beforehand by means of the given relations. Suppose, then, that
fa is given. The other coordinates fa, fa, ... may be so chosen that no
product of their velocities enters into the expressions for T7 and F, although
products with fa, such as fa fa, will enter. These coordinates are, in fact,
the normal coordinates of the system when fa is constrained to vanish.
Thus simplified F becomes
F=lb1fa*+...+$bsfa+...+brsfafa + (1)
and a similar expression applies to T with a written for b. Lagrange's
equation is now
asfa + argfa + bsfa + brgfa = 0,
fa being any one of the coordinates fa, fa, .... In this equation fa = Q,
and fa has a prescribed value; so that
agfa+bsfa = -brsfa (2)
is the equation giving fa. The solution of (2) is well known, and it appears
that fa settles gradually down to the value given by
bsfa = -brsfa, (3)
since as, bg are intrinsically positive. Further,
^= 2 {bsfafa + brs(fafa + fa fa)},
in which the summation extends to all values of s other than r. In this
fa = ®, so that
^ = 2fa{bsfa + brsfa}=-?.asfa*, (4)
by (2). The last expression is intrinsically negative, proving that until the
steady motion is reached F continually decreases. Korteweg's theorem is
thus shown to be of general application to systems devoid of potential
energy for which T and F can be expressed as quadratic functions of the
velocities with constant coefficients.
It may be mentioned in passing that a similar theorem holds for systems
devoid of kinetic energy, for which, however, F and V (the potential energy)
are sensible, and may be proved in the same way. If such a system be
subjected to given displacements, it settles down into the configuration of
minimum V; and during the progress of the motion V continually decreases.
The theorem of Korteweg places in a clear light the general question of
the slow motion of a viscous liquid under given boundary conditions, and the
only remaining difficulty lies in finding the analytical expressions suitable
for special problems. It is proposed to consider a few simple cases relating
to motion in two dimensions.
80 ON THE FLOW OF VISCOUS LIQUIDS, [208
Under the above restriction, as is well known, the motion may be ex-
pressed by means of Earnshaw's current function (•$•), which satisfies
V4>/r = 0, (5)
the same equation as governs the transverse displacement of an elastic plate,
when in equilibrium*. Of this analogy we shall avail ourselves in the sequel.
At a fixed wall ^ retains a constant value, and, further, in consequence of
the friction dty/dn, representing the tangential velocity, is evanescent. The
boundary conditions for a fixed wall in the fluid problem are therefore analo-
gous to those of a clamped edge in the statical problem.
The motion within a simply connected area is determined by (5) and by
the values of the component velocities over the boundary. If we suppose
that two such motions are possible, their difference constitutes a motion also
satisfying (5), and making i/r and d^rjdn zero over the boundary. Consider-
ations respecting energy in this or in the analogous problem of the elastic
plate are then sufficient to show that i/r must vanish throughout ; and an
analytical proof may readily be given by means of Green's theorem. For if
t/r and % are any two functions of x and y,
the integrations being taken round and over the area in question. If we
suppose that i/r and d\Jr/dn are zero over the boundary, the left-hand member
vanishes. If, further, % = V2-^, we have
f
of which the right-hand member vanishes by (5). Hence V2-\Jr vanishes all
over the area, and by a known theorem, as ^ vanishes on the contour, this
requires that i/r vanish throughout.
We will now investigate in detail the slow motion of viscous fluid within
a circular boundary. In virtue of (5) V2-^, which represents the vorticity,
satisfies Laplace's equation, and may therefore be expanded in positive and
negative integral powers of r, each term such as rn, or r~n, being accom-
panied by the factor cos (nd + a). But if, as we shall suppose, the vorticity
be finite at the centre of the circle, where r = 0, the negative powers are
excluded, and we have to consider only such terms as
[1902. If w be the displacement, parallel to z, at any point of a plane elastic plate in the
plane of xy, the differential equation of equilibrium is V% = 0, impressed forces being absent.]
1893] ESPECIALLY IN TWO DIMENSIONS. 81
The solution of this is readily obtained. If we assume
i/r = rm cos (nO + a), ........................... (9)
we find m = n + <2. To this may be added, as satisfying V2i|r = 0, a term
corresponding to m = n; so that the type of solution for nQ is
^ = Anrn+z cos (n0 + a) + Bnrn cos (n0 + (3) ............. (10)
By differentiation,
d-b
-^ = (n + 2) Anrn+> cos (n0 + a) + nBnrn~l cos (nO + 0) ....... (11)
The first problem to which we will apply these equations is that of motion
within the circle r = 1 under the condition that the tangential motion
vanishes at every part of the circumference. By (11) /8 = a, and
(n + 2)An + nBn = 0 ......................... (12)
The normal velocity at the boundary is represented by d-^/dO, and we might
be tempted, in our search after simplicity, to suppose that this is sensible in
the neighbourhood of one point only, for example 9 = 0. But in that case
the condition of incompressibility would require that the total flow of fluid
at the place in question should be zero. If the total quantity of fluid
entering the enclosure at 6 = 0 is to be finite, provision must be made for
its escape elsewhere. This might take the form of a sink at the centre of
the circle ; but it will come to much the same thing, and be more in harmony
with our equations, as already laid down, to suppose that the escape takes
place uniformly over the entire circumference. This state of things will be
represented analytically by ascribing to ty a sudden change of value from
— 1 to +1 at 0 = 0, with a gradual passage from the one value to the other
as 6 increases from 0 to 2?r, or, as it may be more conveniently expressed
for our present purpose, ty is to be regarded as an odd function of 6 such
that from 6 = 0 to 6 = ir its value is
0 = 1-01-* .................................. (13)
The symmetry with respect to 0 — 0 shows that we are concerned in (10)
only with the sines of multiples of 0, so that having regard to (12) we may
take as the form of -\Jr applicable in the present problem,
.................. (14)
in which n is any integer and Cn an arbitrary constant. It remains to
determine the coefficients G in accordance with (13). -When r=l,
and this must hold good for all values of 0 from 0 to TT. Multiplying by
sin m0 and integrating as usual, we find
~; ................................. (15)
iv.
82
ON THE FLOW OF VISCOUS LIQUIDS,
that
is the value of -$r ex
:2sinn0{(l-2/tt)rn-rn+2}
in series.
[208
.(16)
These series may be summed. In the first place, 2r"sinn# is the real
partof-t2(re»)n, or of
Thus
Again, Sri"1
so that
r sin 8
1 - 2r cos 6 + r3
..(17)
is the real part of - iSn-1 (rew)n, or of t log (1 - r
r sm ^
1 — r cos a
Thus, as the expression for i/r in finite terms, we have
r sin 6
(18)
.(19)
In (19) the separate parts admit of simple geometrical interpretation.
The second represents simply twice the angle PAG,
Fig. 1, which is known to constitute a solution of
V2i/r = 0. In the first term,
rsintf PM sinPAO
Fig. 1.
AP*
AP
which is also obviously a solution of V2\|r = 0. The
remaining part of (19) is not a solution of V2\/r = 0 ;
but it satisfies V4\Jr = 0, as being derived from a
solution of V2i/r = 0 by multiplication with ?*2.
On the foundation of (19) we may build up by simple integration the
general expression for i^, subject to the conditions that d-^r/dr vanishes
over the whole circumference, and that d^jdO has any prescribed values
consistent with the recurrence of -Jr.
Fig. 2.
A simple example is afforded by the case of a source at A and an equal
sink at B, where 6 = TT (Fig. 2). The fluid enters and
leaves the enclosure by two perforations situated at
opposite ends of a diameter, the walls being else-
where impenetrable. The solution may be found
independently, or from (19), by changing the sign ,
of cos 6, and adding the equations together. Thus
1893] ESPECIALLY IN TWO DIMENSIONS. 83
In this case the walls of the enclosure are of necessity stream-lines, the
value of -^ being + 1 from 0 to TT, and — 1 from 0 to — TT.
When 6 = %Tr, that is along OD (Fig. 2),
, ........................ (21)
From (21) we obtain by interpolation the following corresponding values: —
•00 -25 -50 -75 1-00
•00 -1330 -2800 -4698 I'OOOO
In the neighbourhood of A or B, Fig. 1, (20) assumes a special form. Thus
in the former case,
1 - 2r2 cos 26 + r4 - (1 - r2)2 + 4r2 sin2 (9 = 4 {^1M2 + PM "},
, 2r sin 0 . „ . ,k
tan-1 — - — = angle PA 0.
Thus if PAO be denoted by <f>, the value of -»Jr in the neighbourhood of
A is given by
TT . ^ = sin 2<£ + 2<£ ............................ (22')
That the functions of <j> which occur in (22') satisfy the fundamental
equation may be readily seen.
By calculation from (22') we get the following values for <f> expressed as
fractions of degrees: —
0 -25 -50 -75 1-00
0 11°'40 23°'83 39°-40 90C'00
This example is of interest, from its bearing upon the laws of flow at a
place where a channel is enlarged. In actual fluids there would be a ten-
dency to shoot directly across from A to B, the region about C being occupied
by an eddy, or backwater, such that the motion of the fluid near the wall is
reversed. Nothing of the kind is indicated by the present solution. In
(22) d-^rfdr represents the velocity across the line 0 = £?r, and we see that
there is no change of sign. In fact the velocity decreases, as r increases,
all the way from r = 0 to r = l. The formation of a backwater may thus
be connected with the terms involving the squares of the velocities, which
are neglected in the present solution. And we may infer that if the motion
were slow enough, or if the fluid were viscous enough, the backwater, usually
observed in practice, would disappear.
6—2
84 ON THE FLOW OF VISCOUS LIQUIDS, [208
Another particular case of some interest, included in the general solution
already indicated, would be obtained by supposing similar sources to be
situated at 0 = 0, 0 = 7T, and equal sinks at # = ^TT, O = ^TT.
We will now suppose that it is the radial velocity which vanishes at
every point of the circumference r = l, and that the tangential velocity also
vanishes except in the neighbourhood of 0 = 0. In this case, by the sym-
metry, \lr in (10) reduces to a series of cosines. And
- d^/dO = 2ra sin nB (Anrn+2 + Bnrn),
which is to vanish when r = 1 for all values of 6. Hence
An + Bn = 0; .............................. (23)
so that
^•=(1 -r*)2Bnrncosn6, ........................ (24)
d^r/dr = ^Bn cos n6 [nr^1 - (n + 2) rn+l] ............. (25)
When r = 1,
d^/dr = -2^Bncosn0, ....................... (26)
and is to be made to vanish for all values of 0 except in the neighbourhood
of 0 = 0. If we suppose that the integral of dty/dr with respect to d over
the whole region where dty/dr is sensible, is 2, we find
5. = -l/2w, Bn = -l/7r, ..................... (27)
the second equation applying to all values of n other than 0. Hence,
-7r.>/r = -£(l-r2) + (l-r2)2"rwcosw0, ............... (28)
or in finite terms,
-^ = -l(l-0 + (l-.%^co^ .......... (29)
The equation may also be written
-2-*=i
In (29),
1 - r cos 6 AM cosPAO
~ AP2~ AP '
which is a solution of V2i/r = 0. When multiplied by r2, or by (1 - r2), it
remains a solution of V4-^- = 0.
In (30) we may write x for r cos 0, and if the point under consideration
lie upon the axis, a? = r2. Hence on the axis,
-27T.A/T = (!+#)', ........................... (31)
-•n-ctyr/cfo = (!+#), ........................... (32)
equations which may be applied at all points except near #= 1. It appears
from (32) that the velocity transverse to the axis increases continuously from
x = — 1 to the neighbourhood of a? = + 1.
1893]
ESPECIALLY IN TWO DIMENSIONS.
85
The lines of flow are readily constructed from (30), which we may write
in the form
(33)
showing how P may be determined by the intersection of circles struck
from 0 and A. A few of the lines of flow are shown in Fig. 3. The external
circle AB corresponds to -^- = 0; AC, AO, AD correspond respectively to
-27TT/r = £, 1, 2. As appears from (31), the highest value of - 27r>/r is 4,
and gives a curve at A of infinitely small area.
Fig. 3.
In the neighbourhood of A (Fig. 1), (30) reduces to a simpler form. Thus
(33')
where <}> = PAO. The second term here satisfies the fundamental equation
as being derived by multiplication with AP2 from a solution, AP~2cos2<j>,
of V2 = 0.
86 ON THE FLOW OF VISCOUS LIQUIDS, [208
Equations (19), (30) give the means of expressing the stream-function
subject to the conditions that both •$• and d-^r/dr shall have values arbitrarily
given at all points of the circumference of the circle. It is not necessary
actually to write down the formulae : but it may be well to notice that the
same solution applies to the question of determining the transverse displace-
ment w of a thin circular plate when iv and dw/dr have arbitrarily prescribed
values on the boundary.
As a preliminary to further questions, it will be desirable to consider for
a moment the form of the general equations of viscous motion. In the usual
notation,
du du du du v 1 dp _0
-y- +u-r+v-r+w-r = X-- -f- + vV2u, ............ (34)
dt dx dy dz p dec
with two similar equations. Further, if q2 denote the resultant velocity, and
£, i), £ be the component rotations,
..... (35)
dy dz * dx
In steady motion du/dt = 0; and if the terms of the second order in velocity
(35) be omitted and there be no impressed forces except such as have a
potential, the equations reduce to the form already considered. A solution
thus obtained for small velocities will fail to satisfy the conditions when the
velocities are increased; but the equations lead readily to an instructive
expression for the forces X, Y, Z, which must be introduced in order that
the solution applicable without impressed forces to small velocities may still
continue to hold good. From (35) we see that the necessary forces are
(36)
with two similar equations. In this the term \d(f\dx need not be regarded,
as it tells only upon the pressure and does not influence the motion. We
may therefore write
=<2v%-2ur] ....... (37)
These equations show that
uX + vY+wZ=0, £X + -nY+ZZ = 0, ............... (38)
signifying that the force whose components are X, Y, Z, acts at every point
in a direction perpendicular both to the velocity and to the axis of rotation.
As regards its magnitude,
i(JT2 + Y* + Z2) = (u2 + v2 + O(£2 + 7?2 + £2) - (t*£ -vi)- w%)\ . . .(39)
If the motion take place in two dimensions, w = 0, f = 77 = 0, and
(40)
1893] ESPECIALLY IN TWO DIMENSIONS. 87
In the case of symmetry round an axis,
= 0,
and (39) reduces to
%(X* + Y* + Z*) = (u* + v* + w-)(Z* + rj* + f2) ............ (41)
These expressions for the forces necessary to the maintenance of a motion
similar to the infinitely small motion give us in simple cases an idea of the
direction in which the law is first departed from as the motion increases.
There are very few cases in which the problem of the rapid motion of a
viscous fluid has been dealt with. When the motion is in one dimension,
the troublesome terms do not present themselves, and the same solution
holds good mathematically for the steady motion at all velocities. When
the motion is so small that the laws appropriate to infinitely small motion
hold good as a first approximation, a correction may be calculated. This has
been effected by Whitehead*, and in an unpublished paper by Rowland, for
the problem, first investigated by Stokes, of a sphere moving with velocity V
through viscous liquid. For infinitely small motion the velocity of the
fluid in the neighbourhood of the sphere is of order V. It follows from
the solution referred to, or may be proved independently by considerations
of dimensions, that in the second approximation involving F2, the terms
are of the order V*a/v, a being the radius of the sphere, and v, equal to
//,/ p, the kinematic coefficient of viscosity. This method of approximation
is thus only legitimate when Va/v is small, a condition of a very restricting
character. In the case of water j/ = '01 c.G.s., and if Fa/i> = 'l, it is required
that Fa ='001.
Thus even if a were as small as one millimetre (-1), F should not exceed
'01 centimetre per second. With such diameters and velocities as often
occur in practice, Va,\v would be a large, instead of a small, quantity; and
a solution founded upon the type of infinitely slow motion is wholly in-
applicable.
We will now recur to the suppositions that the motion is steady, is in
two dimensions, and that its square may be neglected. Thus, writing as usual
u = d^jdy, v = - d^r / dx,
we get from (34)
Forces derivable from a potential do not disturb the equation V4-^ = 0.
In the analogy with a thin elastic plate, already referred to, a place where
dY/dx-dX/dy assumes a finite value in the fluid problem corresponds to
a place where transverse force acts upon the plate.
* Quart. Journ. of Math. Vol. xxm. p. 153 (1889).
88 ON THE FLOW OF VISCOUS LIQUIDS, [208
The simplest example of the finiteness of the second member of (42)
occurs when it is sensible at one point only. This is the case of forces
derivable from a potential 9, where 6 denotes the angle measured round
the point in question. It is to be observed that in the fluid problem the
forces themselves are not limited to the one point, but they have no "cir-
culation" except round that point. In the elastic problem, on the other
hand, the transverse force is limited to the one point.
The circumstance last mentioned renders the elastic problem the easier of
the two to deal with in thought and expression, and we will accordingly
avail ourselves of the analogy in the investigation which follows. It is pro-
posed to examine the infinitely slow motion of fluid within an enclosure,
which is maintained by forces having circulation at one point only, with the
view of determining whether a contrary flow, or backwater, is possible. In the
analogous elastic problem we have to consider a plate, subject at the boundary
to the conditions that w (the transverse displacement) and dw/dn shall every-
where vanish, and disturbed from its original plane condition by a force
acting transversely at a single point P. For distinctness we may suppose
that the plane is horizontal and that the force at P acts downwards, in
which direction the displacements are reckoned positive. At the point P
itself the principle of energy shows that the displacement is positive, and
it might appear probable that the displacement would be also positive at
all other points of the plate. A similar conclusion is readily proved to be
true in the case of a stretched membrane of any shape subjected to trans-
verse force at any point, and also in one dimension for a bar resisting flexure
by its stiffness. But a consideration of particular cases suffices to show that
the theorem cannot be generally true in the present case.
For suppose that the plate (Fig. 4) is almost divided into two independent
parts by a straight partition CD extending across, but perforated ^
by an aperture AB; and that the force is applied at a distance
from CD on the left. If the partition were complete, w and dw/dn
would be zero over the whole, and the displacement in the neigh-
bourhood on the left would be simple one-dimensional bending, with
w positive throughout. On the right w would vanish throughout.
In order to maintain this condition of things a certain couple acts
upon the plate in virtue of the supposed constraints along CD.
Along the perforated portion AB the couple required to produce
the one-dimensional bending fails. The actual deformation accord-
ingly differs from the one-dimensional bending by the deformation
that would be produced by a couple over AB acting upon the plate
as clamped along CA, BD, but otherwise free from force. This
deformation is evidently symmetrical with change of sign upon the
two sides of CD, w being positive on the left, negative on the right, and
1893]
ESPECIALLY IN TWO DIMENSIONS.
89
Fig. 5.
vanishing on AB itself. Thus upon the whole a downward force acting on
the left gives rise to an upward motion on the right, in opposition to the
general law proposed for examination.
In the application to the hydrodynamical problem we see that the fluid
moving on the left from D to B passes on in a
straight course to A, and thence along AC, and
that on the right an eddy, or backwater, is formed.
At distances from the aperture large in com-
parison with AB the supplementary motion is of
the character expressed in (33').
A similar argument may be applied to the .
case (Fig. 5) where fluid moves along a wall DC
into which a channel AF opens, and it leads to
the conclusion that the fluid on arrival at B will
refuse to follow the wall BF, but will rather shoot
across towards A.
These examples are of some interest as estab-
lishing that the formation of eddies observed in
practice is not wholly due to the influence of the terms involving the squares
of the velocities, but would persist in certain cases even though the motion
were made infinitely slow.
We will now investigate the motion in two dimensions of a viscous
Fig. 6.
incompressible fluid past a corrugated wall AB (Fig. 6), whose equation may
be taken to be
y = @coskx ............................... (43)
In this kft will be supposed to be a small quantity; in other words, the
depth of the corrugations small in comparison with their wave-length
(ITT Ik). Further we shall suppose, in the first instance, that the motion
is slow enough to allow the terms involving squares of the velocities to
be neglected; in which case the equation for the stream-function may be
written
0 .................................. (44)
At a distance from the wall we suppose the motion to take place in plane
strata, as defined by
+ = Lf .................................. (45)
90 ON THE FLOW OF VISCOUS LIQUIDS, [208
In the absence of corrugations this value of \|r might hold good throughout,
up to the wall at y = 0. The effect of the corrugations will be to introduce
terms periodic with respect to x ; but the influence of these will be confined
to the neighbourhood of the wall. For any term in ty, proportional to
cos mar, (44) gives
or ^ = A e~my + By e-™* + Cemy + Dyemv ;
but the condition last named requires that of the four arbitrary constants C
and D vanish. Also for our present purpose m is limited to be a multiple
of k.
The form of i/r applicable to our present purpose is accordingly
^ = A0 + B0y + Lf + cos kx (A.e^ + Biye~ky)
..., ...... (47)
in which the constants A0, B0, Alt ... are to be determined from the con-
ditions that i|r and dty/dy vanish when y = ftcosko;. It may be observed
that the problem is mathematically identical with that of an elastic plate
clamped at a sinuous edge, and deformed in such a manner that if there
were no sinuosity the bending would be one-dimensional.
The boundary conditions are
cos kx) e~k^ °°» kx
+ cos 2kx (A2 + B2ft cos kx) e-8tf «»te
+ ...... =0 ...................................................... (48)
and
B0 + 2Lj3 cos kx
+ cos kx (Bi -kAl-Blkft cos kx) e~k^ cos **
+ cos 2kx(Ba -2kA2- 2B2k0 cos kx) g-u*™**
+ ...... = 0; ...................................................... (49)
or, with use of (48),
kA0 + B0 + (BJcft + 2L/3) cos kx + Lk/32 cos"kx
+ (B.2 - kAt - B2k/3 cos kx) e~^ °°skx
+ ...... =0 ....................................................... (50)
The exponentials in (48), (50) could be expanded in Fourier's series by
means of Bessel's functions of an imaginary argument, and the complete
1893] ESPECIALLY IN TWO DIMENSIONS. 91
equations formed which express the evanescence of the various Fourier terms.
But the results are too complicated to be useful in the general case ; and, if we
regard kfi as small, it is hardly worth while to introduce the Bessel's functions
at all. The first approximation, in which /32 is neglected in (48), (50), gives
and the second approximation, in which ft2 is retained, gives
'
the coefficients with higher suffixes than 2 vanishing to this order of ap-
proximation. Thus
TJr/L = /32 (I - 2%) + 7/2 - 2/3?/e-^ cos kas
+ /3*($-ky) e~*y cos 2kx, ......... (53)
i ^ = - 2 kft* + 2y - 2(3 (1 - ky) er* cos kx
-2k/3-2(l-ky)e-aJcycos2kx, ........................ (54)
solutions applicable also to the problem of the elastic plate, if ty be under-
stood to mean the transverse displacement.
In the above investigation, so far as it applies to the hydrodynamical
question, L? has been supposed to be negligible. We will now retain the
square of L, but simplify the problem in another direction by neglecting the
square of /3, so that the first approximation is
(55)
The exact equation (derivable from (34)) for the motion of a viscous fluid
in two dimensions is
v dx v dy
From (55),
2L + 4>Lkfie-kv cos kx,
V2-\//-
(57)
Using this in (56) we have
«k2/Q T.2
i (58)
v
The solution of
<59>
92 ON THE FLOW OF VISCOUS LIQUIDS, [208
so that the required solution of (58), correct as far as the term involving
Z2, is
ty = Lf- 2L@ye-k* cos kx - —^ (y2 + ±ky3) e~^ sin kx. . . .(61)
It may be well to repeat that, though Z2 is retained. /32 is neglected in
(61); that is, the depth of the corrugations is supposed to be infinitely
small.
The part of the motion proportional to Z2 is, of course, independent of
the direction of the principal motion of the fluid, and is thus in a manner
applicable even when the principal motion is alternating. With regard to
the relative importance of the third and second terms in (61), we have to
consider the value of
and the conclusion will depend upon the value of y. If we suppose that
ky = \, the ratio is 2L : 3k'2 v, or, if we denote by V the undisturbed velocity
of the fluid when ky = l, V/Skv, or VX/Girv, X being the wave-length of
the corrugation. With ordinary liquids and moderate values of X, V would
have to be very small in order to permit the success of the method of
approximation.
The character of the motion proportional to L* is easily seen from the
value of v. We have
(62)
indicating a motion directed outwards from the wall over the places where
the sinuosities encroach upon the fluid, and an inward motion where the
sinuosities recede.
The application of the results towards the explanation of such phenomena
as ripple-mark and wave-formation requires a calculation of the forces operative
upon the boundary. We will confine ourselves to the first term in /3 and L,
in (55).
The normal stress, parallel to y, is given by
q~-*+*%~j*-%8i'' (63)
and the tangential stress, parallel to x, is
dv
(64)
1893] ESPECIALLY IN TWO DIMENSIONS. 93
From (34), (55) we find
p = — 4/iA^e-*^ sin kx,
or when y = 0,
£> = — 4<k/3 sin <r, simply.
Also, when y = 0,
2/A -?-•£• = - 4fc/3 sin kx ;
r dxdy
so that Q = 0 ..................................... (65)
In like manner, when y = 0,
sAw} ......................... (66)
So far as the first power of /3 the action upon the boundary is thus purely
tangential, and of magnitude given by (66). The periodic part has the same
sign as the constant part at the places where the boundary encroaches upon
the fluid.
This result finds immediate application to the question of wave-formation
under the action of wind, especially if we suppose that the waves move very
slowly, as they would do if gravity (and cohesion) were small. The main-
tenance or augmentation of the waves requires that the forces operative at
the surface be of suitable phase. Thus pressures acting upon the retreating
shoulders are favourable, as are also tangential forces acting forwards at the
crests of the waves, where the internal motion is itself in the forward direction.
Equation (65) shows that the pressures produce no effect, and that we have
only to consider the action of the tangential stress. We see from (66) that
when the waves move in the same direction as the wind, the effect of the
latter is to favour the development of the former. Whether the waves will
actually increase depends upon whether the supply of energy, proportional
to yS2, is greater or less than the loss from internal dissipation, itself propor-
tional to the same quantity. If the waves are moving against the wind, the
tendency is to a more rapid subsidence than would occur in a calm.
209.
THE SCIENTIFIC WORK OF TYNDALL.
[Proceedings of the Royal Institution, xiv. pp. 216—224, 1894.]
IT is fitting that the present season should not pass without a reference
on these evenings to the work of him whose tragic death a few months since
was felt as a personal grief and loss by every member of the Royal Institution.
With much diffidence I have undertaken the task to-night, wishing that it
had fallen to one better qualified by long and intimate acquaintance to do
justice to the theme. For Tyndall was a personality of exceeding interest.
He exercised an often magical charm upon those with whom he was closely
associated, but when his opposition was aroused he showed himself a keen
controversialist. My subject of to-night is but half the story.
Even the strictest devotion of the time at my disposal to a survey of the
scientific work of Tyndall will not allow of more than a very imperfect and
fragmentary treatment. During his thirty years of labour within these
walls he ranged over a vast field, and accumulated results of a very varied
character, important not only to the cultivators of the physical sciences, but
also to the biologist. All that I can hope to do is to bring back to your
recollection the more salient points of his work, and to illustrate them where
possible by experiments of his own devising.
In looking through the catalogue of scientific papers issued by the Royal
Society, one of the first entries under the name of Tyndall relates to a matter
comparatively simple, but still of some interest. It has been noticed that
when a jet of liquid is allowed to play into a receiving vessel, a good deal
of air is sometimes carried down with it, while at other times this does not
happen. The matter was examined experimentally by Tyndall, and he found
that it was closely connected with the peculiar transformation undergone by
a jet of liquid which had been previously investigated by Savart. A jet as
it issues from the nozzle is at first cylindrical, but after a time it becomes
what the physiologists call varicose; it swells in some places and contracts
in others. This effect becomes more exaggerated as the jet descends, until
1894] THE SCIENTIFIC WORK OF TYNDALL. 95
the swellings separate into distinct drops, which follow one another in single
file. Savart showed that under the influence of vibration the resolution into
drops takes place more rapidly, so that the place of resolution travels up
closer to the nozzle.
Tyndall's observation was that the carrying down of air required a jet
already resolved into drops when it strikes the liquid. I hope to be able to
show you the experiment by projection upon the screen. At the present
moment the jet is striking the water in the tank previously to resolution
into drops, and is therefore carrying down no air. If I operate on the nozzle
with a vibrating tuning-fork, the resolution occurs earlier, and the drops now
carry down with them a considerable quantity of air.
Among the earlier of Tyndall's papers are some relating to ice, a subject
which attracted him much, probably from his mountaineering experiences.
About the time of which I am speaking Faraday made interesting observa-
tions upon a peculiar behaviour of ice, afterwards called by the name of
regelation. He found that if two pieces of ice were brought into contact
they stuck or froze together. The pressure required to produce this effect
need not be more than exceedingly small. Tyndall found that if fragments
of ice are squeezed they pack themselves into a continuous mass. We have
here some small ice in a mould, where it can be subjected to a powerful
squeeze. The ice under this operation will be regelated, and a mass obtained
which may appear almost transparent, and as if it had never been fractured
at all. The flow of glaciers has been attributed to this action, the fractures
which the stresses produce being mended again by regelation. I should say,
perhaps, that the question of glacier motion presents difficulties not yet
wholly explained. There can be no doubt, however, that regelation plays an
important part.
Another question treated by Tyndall is the manner in which ice first
begins to melt under the action of a beam of light passing into it from an
electric lamp. Ice usually melts by conducted heat, which reaches first the
outside layers. But if we employ a beam from an electric lamp, the heat
will reach the ice not only outside but internally, and the melting will begin
at certain points in the interior. Here we have a slab of ice which we
project upon the screen. We see that the melting begins at certain points,
which develop a crystallised appearance resembling flowers. They are points
in the interior of the ice, not upon the surface. Tyndall found that when
the ice gives way at these internal points there is a formation of apparently
empty space. He carefully melted under water such a piece of ice, and
found that when the cavity was melted out there was no escape of air,
proving that the cavity was really vacuous.
Various speculations have been made as to the cause of this internal
melting at definite points, but here again I am not sure if the difficulty has
96 THE SCIENTIFIC WORK OF TYNDALL. [209
been altogether removed. One point of importance brought out by Tyndall
relates to the plane of the flowers. It is parallel to the direction in which
the ice originally froze, that is, parallel to the original surface of the water
from which it was formed.
I must not dwell further upon isolated questions, however interesting;
but will pass on at once to our main subject, which may be divided into
three distinct parts, relating namely to heat, especially dark radiation, sound,
and the behaviour of small particles, such as compose dust, whether of living
or dead matter.
The earlier publications of Tyndall on the subject of heat are for the
most part embodied in his work entitled Heat as a Mode of Motion. This
book has fascinated many readers. I could name more than one now distin-
guished physicist who drew his first scientific nutriment from it. At the
time of its appearance the law of the equivalence of heat and work was
quite recently established by the labours of Mayer and Joule, and had taken
firm hold of the minds of scientific men ; and a great part of Tyndall's book
may be considered to be inspired by and founded upon this first law of
thermodynamics. At the time of publication of Joule's labours, however,
there seems to have been a considerable body of hostile opinion, favourable
to the now obsolete notion that heat is a distinct entity called caloric.
Looking back, it is a little difficult to find out who were responsible for
this reception of the theory of caloric. Perhaps it was rather the popular
writers of the time than the first scientific authorities. A scientific worker,
especially if he devotes himself to original work, has not time to examine
for himself all questions, even those relating to his own department, but
must take something on trust from others whom he regards as authorities.
One might say that a knowledge of science, like a knowledge of law, consists
in knowing where to look for it. But even this kind of knowledge is not
always easy to obtain. It is only by experience that one can find out who
are most entitled to confidence. It is difficult now to understand the hesita-
tion that was shown in fully accepting, the doctrine that heat is a mode of
motion, for all the great authorities, especially in England, seem to have
favoured it. Not to mention Newton and Cavendish, we have Rumford
making almost conclusive experiments in its support, Davy accepting it,
and Young, who was hardly ever wrong, speaking of the antagonistic theory
almost with contempt. On the Continent perhaps, and especially among
the French school of chemists and physicists, caloric had more influential
support.
As has been said, a great part, though not the whole of Tyndall's work
was devoted to the new doctrine. Much relates to other matters, such as
radiant heat. Objection has been taken to this phrase, not altogether
without reason; for it may be said that when heat it is not radiant, and
1894] THE SCIENTIFIC WORK OF TYNDALL. 97
while radiant it is not heat. The term dark radiation, or dark radiance as
Newcomb calls it, is preferable, and was often used by Tyndall. If we
analyse, as Newton did, the components of light, we find that only certain
parts are visible. The invisible parts produce, however, as great, or greater,
effects in other ways than do the visible parts. The heating effect, for
example, is vastly greater in the invisible region than in the visible. One
of the experiments that Tyndall devised in order to illustrate this fact I
hope now to repeat. He found that it was possible by means of a solution
of iodine in bisulphide of carbon to isolate the invisible rays. This solution
is opaque to light ; even the sun could not be seen through it ; but it is very
fairly transparent to the invisible ultra-red radiation. By means of a concave
reflector I concentrate the rays from an arc lamp. In their path is inserted
the opaque solution, but in the focus of invisible radiation the heat developed
is sufficient to cause the inflammation of a piece of gun-cotton.
Tyndall varied this beautiful experiment in many ways. By raising to
incandescence a piece of platinum foil, he illustrated the transformation of
invisible into visible radiation.
The most important work, however, that we owe to Tyndall in connexion
with heat is the investigation of the absorption of invisible radiation by
gaseous bodies. Melloni had examined the behaviour of solid and liquid
bodies, but not of gases. He found that transparent bodies like glass might
be very opaque to invisible radiation. Thus, as we all know, a glass screen
will keep off the heat of a fire, while if we wish to protect ourselves from
the sun, the glass screen would be useless. On the other hand rock-salt
freely transmitted invisible radiation. But nothing had been done on the
subject of gaseous absorption, when Tyndall attacked this very difficult
problem. Some of his results are shown in the accompanying table. The
absorption of the ordinary non-condensable, or rather, not easily condensable
gases — for we must not talk of non-condensable gases now, least of all in
this place — the absorption of these gases is very small ; but when we pass
to the more compound gases, such as nitric oxide, we find the absorption
much greater — and in the case of olefiant gas we see that the absorbing
power is as much as 6000 times that of the ordinary gases.
Relative Absorption at
1 inch Pressure
Air 1
Oxygen 1
Nitrogen 1
Hydrogen .' 1
Carbonic acid 972
Nitric oxide 1590
Ammonia 5460
Olefiant gas G030
98 THE SCIENTIFIC WORK OF TYNDALL. [209
There is one substance as to which there has been a great diversity of
opinion — aqueous vapour. Tyndall found that aqueous vapour exercises a
strong power of absorption — strong relatively to that of the air in which it
is contained. This is of course a question of great importance, especially
in relation to meteorology. Tyndall's conclusions were vehemently contested
by many of the authorities of the time, among whom was Magnus, the
celebrated physicist of Berlin. With a view to this lecture I have gone
somewhat carefully into this question, and I have been greatly impressed
by the care and skill showed by Tyndall, even in his earlier experiments
upon this subject. He was at once sanguine and sceptical — a combination
necessary for success in any branch of science. The experimentalist who is
not sceptical will be led away on a false tack and accept conclusions which
he would find it necessary to reject were he to pursue the matter further; if
not sanguine, he will be discouraged altogether by the difficulties encountered
in his earlier efforts, and so arrive at no conclusion at all. One criticism,
however, may be made. Tyndall did not at first describe with sufficient
detail the method and the precautions which he used. There was a want of
that precise information necessary to allow another to follow in his steps.
Perhaps this may have been due to his literary instinct, which made him
averse from overloading his pages with technical experimental details.
The controversy above referred to I think we may now consider to be
closed. Nobody now doubts the absorbing power of aqueous vapour. In-
deed the question seems to have entered upon a new phase ; for in a recent
number of Wiedemann's Annalen, Paschen investigates the precise position
in the spectrum of the rays which are absorbed by aqueous vapour.
I cannot attempt to show you here any of the early experiments on the
absorption of vapours. But some years later Tyndall contrived an experi-
ment, which will allow of reproduction. It is founded on some observations
of Graham Bell, who discovered that various bodies become sonorous when
exposed to intermittent radiation.
The radiation is supplied from incandescent lime, and is focused by a
concave reflector. In the path of the rays is a revolving wheel provided
with projecting teeth. When a tooth intervenes, the radiation is stopped;
but in the interval between the teeth the radiation passes through, and
falls upon any object held at the focus. The object in this case is a small
glass bulb containing a few drops of ether, and communicating with the ear
by a rubber tube. Under the operation of the intermittent radiation, the
ether vapour expands and contracts ; in other words a vibration is established,
and a sound is heard by the observer. But if the vapour were absolutely
diathermanous, no sound would be heard.
I have repeated the experiment of Tyndall which allowed him to distin-
guish between the behaviour of ordinary air and dry air. If, dispensing with
1894] THE SCIENTIFIC WORK OF TYNDALL. 99
ether, we fill the bulb with air in the ordinary moist state, a sound is heard
with perfect distinctness, but if we drop in a little sulphuric acid, so as to
dry the air, the sound disappears.
According to the law of exchanges, absorption is connected with radiation ;
so that while hydrogen or oxygen do not radiate, from ammonia we might
expect to get considerable radiation. In the following experiment I aim at
showing that the radiation of hot coal gas exceeds the radiation of equally
hot air.
The face of the thermopile, protected by screens from the ball itself, is
exposed to the radiation from the heated air which rises from a hot copper
ball. The effect is manifested by the [spot of] light reflected from a galva-
nometer mirror. When we replace the air by a stream of coal gas, the
galvanometer indicates an augmentation of heat, so that we have before us a
demonstration that coal gas when heated does radiate more than equally hot
air, from which we conclude that it would exercise more absorption than air.
I come now to the second division of my subject, that relating to Sound.
Tyndall, as you know, wrote a book on Sound, founded on lectures delivered
in this place. Many interesting and original discoveries are there embodied.
One that I have been especially interested in myself, is on the subject of
sensitive flames. Professor Leconte in America made the first observations
at an amateur concert, but it was Tyndall who introduced the remarkable
high-pressure flame now before you. It issues from a pin-hole burner, and
the sensitiveness is entirely a question of the pressure at which the gas is
supplied. Tyndall describes the phenomenon by saying that the flame
under the influence of a high pressure is like something on the edge of a
precipice. If left alone, it will maintain itself ; but under the slightest touch
it will be pushed over. The gas at high pressure will, if undisturbed, burn
steadily and erect, but if a hiss is made in its neighbourhood it becomes at
once unsteady, and ducks down. A very high sound is necessary. Even a
whistle, as you see, does not act. Smooth pure sounds are practically without
effect unless of very high pitch.
I will illustrate the importance of the flame as a means of investigation
by an experiment in the diffraction of sound. I have here a source of sound,
but of pitch so high as to be inaudible. The waves impinge perpendicularly
upon a circular disc of plate glass. Behind the disc there is a sound shadow,
and you might expect that the shadow would be most complete at the centre.
But it is not so. When the burner occupies this [central] position the flame
flares ; but when by a slight motion of the disc the position of the flame is
made eccentric, the existence of the shadow is manifested by the recovery
of the flame. At the centre the intensity of sound is the same as if no
obstacle were interposed.
7—2
100 THE SCIENTIFIC WORK OF TYNDALL. [209
The optical analogue of the above experiment was made at the suggestion
of Poisson, who had deduced the result theoretically, but considered it so
unlikely that he regarded it as an objection to the undulatory theory of
light. Now, I need hardly say, it is regarded as a beautiful confirmation.
It is of importance to prove that the flame is not of the essence of the
matter, that there is no need to have a flame, or to ignite it at the burner.
Thus, it is quite possible to have a jet of gas so arranged that ignition does
not occur until the jet has lost its sensitiveness. The sensitive part is that
quite close to the nozzle, and the flame is only an indicator. But it is not
necessary to have any kind of flame at all. Tyndall made observations on
smoke-jets, showing that a jet of air can be made sensitive to sound. The
difficulty is to see it, and to operate successfully upon it ; because, as Tyndall
soon found, a smoke-jet is much more difficult to deal with than flames, and
is sensitive to much graver sounds. I doubt whether I am wise in trying to
exhibit smoke-jets to an audience, but I have a special means of projection
by which I ought at least to succeed in making them visible. It consists in
a device by which the main part of the light from the lamp is stopped at
the image of the arc, so that the only light which can reach the screen is
light which by diffusion has been diverted out of its course. Thus we shall
get an exhibition of a jet of smoke upon the screen, showing bright on a
dark ground. The jet issues near the mouth of a resonator of pitch 256.
When undisturbed it pursues a straight course, and remains cylindrical. But
if a fork of suitable pitch be sounded in the neighbourhood, the jet spreads
out into a sort of fan, or even bifurcates, as you see upon the screen. The
real motion of the jet cannot of course be ascertained by mere inspection.
It consists in a continuously increasing sinuosity, leading after a while to
complete disruption. If two forks slightly out of unison are sounded to-
gether, the jet expands and re-collects itself, synchronously with the audible
beats. I should say that my jet is a very coarse imitation of Tyndall's.
The nozzle that I am using is much too large. With a proper nozzle, and
in a perfectly undisturbed atmosphere — undisturbed not only by sounds, but
free from all draughts — the sensitiveness is wonderful. The slightest noise
is seen to act instantly and to bring the jet down to a fraction of its former
height.
Another important part of Tyndall's work on Sound was carried out as
adviser of the Trinity House. When in thick weather the ordinary lights
fail, an attempt is made to replace them with sound signals. These are
found to vary much in their action, sometimes being heard to a very great
distance, and at other times failing to make themselves audible even at a
moderate distance. Two explanations have been suggested,- depending upon
acoustic refraction and acoustic reflection.
Under the influence of variations of temperature refraction occurs in the
1894] THE SCIENTIFIC WORK OF TYNDALL. 101
atmosphere. For example, sound travels more quickly in warm than in cold
air. If, as often happens, it is colder above, the upper part of the sound
wave tends to lag behind, and the wave is liable to be tilted upwards and
so to be carried over the head of the would-be observer on the surface of the
ground. This explanation of acoustic refraction by variation of temperature
was given by Prof. Osborne Reynolds. As Sir G. Stokes showed, refraction
is also caused by wind. The difference between refraction by wind and by
temperature variations is that in one case everything turns upon the
direction in which the sound is going, while in the second case this con-
sideration is immaterial. The sound is heard by an observer down wind,
and not so well by an observer up wind. The explanation by refraction of
the frequent failure of sound signals was that adopted by Prof. Henry in
America, a distinguished worker upon this subject. Tyndall's investigations,
however, led him to favour another explanation. His view was that sound
was actually reflected by atmospheric irregularities. He observed, what
appears to be amply sufficient to establish his case, that prolonged signals
from fog sirens give rise to echoes audible after the signal has stopped.
This echo was heard from the air over the sea, and lasted in many cases
a long time, up to 15 seconds. There seems here no alternative but to
suppose that reflection must have occurred internally in the atmosphere.
In some cases the explanation of the occasional diminished penetration of
sound seems to be rather by refraction, and in others by reflection.
Tyndall proved that a single layer of hot air is sufficient to cause
reflection, and I propose to repeat his experiment. The source of sound,
a toy reed, is placed at one end of one metallic tube, and a sensitive flame
at one end of a second. The opposite ends of these tubes are placed near
each other, but in a position which does not permit the sound waves issuing
from the one to enter the other directly. Accordingly the flame shows no
response. If, however, a pane of glass be held suitably, the waves are
reflected back and the flame is excited. Tyndall's experiment consists in
the demonstration that a flat gas flame is competent to act the part of a
reflector. When I hold the gas flame in the proper position, the percipient
flame flares ; when the flat flame is removed or held at an unsuitable angle,
there is almost complete recovery.
It is true that in the atmosphere no such violent transitions of density
can occur as are met with in a flame ; but, on the other hand, the inter-
ruptions may be very numerous, as is indeed rendered probable by the
phenomena of stellar scintillation.
The third portion of my subject must be treated very briefly. The
guiding idea of much of Tyndall's work on atmospheric particles was the
application of an intense illumination to render them evident. Fine particles
102 THE SCIENTIFIC WORK OF TYNDALL. [209
of mastic, precipitated on admixture of varnish with a large quantity of
water, had already been examined by Briicke. Chemically precipitated
sulphur is convenient, and allows the influence of size to be watched as the
particles grow. But the most interesting observations of Tyndall relate to
precipitates in gases caused by the chemical action of the light itself. This
may be illustrated by causing the concentrated rays of the electric lamp to
pass through a flask containing vapour of peroxide of chlorine. Within a
few seconds dense clouds are produced.
When the particles are very small in comparison with the wave-length,
the laws governing the dispersion of the light are simple. Tyndall pursued
the investigation to the case where the particles have grown beyond the
limit above indicated, and found that the polarisation of the dispersed light
was affected in a peculiar and interesting manner.
Atmospheric dust, especially in London, is largely organic. If, following
Tyndall, we hold a spirit lamp under the track of the light from the electric
lamp, the dark spaces, resulting from the combustion of the dust, have all
the appearance of smoke.
In confined and undisturbed spaces the dust settles out. I have here a
large flask which has been closed for some days. If I hold it to the lamp,
the track of the light, plainly visible before entering and after leaving the
flask, is there interrupted. This, it will be evident, is a matter of consider-
able importance in connexion with organic germs.
The question of the spontaneous generation of life occupied Tyndall for
several years. He brought to bear upon it untiring perseverance and
refined experimental skill, and his results are those now generally accepted.
Guarding himself from too absolute statements as to other times and other
conditions, he concluded that under the circumstances of our experiments
life is always founded upon life. The putrefaction of vegetable and animal
infusions, even when initially sterilised, is to be attributed to the intrusion
of organic germs from the atmosphere.
The universal presence of such germs is often regarded as a hypothesis
difficult of acceptance. It may be illustrated by an experiment from the
inorganic world. I have here, and can project upon the screen, glass pots,
each containing a shallow layer of a supersaturated solution of sulphate of
soda. Protected by glass covers, they have stood without crystallising for
forty-eight hours. But if I remove the cover, a few seconds or minutes
will see the crystallisation commence. It has begun, and long needles are
invading the field of view. Here it must be understood that, with a few
exceptions, the crystalline germ required to start the action must be of the
same nature as the dissolved salt ; and the conclusion is that small crystals
of sulphate of soda are universally present in the atmosphere.
1894] THE SCIENTIFIC WORK OF TYNDALL. 103
I have now completed my task. With more or less success I have laid
before you the substance of some of Tyndall's contributions to knowledge.
What I could not hope to recall was the brilliant and often poetic exposition
by which his vivid imagination illumined the dry facts of science. Some
reminiscence of this may still be recovered by the reader of his treatises
and memoirs; but much survives only as an influence exerted upon the
minds of his contemporaries, and manifested in subsequent advances due
to his inspiration.
210.
ON AN ANOMALY ENCOUNTERED IN DETERMINATIONS
OF THE DENSITY OF NITROGEN GAS.
[Proceedings of the Royal Society, i>v. pp. 340—344, April, 1894.]
IN a former communication* I have described how nitrogen, prepared by
Lupton'sf method, proved to be lighter by about 1/1000 part than that
derived from air in the usual manner. In both cases a red-hot tube contain-
ing copper is employed, but with this difference. In the latter method the
atmospheric oxygen is removed by oxidation of the copper itself, while in
[Harcourt's] method it combines writh the hydrogen of ammonia, through
which the air is caused to pass on its way to the furnace, the copper remain-
ing unaltered. In order to exaggerate the effect, the air was subsequently
replaced by oxygen. Under these conditions the whole, instead of only about
one-seventh part of the nitrogen is derived from ammonia, and the dis-
crepancy was found to be exalted to about one-half per cent.
Upon the assumption that similar gas should be obtained by both
methods, we may explain the discrepancy by supposing either that the atmo-
spheric nitrogen was too heavy on account of imperfect removal of oxygen,
or that the ammonia nitrogen was too light on account of contamination with
gases lighter than pure nitrogen. Independently of the fact that the action
ofi the copper in the first case was pushed to great lengths, there are two
arguments which appeared to exclude the supposition that oxygen was still
present in the prepared gas. One of these depends upon the large quantity
of oxygen that would be required, in view of the small difference between the
weights of the two gases. As much as l/30th part of oxygen would be
necessary to raise the density by 1/200, or about one-sixth of all the oxygen
originally present. This seemed to be out of the question. But even if so
high a degree of imperfection in the action of the copper could be admitted,
* " On the Densities of the Principal Gases," Roy. Soc. Proc. Vol. LIII. p. 146, 1893. [Vol. iv.
p. 39. See also p. 1.]
t [1902. The use of ammonia to burn atmospheric oxygen is due to Mr Vernon Harcourt.]
1894] DENSITY OF NITROGEN GAS. 105
the large alteration caused by the substitution of oxygen for air in [Harcourt's]
process would remain unexplained. Moreover, as has been described in the
former paper, the introduction of hydrogen into the gas made no difference,
such hydrogen being removed by the hot oxide of copper subsequently
traversed. It is surely impossible that the supposed residual oxygen could
have survived such treatment.
Another argument may be founded upon more recent results, presently to
be given, from which it appears that almost exactly the same density is found
when the oxygen of air is removed by hot iron reduced with hydrogen,
instead of by copper, or in the cold by ferrous hydrate.
But the difficulties in the way of accepting the second alternative are
hardly less formidable. For the question at once arises, of what gas, lighter
than nitrogen, does the contamination consist ? In order that the reader may
the better judge, it may be well to specify more fully what were the arrange-
ments adopted. The gas, whether air or oxygen, after passing through potash
was charged with ammonia as it traversed a small wash-bottle, and thence
proceeded to the furnace. The first passage through the furnace was in a
tube packed with metallic copper, in the form of fine wire. Then followed a
wash-bottle of sulphuric acid by which the greater part of the excess of
ammonia would be arrested, and a second passage through the furnace in a
tube containing copper oxide. The gas then traversed a long length of pumice
charged with sulphuric acid, and a small wash-bottle containing Nessler
solution. On the other side of the regulating tap the arrangements were
always as formerly described, and included tubes of finely divided potash and
of phosphoric anhydride. The rate of passage was usually about half a litre
per hour.
Of the possible impurities, lighter than nitrogen, those most demanding
consideration are hydrogen, ammonia, and water vapour. The last may be
dismissed at once, and the absence of ammonia is almost equally certain.
The question of hydrogen appears the most important. But this gas, and
hydrocarbons, such as CH4, could they be present, should be burnt by the
copper oxide; and the experiments already referred to, in which hydrogen
was purposely introduced into atmospheric nitrogen, seem to prove conclu-
sively that the burning would really take place. Some further experiments
of the same kind will presently be given.
The gas from ammonia and oxygen was sometimes odourless, but at other
times smelt strongly of nitrous fumes, and, after mixture with moist air,
reddened litmus paper. On one occasion the oxidation of the nitrogen went
so far that the gas showed colour in the blow-off tube of the Toppler, although
the thickness of the layer was only about half an inch. But the presence
of nitric oxide is, of course, no explanation of the abnormal lightness. The
106 ON AN ANOMALY ENCOUNTERED IN DETERMINATIONS [210
conditions under which the oxidation takes place proved to be difficult of
control, and it was thought desirable to examine nitrogen derived by reduc-
tion from nitric and nitrous oxides.
The former source was the first experimented upon. The gas was evolved
from copper and diluted nitric acid in the usual way, and, after passing
through potash, was reduced by iron, copper not being sufficiently active, at
least without a very high temperature. The iron was prepared from black-
smith's scale. In order to get quit of carbon, it was first treated with a
current of oxygen at a red heat, and afterwards reduced by hydrogen, the
reduction being repeated after each employment. The greater part of the
work of reducing the gas was performed outside the furnace, in a tube heated
locally with a Bunsen flame. In the passage through the furnace in a tube
containing similar iron the work would be completed, if necessary. Next
followed washing with sulphuric acid (as required in the ammonia process), a
second passage through the furnace over copper oxide, and further washing
with sulphuric acid. In order to obtain an indication of any unreduced nitric
oxide, a wash-bottle containing ferrous sulphate was introduced, after which
followed the Nessler test and drying tubes, as already described. As thus
arranged, the apparatus could be employed without alteration, whether the
nitrogen to be collected was derived from air, from ammonia, from nitric
oxide, from nitrous oxide, or from ammonium nitrite.
The numbers which follow are the weights of the gas contained by the
globe at zero, at the pressure defined by the manometer when the tempera-
ture is 15°. They are corrected for the errors in the weights, but not for the
shrinkage of the globe when exhausted, and thus correspond to the number
2*31026, as formerly given for nitrogen.
Nitrogen from NO by Hot Iron.
November 29, 1893 2-30143 \
December 2,1893 2-29890 __
December 5,1893 2'29816 Mean' ™™S
December 6, 1893 2'30182 J
Nitrogen from N20 by Hot Iron*.
December 26, 1893 2'29869 ) ..
December 28, 1893 2*29940 } Mean' 2'2"°4
Nitrogen from Ammonium Nitrite passed over Hot Iron.
January 9, 1894 2*29849 }
January 13, 1894 2'29889 } Mean> 2 29S
* The N20 was prepared from zinc and very dilute nitric acid.
1894] OF THE DENSITY OF NITROGEN GAS. 107
With these are to be compared the weights of nitrogen derived from the
atmosphere.
Nitrogen from Air by Hot Iron.
December 12, 1893 2'31017 \
December 14, 1893 2'30986 (H) I
December 19, 1893 2*31010 (H) f an' 23
December 22, 1893 2-31001 J
Nitrogen from Air by Ferrous Hydrate.
January 27, 1894 2'31024 j
January 30, 1894 2-31010 [ Mean, 2*31020
February 1, 1894 2'31028 )
In the last case a large volume of air was confined for several hours in a
glass reservoir with a mixture of slaked lime and ferrous sulphate. The gas
was displaced by deoxygenated water, and further purified by passage through
a tube packed with a similar mixture. The hot tubes were not used.
If we bring together the means for atmospheric nitrogen obtained by
various methods, the agreement is seen to be good, and may be regarded as
inconsistent with the supposition of residual oxygen in quantity sufficient to
influence the weights.
Atmospheric Nitrogen.
By hot copper, 1892 2'31026
By hot iron, 1893 2'31003
By ferrous hydrate, 1894 2'31020
Two of the results relating to hot iron, those of December 14 and Decem-
ber 19, were obtained from nitrogen, into which hydrogen had been purposely
introduced. An electrolytic generator was inserted between the two tubes
containing hot iron, as formerly described. The generator worked under its
own electromotiVe force, and the current was measured by a tangent galvano-
meter. Thus, on December 19, the deflection throughout the time of filling
was 3°, representing about 1/15 ampere. In two hours and a half the hydro-
gen introduced into the gas would be about 70 c.c., sufficient, if retained, to
reduce the weight by about 4 per cent. The fact that there was no sensible
reduction proves that the hydrogen was effectively removed by the copper
oxide.
The nitrogen, obtained altogether in four ways from chemical compounds,
is materially lighter than the above, the difference amounting to about
11 mg., or about 1/200 part of the whole. It is also to be observed that the
agreement of individual results is less close in the case of chemical nitrogen
than of atmospheric nitrogen.
108 DENSITY OF NITROGEN GAS. [210
I have made some experiments to try whether the densities were influ-
enced by exposing the gas to the silent electric discharge. A Siemens tube,
as used for generating ozone, was inserted in the path of the gas after desic-
cation with phosphoric anhydride. The following were the results : —
Nitrogen from Air by Hot Iron, Electrified.
January 1, 1894 2'31163 ) _
„ > Mean, 2'310o9
January 4, 1894 2*30956 j
Nitrogen from N,O by Hot Iron, Electrified.
January 2, 1894 2'30074 |
January 5, 1894 2'30054 } Mean' 2'3°°64
The somewhat anomalous result of January 1 is partly explained by the
failure to obtain a subsequent weighing of the globe empty, and there is no
indication that any effect was produced by the electrification.
One more observation I will bring forward in conclusion. Nitrogen pre-
pared from oxygen and ammonia, and about one-half per cent, lighter than
ordinary atmospheric nitrogen, was stored in the globe for eight months.
The globe was then connected to the apparatus, and the pressure was re-
adjusted in the usual manner to the standard conditions. On re-weighing
no change was observed, so that the abnormally light nitrogen did not become
dense by keeping.
[1902. For the explanation of the discrepancy here set forth, as due to a
previously unrecognised constituent of the atmosphere, see the memoir by
Rayleigh and Ramsay, Art. 214 below.]
211.
ON THE MINIMUM CURRENT AUDIBLE IN THE TELEPHONE.
[Philosophical Magazine, xxxvm. pp. 285—295, 1894*.]
THE estimates which have been put forward of the minimum current
perceptible in the Bell telephone vary largely. Mr Preece gives 6 x 10~1S
ampere f ; Prof. Tait, for a current reversed 500 times per second, 2 x 10~12
ampere*. De la Rue gives 1 x 10~8 ampere, and the same figure is recorded
by Brough§ as applicable to the strongest current with which the instrument
is worked. Various methods, more or less worthy of confidence, have been
employed, but the only experimenter who has described his procedure with
detail sufficient to allow of criticism is Prof. Ferraris ||, whose results may be
thus expressed : —
Fluency
Do3 ............... 264 23x10-
Fa3 ............... 352 17x10-
La3 ............... 440 10x10-
Do4 ............... 528 7x10-
Re4 ............... 594 5x10-
The currents were from a make-and-break apparatus, and in each case are
reckoned as if only the first periodic term of the Fourier series, representative
of the actual current, were effective. On this account the quantities in the
third column should probably be increased, for the presence of overtones
could hardly fail to favour audibility.
Although a considerable margin must be allowed for varying pitch, vary-
ing acuteness of audition, and varying construction of the instruments, it is
scarcely possible to suppose that all the results above mentioned can be
* Bead at the Oxford Meeting of the British Association.
t Brit. Assoc. Report, Manchester, 1887, p. 611.
J Edin. Proc. Vol. ix. p. 551 (1878). Prof. Tait speaks of a billion B.A. units, and, as he
kindly informs me, a billion here means 1012.
§ Proceedings of the Asiatic Society of Bengal, 1877, p. 255.
|| Atti delta R. Accad. d. Sci. di Torino, Vol. xm. p. 1024 (1877).
110 ON THE MINIMUM CURRENT AUDIBLE IN THE TELEPHONE. [211
correct, even in the roughest sense. The question is of considerable interest
in connexion with the theory of the telephone. For it appears that a, priori
calculations of the possible efficiency of the instrument are difficult to reconcile
with numbers such as those of Tait and of Preece, at least without attributing
to the ear a degree of sensitiveness to aerial vibration far surpassing even the
marvellous estimates that have hitherto been given*.
Under these circumstances it appeared to be desirable to undertake fresh
observations, in which regard should be paid to various sources of error that
may have escaped attention in the earlier days of telephony. The importance
of denning the resistance of the instruments and of employing pure tones of
various pitch need not be insisted upon.
As regards resistance, a low-resistance telephone, although suitable in
certain cases, must not be expected to show the same sensitiveness to current
as an instrument of higher resistance. If we suppose that the total space
available for the windings is given, and that the proportion of it occupied by
the copper is also given, a simple relation obtains between the resistance and
the minimum current. For if 7 be the current, n be the number of convolu-
tions, and r the resistance, we have, as in the theory of galvanometers,
ny = const., n~-r = const., so that y^r — const., or the minimum current is
inversely as the square root of the resistance.
The telephones employed in the experiments about to be narrated were
two, of which one (Tj) is a very efficient instrument of 70-ohms resistance.
The other (T2), of less finished workmanship, was rewound in the laboratory
with comparatively thick wire. The interior diameter of the windings is
9 mm., and the exterior diameter is 26 mm. The width of the groove, or the
axial dimension of the coil, is 8 mm., the number of windings is 160, and the
resistance is '8 ohm. Since the dimensions of the coils are about the same
in the two cases, we should expect, according to the above law, that about
10 times as much current would be required in T2 as in Tt. Both instru-
ments are of the Bell (unipolar) type, and comparison with other specimens
shows that there is nothing exceptional in their sensibility.
In view of the immense discrepancies above recorded, it is evident that
what is required is not so much accuracy of measurement as assured sound-
ness in method. It appeared to me that electromotive forces of the necessary
harmonic type would be best secured by the employment of a revolving
magnet in the proximity of an inductor-coil of known construction. The
electromotive force thus generated operates in a circuit of known resistance ;
and, if the self-induction can be neglected, the calculation of the current
presents no difficulty. The sound as heard in the telephone may be reduced
* Proc. Roy. Soc. Vol. xxvi. p. 248 (1877). [Vol. i. p. 328 ; sec also Art. 213 below.] Also
Wien, Wied. Ann. Vol. xxxvi. p. 834 (1889).
1894] ON THE MINIMUM CURRENT AUDIBLE IN THE TELEPHONE. Ill
to the required point either by varying the distance (B) between the magnet
and the inductor, or by increasing the resistance (R) of the circuit. In fact
both these quantities may be varied ; and the agreement of results obtained
with widely different values of R constitutes an effective test of the legitimacy
of neglecting self-induction. When R is too much reduced, the time-constant
of the circuit becomes comparable with the period of vibration, and the current
is no longer increased in proportion to the reduction of R. This complication
is most likely to occur when the pitch is high.
In order to keep as clear as possible of the complication due to self-induc-
tion, I employed in the earlier experiments a resistance-coil of 100,000 ohms,
constructed as usual of wire doubled upon itself. But it soon appeared that
in avoiding Scylla I had fallen upon Charybdis. The first suspicion of some-
thing wrong arose from the observation that the sound was nearly as loud
when the 100,000 ohms was included as when a 10,000-ohm coil was substi-
tuted for it. The first explanation that suggested itself was that the sound
was being conveyed mechanically instead of electrically, as is indeed quite
possible under certain conditions of experiment. But a careful observation
of the effect of breaking the continuity of the leads, one at a time, proved
that the propagation was really electrical. Subsequent inquiry showed that
the anomaly was due to a condenser, or leyden, like action of the doubled
wire of the 100,000-ohm coil. When the junction at the middle was un-
soldered, so as to interrupt the metallic continuity, the sounds heard in the
telephone were nearly as loud as before. In this condition the resistance
should have been enormous, and was in fact about 12 megohms* as indicated
by a galvanometer. It was evident that the coil was acting principally as a
leyden rather than as a resistance, and that any calculation founded upon
results obtained with it would be entirely fallacious.
It is easy to form an estimate of the point at which the complication due
to capacity would begin to manifest itself. Consider the case of a simple
resistance R in parallel with a leyden of capacity C, and let the currents in
the two branches be x and y respectively. If V be the difference of potential
at the common terminals, proportional to eipt, we have
as = V/R, y = CdV/dt = ipVC;
so that
x + y = 1 + JpRG
V R
The amplitude of the total current is increased by the leyden in the ratio
V(l + p^R^C1) : 1 ; and the action of the leyden becomes important when
pRG= 1. ' With a frequency of 640, p = 4020 ; so that, if R = 1014 C.G.S., the
critical value of C is -fa x 10~15 C.G.S., or about ^ of a microfarad.
* Doubtless the insulation between the wires should have been much higher.
112 ON THE MINIMUM CURRENT AUDIBLE IN THE TELEPHONE. [211
It will be seen that even if the capacity remained unaltered, a reduction
of resistance in the ratio say of 10 to 1 would greatly dimmish the complica-
tion due to condenser-like action ; but perhaps the best evidence that the
results obtained are not prejudiced in this manner is afforded by the experi-
ments in which the principal resistance was a column of plumbago.
The revolving magnet was of clock-spring, about 2| cm. long, and so bent
as to be driven directly, windmill fashion, from an organ bellows. It was
mounted transversely upon a portion of a sewing-needle, the terminals of
which were carried in slight indentations at the ends of a U-shaped piece of
brass. As fitted to the wind-trunk, the axis of rotation was horizontal.
The inductor-coil, with its plane horizontal, was situated so that its centre
was vertically below that of the magnet at distance B. Thus, if A be the
mean radius of the coil, n the number of convolutions, the galvanometer-
constant G of the coil at the place occupied by the magnet is given by
a)
where C"- = A--\- B2; and if m be the magnetic moment of the magnet, and
the angle of rotation, the mutual potential M may be represented by*
(2)
If the frequency of revolution be p/27r, <f> = pt; and then
dM/dt=Gmpcospt (3)
The expression (3) represents the electromotive force operative in the circuit.
If the inductance can be neglected, the corresponding current is obtained on
division of (3) by R, the total resistance of the circuit.
The moment in is deduced by observation of the deflection of a magneto-
meter-needle from the position which it assumes under the operation of the
earth's horizontal force H. If the magnet be situated to the east at distance
r, and be itself directed east and west, the angular deflection 0 from equili-
brium is given by
a 2»i/rs
tan u = — jj .
The relation between the angle 0 and the double deflection d in scale-
divisions, obtained on revel-sal of TO, is approximately 0 = d/4>D, where D is
the distance between mirror and scale ; so that we may take
_Hr*d
* Maxwell, Electricity and Magnetism, Vol. n. § 700.
1894] ON THE MINIMUM CURRENT AUDIBLE IN THE TELEPHONE. 113
The amplitude of the oscillatory current, generated under these conditions, is
accordingly
If C.G.S. units are employed, H='l8. A must of course be measured in
centimetres ; but any units that are convenient may be used for r and C, and
for d and D. The current will then be given in terms of the C.G.S. unit,
which is equal to 10 amperes.
The inductor-coil used in most of the experiments is wound upon an
ebonite ring, and is the one that was employed as the " suspended coil " in
the determination of the electro-chemical equivalent of silver*. The number
of convolutions (n) is 242. The axial dimension of the section is 1'4 cm.
and the radial dimension is '97 cm. The mean radius A is 10'25 cm., and
the resistance is about 10£ ohms.
In making the observations the current from the inductor-coil was led to
a distant part of the house by leads of doubled wire, and was there connected
to the telephone and resistances. Among the latter was a plumbago resist-
ance on Prof. F. J. Smith's planf of about 84,000 ohms; but in .most of the
experiments a resistance-box going up to 10,000 ohms was employed, with
the advantage of allowing the adjustment of sound to be made by the observer
at the telephone. The attempt to hit off the least possible sound was found
to be very fatiguing and unsatisfactory ; and in all the results here recorded
the sounds were adjusted so as to be easily audible after attention for a few
seconds. Experiment showed that the resistances could then be doubled
without losing the sound, although perhaps it would not be caught at once
by an unprepared ear. But it must not be supposed that the observation
admits of precision, at least without greater precautions than could well be
taken. Much depends upon the state of the ear as regards fatigue, and upon
freedom from external disturbance.
The pitch was determined before and after an observation by removing
the added resistance and comparing the loud sound then heard with a harmo-
nium. The octave thus estimated might be a little uncertain. It was verified
by listening to the beats of the sound from the telephone and from a nearly
unisonant tuning-fork, both sounds being nearly pure tones.
When the magnet was driven at full speed the frequency was found to be
307, and at this pitch a series of observations was made with various values
of G and of R. Thus when 5 = 7'75 inches, or (7= 8'7 inches, the resistance
from the box required to produce the standard sound in telephone T^ was
* Phil. Trans. Part n. 1884, p. 421. [Vol. n. p. 290.]
t Phil. Mag. Vol. xxxv. p. 210 (1893).
114 ON THE MINIMUM CURRENT AUDIBLE IN THE TELEPHONE. [211
8000 ohms, so that ,R = 8100xl09. The quantities required for the calcu-
lation of (5) are as follows : —
A = 10-25,
(7=8-7,
= 2?rx307,
= 8-25,
= 81x10",
= •18,
Z = 140,
> = 1370,
r and C being reckoned in inches, d and Z) in scale-divisions of about -^ inch.
From these data the current required to produce the standard sound is found
to be 7'4 x 10~8 C.G.S., or 7*4 x 10~7 amperes, for telephone T^.
The results obtained by the method of the revolving magnet are collected
into the accompanying table. The " wooden coil " is of smaller dimensions
than the " ebonite coil," the mean radius being only 3'5 cm. The number
of convolutions is 370.
Telephone
Frequency = 307. Ebonite coil.
R in ohms Current in amperes Sound
84100 Plumbago
8100 Box
4100 Box
3-8xlO-7
7-4xlO-7
5-2 x 10~7
Below standard
Standard
500 Box
l'2x 10~6
200 Box
1-OxlO-5
Frequency = 307. Wooden coil.
84100 Plumbago
10100 Box
1600 Box
350 Box
3-6xlO~r
3-7x10-7
5'4xlO-y
1-lxlO"5
Standard
Frequency = 192. Ebonite coil.
7\ | 3100 Box | 2-oxlO-6 | Standard
The method of the revolving magnet seemed to be quite satisfactory so
far as it went, but it was desirable to extend the determinations to frequencies
higher than could well be reached in this manner. For this purpose recourse
was had to magnetized tuning-forks, vibrating with known amplitudes. If,
for the moment, we suppose the magnetic poles to be concentrated at the
extremities of the prongs, a vibrating-fork may be regarded as a simple
magnet, fixed in position and direction, but of moment proportional to the
instantaneous distance between the poles. Thus, if the magnetic axis pass
perpendicularly through the centre of the mean plane of the inductor-coil,
the situation is very similar to that obtaining in the case of the revolving
magnet. The angle $ in (2) is no longer variable, but such that sin <f> = 1
throughout. On the other hand m varies harmonically. If I be the mean
distance between the poles, 2# the extreme arc from rest to rest traversed by
1894] ON THE MINIMUM CURRENT AUDIBLE IN THE TELEPHONE. 115
each pole during the vibration, w0 the mean magnetic moment,
M/WO = 1 + 2/3/1 . sinpt,
and
dMJdt=Gm0p.'2fi/l.cospt ,, (6)
The formula corresponding to (5) is thus derived from it by simple introduc-
tion of the factor 2/3/1.
The forks were excited by bowing, and the observation of amplitude was
effected by comparison with a finely divided scale under a magnifying-glass.
It was convenient to observe the extreme end of a prong where the motion is
greatest, but the double amplitude thus measured must be distinguished from
2/3. In order to allow for the distance between the resultant poles and the
extremities of the prongs, the measured amplitude was reduced in the ratio
of 2 to 3. The observation of the magnetic moment at the magnetometer is
not embarrassed by the diffusion of the free polarity.
In order to explain the determination more completely, I will give full
details of an observation with a fork c' of frequency 256. The distance I
between the middles of the prongs was '875 inch, and the double amplitude
of the vibration at the end of one of the prongs was '09 inch. Thus 2/3 is
reckoned as "06 inch. The inductor-coil was the ebonite coil already described,
and the sound was judged to be of the standard distinctness when, for
example, 5 = 15 inches, or (7=15*5 inches, and the added resistance was
1000 ohms, so that R = 1100 x 10°. The quantities required for the compu-
tation of (5) as extended are
n = 242, p = 2-7T x 256, H = 18,
4=10-25, r = 15, d = 410,
(7=15-5, # = 11x10", D = 1370,
2/3 = -06, £ = -875;
and they give for the current corresponding to the standard sound 9'8 x 10~8
C.G.S., or 9'8 x 10~7 amperes.
A summary of the results obtained with forks of pitch c, c', e, g', c", e", g"
is annexed. As the pitch rose, the difficulties of observation increased, both
Telephone R in ohms Current in amperes
C = 128.
71! | 1100 | 2-8 xlO~6
c = 256.
8100 Box
1100
500 ...
6-8x10-7
9-8x10-7
1-1x10-
116
ON THE MINIMUM CURRENT AUDIBLE IN THE TELEPHONE.
[211
Telephone
2*
R in ohms
e = 320.
84000 Plumbago
6100 Box
1600 ...
Current in amperes
3-8xlO~7
2-6x10-7
3-1x10-'
g' = 384.
T^
84000 Plumbago
1-4x10-'
r,
9500 Box
l-6xlO~7
7\
2100
1-4x10-'
7*!
900
1-7 xlO-7
T0
600
1-9 xlO~6
J 2
T7* ...
300 ...
2-2 xlO~6
c"=512.
84000 Plumbago
8-9x10-8
9000 Box
4-8x10-8
3600
5-2xlO-8
700
8-2x10-8
11?
5-2xlO-c?
100 Box
l-9xlO-«
300
l-4x!0-fi
500
2-5x10-°
900 ...
2-4xlO-6
e" = 640.
84000 Plumbago
5100 Box
1100 ...
" = 768.
84000 Plumbago
7100 Box
2100 ...
3-8xlO-8
3-8xlO-8
5-5x10-8
1-1x10-'
•9 x!0~7
1-1 xlO"7
on account of the less duration of the sound and of the smaller amplitudes
available for measurement. In one observation with telephone T.2 at pitch c",
the resistance, estimated at 11 ohms, was that of the coil, telephone, and
leads only. No trustworthy result was to be expected under such conditions,
but the number is included in order to show how small was the influence of
self-induction, even where it had every opportunity of manifesting itself. If
we bring together the numbers* derived with the revolving magnet and with
the forks, we obtain in the case of Tl : —
* The observations recorded were made with my own ears. Mr Gordon obtained very similar
numbers when he took my place.
1894] ON THE MINIMUM CURRENT AUDIBLE IN THE TELEPHONE. 117
Pitch
128
Source
Fork
Current in 10~8 amperes
2800
192
256
307
320
Revolving magnet
Fork
Revolving magnet
Fork
250
83
49
32
384
15
512
7
640
4-4
768...
10
It would appear that the maximum sensitiveness to current occurs in the
region of frequency 640 ; but observations at still higher frequencies would
be needed to establish this conclusion beyond doubt. Attention must be paid
to the fact that the sounds were not the least that could be heard, and that
before a comparison is made with the numbers given by other experimenters
there should be a division by 2, if not by 3. But this consideration does not
fully explain the difference between the above table and that of Ferraris
already quoted, from which it appears that in his experiments a current of
5 x 10~9 amperes was audible.
It is interesting to note that the sensitiveness of the telephone to periodic
currents is of the same order as that of the galvanometer of equal resistance
to steady currents*, viz. that the currents (at pitch 512) just audible in the
telephone would, on commutation, be just easily visible by a deflection in the
latter instrument. But there is probably more room for further refinements
in the galvanometer than in the telephone.
If we compare the performances of the two telephones T^ and T2, we find
ratios of sensitiveness to current ranging from 13 to 30 ; so that T2 shows
itself inferior in a degree beyond what may be accounted for by the resist-
ances. It is singular that an experiment of another kind led to the opposite
conclusion. The circuit of a Daniell cell A was permanently closed through
resistance-coils of 5 ohms and of 1000 ohms. The two telephones in series
with one another and with a resistance-box C were placed in a derived circuit
where was also a scraping contact-apparatus B, as indicated in the figure.
The adjustment was made by varying the resistance in C until the sound was
just easily audible in the telephone under trial. Experiments conducted
* See, for example, Ayrton, Mather, and Sumpnef, Phil. Mag. Vol. xxx. p. 90, 1890, " On
Galvanometers."
118 ON THE MINIMUM CURRENT AUDIBLE IN THE TELEPHONE. [211
upon this plan showed that T, was only about five times as sensitive to cur-
rent as T2. It was noticed, however, that the sounds, though as equal as
could be estimated, were not of the same quality, and in this probably lies
the explanation of the discrepancy between the two methods of experiment-
ing. In the latter the original sound is composite, and the telephone selects
the most favourable elements — that is, those nearly in agreement with the
natural pitch of its own plate. In this way the loudness of the selected sound
becomes a question of the freedom of vibration of the plate, an element which
is almost without influence when the sound is of pitch far removed from that
of the proper tone of the telephone. There was independent reason for the
suspicion that Tl had not so well defined a proper pitch as was met with in
the case of some other telephones.
P. 8. — Measurements with the electro-dynamometer have been made by
Cross and Page* of the currents used in practical telephony. The experi-
ments were varied by the employment of several transmitters, and various
vowel sounds were investigated. The currents found were of the order
2 x 10~4 amperes.
* Electrical Review, Nov. 14, 1885. I owe this reference to Mr Swinburne.
212.
AN ATTEMPT AT A QUANTITATIVE THEORY OF THE
TELEPHONE.
[Philosophical Magazine, xxxvni. pp. 295 — 301, 1894*.]
THE theory of the telephone cannot be said to be understood, in any but
the most general manner, until it is possible to estimate from the data of
construction what its sensitiveness should be, at least so far as to connect the
magnitude of the vibratory current with the resulting condensations and
rarefactions in the external ear-passage. Unfortunately such an estimate is
a matter of extreme difficulty, partly on account of imperfection in our know-
ledge of the . magnetic properties of iron, and partly from mathematical diffi-
culties arising from the particular forms employed in actual construction ; and
indeed the problem does not appear to have been attacked hitherto. In view,
however, of the doubts that have been expressed as to theory, and of the
highly discrepant estimates of actual sensitiveness which have been put
forward, it appears desirable to make the attempt. It will be understood
that at present the question is as to the order of magnitude only, and that
the result will not be without value should it prove to be 10 or even 100
times in error.
One of the elements required to be known, the number (n) of convolutions,
cannot be directly observed in the case of a finished instrument ; but it may
be inferred with sufficient accuracy for the present purpose from the dimen-
sions and the resistance of the coil. Denote the axial dimension by £, the
inner and outer radii by ^ and tj.2, the section of the wire by a- and its total
length by I, so that l<r is the total volume of copper. The area of section of
the coil by an axial plane is f (i)2 — ^a), and of this the area nor is occupied by
* Read at the Oxford Meeting of the British Association.
120 ON A QUANTITATIVE THEORY OF THE TELEPHONE [212
copper. If we suppose the latter to be half the former, we shall not be far
from the mark. Thus
H«--if(lfe-1h) ............................... (1)
On the same assumption,
V) ............................ (2)
Accordingly, if R be the whole resistance of the coil, and r the specific resist-
ance of copper,
As applicable to actual telephones we may take f = 1 centim., rj2 = S^ ; and
then R = 4firrn?. In C.G.S. measure r = 1600, and thus
w'= 477x1600 ............................... (4)
If the resistance be 100 ohms, .R = 10", and n = 2230.
When the resistance varies, other circumstances. remaining the same,
We have now to connect the periodic force upon the telephone-plate with
the periodic current in the coil. As has already been stated, only a very
rough estimate is possible a, priori. We will commence by considering the
case of an unlimited cylindrical core, divided by a transverse fracture into two
parts, and encompassed by an infinite cylindrical magnetizing coil containing
n turns to the centimetre. If 7 be the current, the magnetizing force BH
due to it is
(5)
If we regard the core as composed of soft iron, magnetized strongly by a
constant force H, the mechanical force with which the two parts attract one
another per unit of area is in the usual notation
and what we require is the variation of this quantity, when H becomes
H + 8H. This may be written
(6)
The value of dl/dH to be here employed is that appropriate to small
cyclical changes. It is greatest when 7 is small, and then* amounts to about
100/47T. As 7 increases, dl/dH diminishes, and finally approaches to zero in
the state of saturation. In order to increase (6) it is thus advisable to aug-
ment 7 up to a certain point, but not to approach saturation so nearly as to
* Phil. Mag. XXIH. p. 225 (1887). [Vol. n. p. 579.]
1894] ON A QUANTITATIVE THEORY OF THE TELEPHONE. 121
bring about a great diminution in the value of dl/dH. In the absence of
precise information we may estimate that the maximum of (6) will be reached
when / is about half the saturation value, or equal to 800*; and that dl/dH
also has half its maximum value, or 50/4-Tr. At this rate the force due to SH
is about 40,000 BH, reckoned per unit of area of the divided core, or by (5)
40,000 x 4-7TW7 (7)
But before (7) can be applied to the core of a telephone electromagnet it
must be subjected to large deductions. For in the telephone the total number
of Avindings n is limited to about one centimetre measured parallel to the
axis, whereas in (7) the electromagnet is supposed to be infinitely long, and
n denotes the number of windings per centimetre. If we are to suppose in
(7) that the windings are really limited to one centimetre, lying immediately
on one side of the division, there must be a loss of effect which I estimate at
5 times. We have now further to imagine the second part of the divided
cylinder to be replaced by the plate of the telephone, and that not in actual
contact with the remaining cylindrical part. The reduction of effect on this
account I estimate at 4 times "f*. The force on the telephone-plate per unit
area of core is thus
2000 x 477-717; (8)
or if, as for the telephone of 100-ohms resistance, n = 2200, and area of section
= '31 sq. cm.,
force =1-7 x 1077 (9)
In (9) the force is in dynes, and the current 7 is in c.G.8. measure. If F
denote the current reckoned in amperes,
force = 1-7 xlOT, (10)
and this must be supposed to be operative at the centre of the plate.
We shall presently consider what effect such a force may be expected to
produce ; but before proceeding to this I may record the result of some ex-
periments directed to check the applicability of (10), and made subsequently
to the theoretical estimates. A Bell telephone, similar to T1} was mounted
vertically, mouth downwards, having attached to the centre of its plate a
slender strip of glass. This strip was also vertical and carried at its lower
end a small scale-pan. The whole weight of the attachments was only '44
gram. The movement of the glass strip in the direction of its length was
observed through a reading-microscope focused upon accidental markings.
The telephone, itself of 70-ohms resistance, was connected through a revers-
ing-key with a Daniell cell and with an external resistance varied from time
to time. In taking an observation the current was first sent in such a direc-
tion as to depress the plate, and the web was adjusted upon the mark. The
* Ewing, Magnetic Induction, 1891, p. 136.
t I should say that these estimates were all made in ignorance of the result to which
they would lead.
122 ON A QUANTITATIVE THEORY OF THE TELEPHONE. [212
current was then reversed, by which the plate was drawn up, but by addition
of weights in the pan it was brought back again to the same position as
before. The force due to the current is thus measured by the half of the
weight applied.
The results were as follows : —
External resistance in ohms 100 200 500
Weight in grams 842
When 1000 ohms were included, the displacement on reversal was still just
visible. We may conclude that a force of 1 gram weight corresponds to a
current of about g^ of an ampere. Now, 1 gram weight is equal to 981 dynes,
so that for comparison with (10)
force = -6xlOT (11)
The force observed is thus about the third part of that which had been
estimated, and the agreement is sufficient.
Although not needed for the above comparison, we shall presently require
to know the linear displacement of the centre of the telephone-plate due to a
given force. Observations with the aid of a micrometer-eyepiece showed that
a force of 5 grams weight gave a displacement of 10~4 x 6'62 centim., or
10~4 x 1'32 for each gram, viz. 10~7 x T34 centim. per dyne. Thus by (11)
the displacement x due to a current T expressed in amperes is
#=-080r (12)
We have now to estimate what motion of the telephone-plate may be
expected to result from a given periodic force operating at its centre. The
effect depends largely upon the relation between the frequency of the imposed
vibration and those natural to the plate regarded as a freely vibrating body.
If we attempt to calculate the natural frequencies d priori, we are met by
uncertainty as to the precise mechanical conditions. From the manner in
which a telephone-plate is supported we should naturally regard the ideal
condition as one in which the whole of the circular boundary is clamped. On
this basis a calculation may be made, and it appears* that the frequency of
the gravest symmetrical mode should be about 991 in the case of the tele-
phone in question. But it may well be doubted whether we are justified in
assuming that the clamping is complete, and any relaxation tells in the
direction of a lowered frequency. A more trustworthy conclusion may per-
haps be founded upon the observed connexion between displacement and
force of restitution, coupled with an estimate of the inertia of the moving
parts. The total weight of the plate is 3'4 grams; the outside diameter
is 5'7 centim., and the inside diameter, corresponding to the free portion of
* Theory of Sound, 2nd ed. § 221 a.
1894] ON A QUANTITATIVE THEORY OF THE TELEPHONE. 123
the plate, is 4'5. The effective mass, supposed to be situated at the centre, I
estimate to be that corresponding to a diameter of 2*5 centim., viz. '65 gram.
A force of restitution per unit displacement equal to (10~7 x 1'34)~1, or
106 x 7*5, is supposed to urge the above mass to its position of equilibrium.
The frequency of the resulting vibration is
± /QO'x7-5)
27rV I "65 I
With the aid of a special electric maintenance the plate may be made to
speak on its own account. The frequency so found, viz. 896, corresponds
undoubtedly to a free vibration, but it does not follow that the vibration is
the gravest of which the plate is capable ; and there were indications pointing
to the opposite conclusion.
As it is almost impossible to form an d priori estimate of the amplitude
of vibration (#) when the frequency of the force is in the neighbourhood of
any of the free frequencies, I will take for calculation the case of frequency
256, which is presumably much lower than any of them. Under these
circumstances an " equilibrium theory " may be employed, the displacement
coexisting with any applied force being the same as if the force were perma-
nent. At this pitch the minimum current recorded in the table* is 8'3 x 10~7
amperes; so that by (12) the maximum excursion corresponding thereto is
given by x = '080 x 8'3 x 10~7= 6'8 x 10~8 centim.
The excursion thus found must not be compared with that calculated
formerly -f- for free progressive waves. The proper comparison is rather
between the condensations s in the two cases. In a progressive wave the
connexion between s and v, the maximum velocity, is v = as, where a is the
velocity of propagation. But in the present case the excursion x takes effect
upon a very small volume. If A be the effective area of the plate, and 8 the
whole volume included between the plate and the tympanum of the ear, we
may take s = AxjS. This relation assumes that the condensations and rare-
factions are uniform throughout the space in question, an assumption justified
by the smallness of its dimensions in comparison with the wave-length, and
further that the behaviour is the same as if the space were closed air-tight.
It would seem that a slight deficiency in the latter respect would not be
material.
For the numerical application I estimate that A = 4 sq. cm., S = 20 cub
cm. ; so that with the above value of as
s = l-4x!0-8, (13)
s being reckoned in atmospheres.
* Supra, p. 294. [Vol. iv. p. 117.]
t Proc. Roy. Soc. Vol. xxvi. p. 248 (1877). [Vol. i. p. 328.]
124 ON A QUANTITATIVE THEORY OF THE TELEPHONE. [212
The value of s corresponding to but just audible progressive waves of
frequency 256 was found to be 5'9 x 10~9, in sufficiently good agreement with
(13)*
But if the equilibrium theory be applied to the notes of higher pitch, such
as 512, we find the actual sensitiveness of the telephone greater than accord-
ing to the calculation. In this caset T = 7 x 10~8; so that by (12)
x = 5-6 x 10-9,
and
l-l x 10-9, ........................ (14)
decidedly smaller than that (4'5 x 10~9) deduced from the observations upon
progressive waves. The conclusion seems to be that for these frequencies the
equilibrium theory of the telephone-plate fails, and that in virtue of resonance
the sensitiveness of the instrument is specially exalted.
I will not dwell further upon these calculations, which involve too much
guesswork to be very satisfactory. They suffice, however, to show that the
"push and pull" theory is capable of giving an adequate account of the
action of the telephone, so far at least as my own observations are concerned.
But it is doubtful, to say the least, whether it could be reconciled with
estimates of sensitiveness such as those of Tait and of Preece.
* I hope shortly to publish an account of the observations upon which this statement is
founded. [See following Art. 213.]
t Supra, p. 294. [Vol. iv. p. 117.]
213.
ON THE AMPLITUDE OF AERIAL WAVES WHICH ARE BUT
JUST AUDIBLE. -
[Philosophical Magazine, XXXVIH. pp. 365—370, 1894*.]
THE problem of determining the .absolute value of the amplitude, or
particle velocity, of a sound which is but just audible to the ear, is one of
considerable difficulty. In a short paper published seventeen years agof I
explained a method by which it was easy to demonstrate a superior limit.
A whistle, blown under given conditions, consumes a known amount of
energy per second. Upon the assumption that the whole of this energy
is converted into sound, that the sound is conveyed without loss, and that
it is uniformly distributed over the surface of a hemisphere, it is easy to
calculate the amplitude at any distance; and the result is necessarily a
superior limit to the actual amplitude. In the case of the whistle experi-
mented on, of frequency 2730, the superior limit so arrived at for a sound
just easily audible was 8'1 x 10~8 cm. The maximum particle velocity v and
the maximum condensation s are the quantities more immediately determined
by the observations, and they are related by the well-known equation v = as,
in which a denotes the velocity of propagation. In the experiment above
referred to the superior limit for v was '0014 cm. per second, and that for s
was 4*1 x 10~8. I estimated that on a still night an amplitude, or velocity,
one-tenth of the above would probably be audible. A very similar number
has been arrived at by Wien|, who used an entirely different method§.
In connexion with calculations respecting the sensitiveness of telephones,
I was desirous of checking the above estimates, and made some attempts
to do so by the former method. In order to avoid possible complications of
* Bead at the Oxford Meeting of the British Association.
+ Proc. Roy. Soc. Vol. xxvi. p. 248 (1878). [Vol. i. p. 328.]
J Wied. Ann. xxxvi. p. 834 (1889).
§ The first estimate of the amplitude of but just audible sounds, with which I have only
recently become acquainted, is that of Topler and Boltzmann (Pogg. Ann. CXLI. p. 321 (1870)).
It depends upon an ingenious application of v. Helmholtz's theory of the open organ-pipe to data
relating to the maximum condensation within the pipe as obtained by the authors experimentally.
The value of s was found to be 6-5 x 10~8 for a pitch of 181.— August 21.
126 ON THE AMPLITUDE OF AERIAL WAVES [213
atmospheric refraction which may occur when large distances are in question,
I sought to construct pipes which should generate sound of given pitch upon
a much smaller scale, but with the usual economy of wind. In this I did not
succeed, and it seems as if there is some obstacle to the desired reduction of
scale.
The experiments here to be recorded were conducted with tuning-forks.
A fork of known dimensions, vibrating with a known amplitude, may be
regarded as a store of energy of which the amount may readily be calculated.
This energy is gradually consumed by internal friction and by generation
of sound. When a resonator is employed the latter element is the more
important, and in some cases we may regard the dying down of the amplitude
as sufficiently accounted for by the emission of sound. Adopting this view
for the present, we may deduce the rate of emission of sonorous energy from
the observed amplitude of the fork at the moment in question and from the
rate at which the amplitude decreases. Thus if the law of decrease be erW*
for the amplitude of the fork, or e~kt for the energy, and if E be the total
energy at time t, the rate at which energy is emitted at that time is —dE/dt,
or kE. The value of k is deducible from observations of the rate of decay,
e.g. of the time during which the amplitude is halved. With these arrange-
ments there is no difficulty in converting energy into sound upon a small
scale, and thus in reducing the distance of audibility to such a figure as
30 metres. Under these circumstances the observations are much more
manageable than when the operators are separated by half a mile, and there
is no reason to fear disturbance from atmospheric refraction.
The fork is mounted upon a stand to which is also firmly attached the
observing-microscope. Suitable points of light are obtained from starch
grains, and the line of light into which each point is extended by the
vibration is determined with the aid of an eyepiece-micrometer. Each
division of the micrometer-scale represents '001 centim. The resonator,
when in use, is situated in the position of maximum effect, with its mouth
under the free ends of the vibrating prongs.
The course of an experiment was as follows : — In the first place the rates
of dying down were observed, with and without the resonator, the stand being
situated upon the ground in the middle of a lawn. The fork was set in
vibration with a bow, and the time required for the double amplitude to fall
to half its original value was determined. Thus in the case of a fork of
frequency 256, the time during which the vibration fell from 20 micrometer-
divisions to 10 micrometer-divisions was 16s without the resonator, and 9s
when the resonator was in position. These times of halving were, as far as
could be observed, independent of the initial amplitude. To determine the
minimum audible, one observer (myself) took up a position 30 yards (27'4
metres) from the fork, and a second (Mr Gordon) communicated a large
1894] WHICH ARE BUT JUST AUDIBLE. 127
vibration to the fork. At the moment when the double amplitude measured
20 micrometer-divisions the second observer gave a signal, and immediately
afterwards withdrew to a distance. The business of the first observer was
to estimate for how many seconds after the signal the sound still remained
audible. In the case referred to the time was 12s. When the distance was
reduced to 15 yards (13*7 metres), an initial double amplitude of 10 micro-
meter-divisions was audible for almost exactly the same time.
These estimates of audibility are not made without some difficulty. There
are usually 2 or 3 seconds during which the observer is in doubt whether
he hears or only imagines, and different individuals decide the question in
opposite ways. There is also of course room for a real difference of hearing,
but this has not obtruded itself much. A given observer on a given day will
often agree with himself surprisingly well, but the accuracy thus suggested
is, I think, illusory. Much depends upon freedom from disturbing noises.
The wind in the trees or the twittering of birds embarrasses the observer,
and interferes more or less with the accuracy of results.
The equality of emission of sound in various horizontal directions was
tested, but no difference could be found. The sound issues almost entirely
from the resonator, and this may be expected to act as a simple source.
When the time of audibility is regarded as known, it is easy to deduce
•the amplitude of the vibration of the fork at the moment when the sound
ceases to impress the observer. From this the rate of emission of sonorous
energy and the amplitude of the aerial vibration as it reaches the observer
are to be calculated.
The first step in the calculation is the expression of the total energy
of the fork as a function of the amplitude of vibration measured at the
extremity of one of the prongs. This problem is considered in § 164 of
my Theory of Sound. If I be the length, p the density, and » the sectional
area of a rod clamped at one end and free at the other, the kinetic energy T
is connected with the displacement 17 at the free end by the equation (10)
At the moment of passage through the position of equilibrium 77 = 0 and
drjjdt has its maximum value, the whole energy being then kinetic. The
maximum value of drjfdt is connected with the maximum value of 77 by the
equation
so that if we now denote the double amplitude by 2?;, the whole energy of
the vibrating bar is
or for the two bars composing the fork
E=%pa>l7r*/T*.(2r))'2, ........................... (A)
where pwl is the mass of each prong.
128 ON THE AMPLITUDE OF AERIAL WAVES [213
The application of (A) to the 256-fork, vibrating with a double amplitude
of 20 micrometer-divisions, is as follows. We have
1 = 14-0 cm., «0 = '6xl-l=-66sq. cm.,
l/r = 256, p = 7'8, 277 = -050 cm.;
and thus
E = 4-06 x 103 ergs.
This is the whole energy of the fork when the actual double amplitude at
the ends of the prongs is '050 centim.
As has already been shown, the energy lost per second is kE, if the
amplitude vary as e~*kt. For the present purpose k must be regarded as
made up of two parts, one kt representing the dissipation which occurs in
the absence of the resonator, the other k2 due to the resonator. It is the
latter part only which is effective towards the production of sound. For
when the resonator is out of use the fork is practically silent; and, indeed,
even if it were worth while to make a correction on account of the residual
sound, its phase would only accidentally agree with that of the sound issuing
from the resonator.
The values of jfc, and k are conveniently derived from the times, ^ and t,
during which the amplitude falls to one-half. Thus
so that
&2 = 2 log,2 . (l/t - I/*,) = 1-386 (lit - I/tJ.
And the energy converted into sound per second is kzE.
We may now apply these formulae to the case, already quoted, of the
256-fork, for which £ = 9, ^ = 16. Thus t.2, the time which would be occupied
in halving the amplitude were the dissipation due entirely to the resonator,
is 20-6; and k, = '0674. Accordingly,
k*E = 267 ergs per second,
corresponding to a double amplitude represented by 20 micrometer-divisions.
In the experiment quoted the duration of audibility was 12 seconds, during
which the amplitude would fall in the ratio 212/9 : 1, and the energy in the
ratio 412/9 : 1. Hence at the moment when the sound was just becoming
inaudible the energy emitted as sound was 42'1 ergs per second*.
* It is of interest to compare with the energy-emission of a source of light. An incandescent
electric-lainp of 200 candles absorbs about a horse-power, or say 1010 ergs per second. Of the
total radiation only about TJff part acts effectively upon the eye ; so that radiation of suitable
quality consuming 5 x 103 ergs per second corresponds to a candle-power. This is about 104 times
that emitted as sound by the fork in the experiment described above. At a distance of 102 x 30,
or 3000 metres the stream of energy from the ideal candle would be about equal to the stream of
energy just audible to the ear. It appears that the streams of energy required to influence the
eye and the ear are of the same order of magnitude, a conclusion already drawn by Topler and
Boltzmann. — August 21.
1894] WHICH ARE BUT JUST AUDIBLE. 129
The question now remains, What is the corresponding amplitude or
condensation in the progressive aerial waves at 27 '4 metres from the source ?
If we suppose, as in my former calculations, that the ground reflects well,
we are to treat the waves as hemispherical. On the whole this seems to be
the best supposition to make, although the reflexion is doubtless imperfect.
The area S covered at the distance of the observer is thus 2?r x 27402 sq.
centim., and since*
S . %apv* = 8 . ^ptfs* = 421,
2_ 421
~7rx27402x -00125 x341003'
and s = 6'0 x 10~*.
The condensation s is here reckoned in atmospheres; and the result shows
that the ear is able to recognize the addition and subtraction of densities
far less than those to be found in our highest vacua.
The amplitude of aerial vibration is given by asrf^Tr, where l/r = 256,
and is thus equal to T27 x 10~7 cm.
It is to be observed that the numbers thus obtained are still somewhat
of the nature of superior limits, for they depend upon the assumption that
all the dissipation due to the resonator represents production of sound. This
may not be strictly the case even with the moderate amplitudes here in
question, but the uncertainty under this head is far less than in the case
of resonators or organ-pipes caused to speak by wind. From the nature of
the calculation by which the amplitude or condensation in the aerial waves
is deduced, a considerable loss of energy does not largely influence the final
numbers.
Similar experiments have been tried at various times with forks of pitch
384 and 512. The results were not quite so accordant as was at first hoped
might be the case, but they suffice to fix with some approximation the con-
densation necessary for audibility. The mean results are as follows : —
c', frequency = 256, s = 6'0 x 10~9,
gf, „ = 384, s = 4-6 x 10~9,
c", „ =512, s = 4-6 x 10-9,
no reliable distinction appearing between the two last numbers. Even the
distinction between 6'0 and 4'6 should be accepted with reserve ; so that the
comparison must not be taken to prove much more than that the condensation
necessary for audibility varies but slowly in the singly dashed octave.
* Theory of Sound, § 245.
214.
ARGON, A NEW CONSTITUENT OF THE ATMOSPHERE*
BY LORD RAYLEIGH, SEC. R.S., AND PROFESSOR WILLIAM RAMSAY, F.R.S.
[Philosophical Transactions, 186 (A), pp. 187—241, 1895.]
" Modern discoveries have not been made by large collections of facts, with subsequent
discussion, separation, and resulting deduction of a truth thus rendered perceptible. A few
facts have suggested an hypothesis, which means a supposition, proper to explain them.
The necessary results of this supposition are worked out, and then, and not till then, other
facts are examined to see if their ulterior results are found in Nature." — De Morgan,
A Budget of Paradoxes, Ed. 1872, p. 55.
1. Density of Nitrogen from Various Sources.
IN- a former paper-f- it has been shown that nitrogen extracted from
chemical compounds is about one-half per cent, lighter than " atmospheric
nitrogen."
The mean numbers for the weights of gas contained in the globe used
were as follows : —
grams.
From nitric oxide 2*3001
From nitrous oxide 2'2990
From ammonium nitrite .... 2'2987
while for " atmospheric " nitrogen there was found —
By hot copper, 1892 2'3103
By hot iron, 1893 2*3100
By ferrous hydrate, 1894 .... 2'3102
At the suggestion of Professor Thorpe, experiments were subsequently
tried with nitrogen liberated from urea by the action of sodium hypobromite.
* This memoir is included in the present collection by kind permission of Prof. Ramsay,
t Rayleigh, " On an Anomaly encountered in Determinations of the Density of Nitrogen Gas,"
Proc. Roy. Soc. Vol. LV. p. 340, 1894. [Vol. iv. p. 104.]
1895] ARGON, A NEW CONSTITUENT OF THE ATMOSPHERE. 131
The carbon and hydrogen of the urea are supposed to be oxidized by the
reaction to CO2 and H2O, the former of which would be retained by the
large excess of alkali employed. It was accordingly hoped that the gas
would require no further purification than drying. If it proved to be light,
it would at any rate be free from the suspicion of containing hydrogen.
The hypobromite was prepared from commercial materials in the pro-
portions recommended for the analysis of urea — 100 grams, caustic soda,
250 cub. centims. water, and 25 cub. centims. of bromine. For our purpose
about one and a half times the above quantities were required. The gas
was liberated in a bottle of about 900 cub. centims. capacity, in which a
vacuum was first established. The full quantity of hypobromite solution
was allowed to run in slowly, so that any dissolved gas might be at once
disengaged. The urea was then fed in, at first in a dilute condition, but,
as the pressure rose, in a 10 per cent, solution. The washing out of the
apparatus, being effected with gas in a highly rarefied state, made but a slight
demand upon the materials. The reaction was well under control, and the
gas could be liberated as slowly as desired.
In the first experiment, the gas was submitted to no other treatment
than slow passage through potash and phosphoric anhydride, but it soon
became apparent that the nitrogen was contaminated. The " inert and
inodorous " gas attacked vigorously the mercury of the Topler pump, and was
described as smelling like a dead rat. As to the weight, it proved to be in
excess even of the weight of atmospheric nitrogen.
The corrosion of the mercury and the evil smell were in great degree
obviated by passing the gas over hot metals. For the fillings of June 6,
9, 13, the gas passed through a short length of tube containing copper in
the form of fine wire, heated by a flat Bunsen burner, then through the
furnace over red-hot iron, and back over copper oxide. On June 19 the
furnace tubes were omitted, the gas being treated with the red-hot copper
only. The results, reduced so as to correspond with those above quoted,
were —
June 6 2-2978
„ 9 2-2987
„ 13 '. . 2-2982
„ 19 2-2994
Mean . . . . 2'2985
Without using heat it has not been found possible to prevent the cor-
rosion of the mercury. Even when no urea is employed, and air simply
bubbled through the hypobromite solution is allowed to pass with constant
shaking over mercury contained in a U-tube, the surface of the metal was
soon fouled. When hypochlorite was substituted for hypobromite in the last
9-2
132 ARGON, A NEW CONSTITUENT OF THE ATMOSPHERE. [214
experiment there was a decided improvement, and it was thought desirable
to try whether the gas prepared from hypochlorite and urea would be pure
on simple desiccation. A filling on June 25 gave as the weight 2-3343,
showing an excess of 36 mgs., as compared with other chemical nitrogen,
and of about 25 mgs. as compared with atmospheric nitrogen. A test with
alkaline pyrogallate appeared to prove the absence from this gas of free
oxygen, and only a trace of carbon could be detected when a considerable
quantity of the gas was passed over red-hot cupric oxide into solution of
baryta.
Although the results relating to urea nitrogen are interesting for com-
parison with that obtained from other nitrogen compounds, the original
object was not attained on account of the necessity of retaining the treatment
with hot metals. We have found, however, that nitrogen from ammonium
nitrite may be prepared without the employment of hot tubes, whose weight
agrees with that above quoted. It is true that the gas smells slightly of
ammonia, easily removable by sulphuric acid, and apparently also of oxides
of nitrogen. The solution of potassium nitrite and ammonium chloride was
heated in a water-bath, of which the temperature rose to the boiling-point
only towards the close of operations. In the earlier stages the temperature
required careful watching in order to prevent the decomposition taking place
too rapidly. The gas was washed with sulphuric acid, and after passing a
Nessler test, was finally treated with potash and phosphoric anhydride in the
usual way. The following results have been obtained : —
July 4 2-2983
„ 9 ....... 2-2989
13 . 2-2990
Mean .... 2'2987
It will be seen that in spite of the slight nitrous smell there is no appreciable
difference in the densities of gas prepared from ammonium nitrite with and
without the treatment by hot metals. The result is interesting, as showing
that the agreement of numbers obtained for chemical nitrogen does not
depend upon the use of a red heat in the process of purification.
The five results obtained in more or less distinct ways for chemical
nitrogen stand thus: —
From nitric oxide 2'3001
From nitrous oxide 2-2990
From ammonium nitrite purified at a red heat . . . 2'2987
From urea 2'2985
From ammonium nitrite purified in the cold . . . 2'2987
Mean . , . 2'2990
1895] ARGON, A NEW CONSTITUENT OF THE ATMOSPHERE. 133
These numbers, as well as those above quoted for " atmospheric nitrogen,"
are subject to a correction (additive)* of '0006 for the shrinkage of the globe
when exhausted -f-. If they are then multiplied in the ratio of 2'3108 : 1-2572,
they will express the weights of the gas in grams, per litre. Thus, as regards
.the mean numbers, we find as the weight per litre under standard conditions
of chemical nitrogen 1-2511, that of atmospheric nitrogen being 1'2572.
It is of interest to compare the density of nitrogen obtained from chemical
compounds with that of oxygen. We have N2 : 02 = 2'2996 : 2'6276 = O87517 ;
so that if 02 = 16, N2 = 14'003. Thus, when the comparison is with chemical
nitrogen, the ratio is very nearly that of 16 : 14. But if " atmospheric nitro-
gen " be substituted, the ratio of small integers is widely departed from.
The determination by Stas of the atomic weight of nitrogen from synthesis
of silver nitrate is probably the most trustworthy, inasmuch as the atomic
weight of silver was determined with reference to oxygen with the greatest
care, and oxygen is assumed to have the atomic weight 16. If, as found by
Stas, AgN03 : Ag = 1-57490 : 1, and Ag : 0 = 107'930 : 16, then
N : O = 14-049 : 16.
To the above list may be added nitrogen, prepared in yet another manner,
whose weight has been determined subsequently to the isolation of the new
dense constituent of the atmosphere. In this case nitrogen was actually
extracted from air by means of magnesium. The nitrogen thus separated
was then converted into ammonia by action of water upon the magnesium
nitride, and afterwards liberated in the free state by means of calcium hypo-
chlorite. The purification was conducted in the usual way, and included
passage over red-hot copper and copper oxide. The following was the
result : —
Globe empty, October 30, November 5 . . 2-82313
Globe full, October 31 "52395
Weight of gas 2-29918
It differs inappreciably from the mean of other results, viz., 2'2990, and is
of special interest as relating to gas which, at one stage of its history, formed
part of the atmosphere.
Another determination with a different apparatus of the density of
" chemical " nitrogen from the same source, magnesium nitride, which had
been prepared by passing " atmospheric " nitrogen over ignited magnesium,
may here be recorded. The sample differed from that previously mentioned,
inasmuch as it had not been subjected to treatment with red-hot copper.
[* In the Abstract of this paper (Proc. Roy. Soc. Vol. LVII. p. 265) the correction of -0006 was
erroneously treated as a deduction. — April, 1895.]
t Rayleigh, " On the Densities of the Principal Gases," Proc. Roy. Soc. Vol. Lin. p. 134, 1893.
[Vol. iv. p. 39.]
134 ARGON, A NEW CONSTITUENT OF THE ATMOSPHERE. [214
After treating the nitride with water, the resulting ammonia was distilled
off, and collected in hydrochloric acid ; the solution was evaporated to dry-
ness ; the dry ammonium chloride was dissolved in water, and its concentrated
solution added to a freshly prepared solution of sodium hypobromite. The
nitrogen was collected in a gas-holder over water which had previously been
boiled, so as at all events partially to expel air. The nitrogen passed into
the vacuous globe through a solution of potassium hydroxide, and through
two drying-tubes, one containing soda-lime, and the other phosphoric an-
hydride.
At 18'38° C. and 754'4 mgs. pressure, 162'843 cub. centims. of this nitrogen
weighed 0-18963 gram. Hence:—
Weight of 1 litre at 0°C. and 760 millims. pressure ... T2521 gram.
The mean result of the weight of 1 litre of "chemical" nitrogen has
been found to equal 1'2511. It is therefore seen that "chemical" nitrogen,
derived from "atmospheric" nitrogen, without any exposure to red-hot
copper, possesses the usual density.
Experiments were also made, which had for their object to prove that the
ammonia, produced from the magnesium nitride, is identical with ordinary
ammonia, and contains no other compound of a basic character. For this
purpose, the ammonia was converted into ammonium chloride, and the
percentage of chloride determined by titration with a solution of silver
nitrate which had been standardized by titrating a specimen of pure
sublimed ammonium chloride. The silver solution was of such a strength
that 1 cub. centim. precipitated the chlorine from O'OOITOI gram, of am-
monium chloride.
1. Ammonium chloride from orange-coloured sample of magnesium
nitride.
0'1106 gram, required 43'10 cub. centims. of silver nitrate = 66'35 per
cent, of chlorine.
2. Ammonium chloride from blackish magnesium nitride.
O'lllS gram, required 43'6 cub. centims. of silver nitrate = 66'35 per
cent, of chlorine.
3. Ammonium chloride from nitride containing a large amount of
unattacked magnesium.
0'0630 gram, required 24'55 cub. centims. of silver nitrate = 66'30 per
cent, of chlorine.
Taking for the atomic weights of hydrogen, H = T0032, of nitrogen,
N = 14-04, and of chlorine, Cl = 35'46, the theoretical amount of chlorine
in ammonium chloride is 66 '27 per cent.
1895] ARGON, A NEW CONSTITUENT OF THE ATMOSPHERE. 135
From these results — that nitrogen prepared from magnesium nitride
obtained by passing " atmospheric " nitrogen over red-hot magnesium has
the density of " chemical " nitrogen, and that ammonium chloride prepared
from magnesium nitride contains practically the same percentage of chlorine
as pure ammonium chloride — it may be concluded that red-hot magnesium
withdraws from " atmospheric " nitrogen no substance other than nitrogen
capable of forming a basic compound with hydrogen.
In a subsequent part of this paper, attention will again be called to this
statement. (See addendum, p. 240.)
2. Reasons for Suspecting a hitherto Undiscovered Constituent in Air.
When the discrepancy of weights was first encountered, attempts were
naturally made to explain it by contamination with known impurities. Of
these the most likely appeared to be hydrogen, present in the lighter gas,
in spite of the passage over red-hot cupric oxide. But, inasmuch as the
intentional introduction of hydrogen into the heavier gas, afterwards treated
in the same way with cupric oxide, had no effect upon its weight, this
explanation had to be abandoned; and, finally, it became clear that the
difference could not be accounted for by the presence of any known impurity.
At this stage it seemed not improbable that the lightness of the gas extracted
from chemical compounds Avas to be explained by partial dissociation of
nitrogen molecules N2 into detached atoms. In order to test this suggestion,
both kinds of gas were submitted to the action of the silent electric discharge,
with the result that both retained their weights unaltered. This was
discouraging, and a further experiment pointed still more markedly in the
negative direction. The chemical behaviour of nitrogen is such as to suggest
that dissociated atoms would possess a higher degree of activity, and that,
even though they might be formed in the first instance, their life would
probably be short. On standing, they might be expected to disappear, in
partial analogy with the known behaviour of ozone. With this idea in view,
a sample of chemically-prepared nitrogen was stored for eight months. But,
at the end of this time, the density showed no sign of increase, remaining
exactly as at first*.
Regarding it as established that one or other of the gases must be a
mixture, containing, as the case might be, an ingredient much heavier or
much lighter than ordinary nitrogen, we had to consider the relative pro-
babilities of the various possible interpretations. Except upon the already
discredited hypothesis of dissociation, it was difficult to see how the gas of
chemical origin could be a mixture. To suppose this would be to admit two
kinds of nitric acid, hardly reconcilable with the work of Stas and others
* Rayleigh, Proc. Bay. Soc. Vol. LV. p. 344, 1894. [Vol. iv. p. 108.]
136 ARGON, A NEW CONSTITUENT OF THE ATMOSPHERE. [214
upon the atomic weight of that substance. The simplest explanation in
many respects was to admit the existence of a second ingredient in air from
which oxygen, moisture, and carbonic anhydride had already been removed.
The proportional amount required was not great. If the density of the sup-
posed gas were double that of nitrogen, one-half per cent, only by volume
would be needed; or, if the density were but half as much again as that
of nitrogen, then one per cent, would still suffice. But in accepting this
explanation, even provisionally, we had to face the improbability that a
gas surrounding us on all sides, and present in enormous quantities, could
have remained so long unsuspected.
The method of most universal application by which to test whether a gas
is pure or a mixture of components of different densities is that of diffusion.
By this means Graham succeeded in effecting a partial separation of the
nitrogen and oxygen of the air, in spite of the comparatively small difference
of densities. If the atmosphere contain an unknown gas of anything like
the density supposed, it should be possible to prove the fact by operations
conducted upon air which had undergone atmolysis. If, for example, the
parts least disposed to penetrate porous walls were retained, the " nitrogen "
derived from it by the usual processes should be heavier than that derived
in like manner from unprepared air. This experiment, although in view
from the first, was not executed until a later stage of the inquiry (§ 6), when
results were obtained sufficient of themselves to prove that the atmosphere
contains a previously unknown gas.
But although the method of diffusion was capable of deciding the main,
or at any rate the first question, it held out no prospect of isolating the new
constituent of the atmosphere, and we therefore turned our attention in the
first instance to the consideration of methods more strictly chemical. And
here the question forced itself upon us as to what really was the evidence
in favour of the prevalent doctrine that the inert residue from air after
withdrawal of oxygen, water, and carbonic anhydride, is all of one kind.
The identification of " phlogisticated air " with the constituent of nitric
acid is due to Cavendish, whose method consisted in operating with electric
sparks upon a short column of gas confined with potash over mercury at
the upper end of an inverted U-tube*. This tube (M) was only about
^ inch in diameter, and the column of gas was usually about 1 inch in
length. After describing some preliminary trials, Cavendish proceeds : —
" I introduced into the tube a little soap-lees (potash), and then let up some
dephlogisticatedf and common air, mixed in the above-mentioned proportions
* "Experiments on Air," Phil. Trans. Vol. LXXV. p. 372, 1785.
[t The explanation of combustion in Cavendish's day was still vague. It was generally
imagined that substances capable of burning contained an unknown principle, to which the name
" phlogiston " was applied, and which escaped during combustion. Thus, metals and hydrogen
and other gases were said to be " phlogisticated " if they were capable of burning in air. Oxygen
1895] ARGON, A NEW CONSTITUENT OF THE ATMOSPHERE. 137
which rising to the top of the tube M, divided the soap-lees into its two
legs. As fast as the air was diminished by the electric spark, I continued
adding more of the same kind, till no further diminution took place : after
which a little pure dephlogisticated air, and after that a little common air,
were added, in order to see whether the cessation of diminution was not
owing to some imperfection in the proportion of the two kinds of air to
each other; but without effect. The soap-lees being then poured out of
the tube, and separated from the quicksilver, seemed to be perfectly neutra-
lised, and they did not at all discolour paper tinged with the juice of blue
flowers. Being evaporated to dryness, they left a small quantity of salt,
which was evidently nitre, as appeared by the manner in which paper,
impregnated with a solution of it, burned."
Attempts to repeat Cavendish's experiment in Cavendish's manner have
only increased the admiration with which we regard this wonderful investi-
gation. Working on almost microscopical quantities of material, and by
operations extending over days and weeks, he thus established one of the
most important facts in chemistry. And what is still more to the purpose,
he raises as distinctly as we could do, and to a certain extent resolves, the
question above suggested. The passage is so important that it will be
desirable to quote .it at full length.
" As far as the experiments hitherto published extend, we scarcely know
more of the phlogisticated part of our atmosphere than that it is not
diminished by lime-water, caustic alkalies, or nitrous air; that it is unfit
to support fire or maintain life in animals ; and that its specific gravity is
not much less than that of common air; so that, though the nitrous acid,
by being united to phlogiston, is converted into air possessed of these
properties, and consequently, though it was reasonable to suppose, that part
at least of the phlogisticated air of the atmosphere consists of this acid
united to phlogiston, yet it was fairly to be doubted whether the whole
is of this kind, or whether there are not in reality many different substances
confounded together by us under the name of phlogisticated air. . I therefore
made an experiment to determine whether the whole of a given portion of
the phlogisticated air of the atmosphere could be reduced to nitrous acid, or
whether there was not a part of a different nature to the rest which would
refuse to undergo that change. The foregoing experiments indeed in some
measure decided this point, as much the greatest part of the air let up into
the tube lost its elasticity; yet as some remained unabsorbed it did not
appear for certain whether that was of the same nature as the rest or not.
being non-inflammable was named " dephlogisticated air," and nitrogen, because it was incapable
of supporting combustion or life was named by Priestley " phlogisticated air," although up till
Cavendish's time it had not been made to unite with oxygen.
The term used for oxygen by Cavendish is " dephlogisticated air," and for nitrogen, " phlogis-
ticated air."— April, 1895.]
138 ARGON, A NEW CONSTITUENT OF THE ATMOSPHERE. [214
For this purpose I diminished a similar mixture of dephlogisticated and
common air, in the same manner as before, till it was reduced to a small
part of its original bulk. I then, in order to decompound as much as I could
of the phlogisticated air which remained in the tube, added some dephlo-
gisticated air to it and continued the spark until no further diminution
took place. Having by these means condensed as much as I could of the
phlogisticated air, I let up some solution of liver of sulphur to absorb the
dephlogisticated air ; after which only a small bubble of air remained
unabsorbed, which certainly was not more than T|7 of the bulk of the
phlogisticated air let up into the tube ; so that, if there is any part of the
phlogisticated air of our atmosphere which differs from the rest, and cannot
be reduced to nitrous acid, we may safely conclude that it is not more than'
T27 Part °f tne whole."
Although Cavendish was satisfied with his result, and does not decide
whether the small residue was genuine, our experiments about to be related
render it not improbable that his residue was really of a different kind from
the main bulk of the " phlogisticated air," and contained the gas now called
argon.
Cavendish gives data* from which it is possible to determine the rate of
absorption of the mixed gases in his experiment. The electrical machine
used " was one of Mr Nairne's patent machines, the cylinder of which is
12£ inches long and 7 in diameter. A conductor, 5 feet long and 6 inches
in diameter, was adapted to it, and the ball which received the spark was
placed two or three inches from another ball, fixed to the end of the
conductor. Now, when the machine worked well, Mr Gilpin supposes he
got about two or three hundred sparks a minute, and the diminution of the
air during the half hour which he continued working at a time varied in
general from 40 to 120 measures, but was usually greatest when there was
most air in the tube, provided the quantity was not so great as to prevent
the spark from passing readily." The " measure " spoken of represents the
volume of one grain of quicksilver, or '0048 cub. centim., so that an absorp-
tion of one cub. centim. of mixed gas per hour was about the most favourable
rate. Of the mixed gas about two-fifths would be nitrogen.
3. Methods of Causing Free Nitrogen to Combine.
The concord between the determinations of density of nitrogen obtained
from sources other than the atmosphere, having made it at least probable
that some heavier gas exists in the atmosphere, hitherto undetected, it
became necessary to submit atmospheric nitrogen to examination, with a
view of isolating, if possible, the unknown and overlooked constituent, or it
might be constituents.
* Phil. Trans. Vol. LXXVHI. p. 271, 1788.
1895] ARGON, A NEW CONSTITUENT OF THE ATMOSPHERE. 139
Nitrogen, however, is an element which does not easily enter into direct
combination with other elements; but with certain elements, and under
certain conditions, combination may be induced. The elements which have
been directly united to nitrogen are (a) boron, (6) silicon, (c) titanium,
(d) lithium, (e) strontium and barium, (/) magnesium, (g) aluminium,
(h) mercury, (i) manganese, (j) hydrogen, and (k) oxygen, the last two by
help of an electrical discharge.
(a) Nitride of boron was prepared by Wohler and Deville* by heating
amorphous boron to a white heat in a current of nitrogen. Experiments
were made to test whether the reaction would take place in a tube of
difficultly fusible glass; but it was found that the combination took place
at a bright red heat to only a small extent, and that the boron, which had
been prepared by heating powdered boron oxide with magnesium dust, was
only superficially attacked. Boron is, therefore, not a convenient absorbent
for nitrogen. [M. Moissan informs us that the reputation it possesses is
due to the fact that early experiments were made with boron which had
been obtained by means of sodium, and which probably contained a boride
of that metal— April, 1895.]
(6) Nitride of silicon^ also requires for its formation a white heat, and
complete union is difficult to bring about. Moreover, it is not easy to obtain
large quantities of silicon. This method was therefore not attempted.
(c) Nitride of titanium is said to have been formed by Deville and
CaronJ, by heating titanium to whiteness in a current of nitrogen. This
process was not tried by us. As titanium has an unusual tendency to unite
with nitrogen, it might, perhaps, be worth while to set the element free in
presence of atmospheric nitrogen, with a view to the absorption of the
nitrogen. This has, in effect, been already done by Wohler and Deville§;
they passed a mixture of the vapour of titanium chloride and nitrogen over
red-hot aluminium, and obtained a large yield of nitride. It is possible that
a mixture of the precipitated oxide of titanium with magnesium dust might
be an effective absorbing agent at a comparatively low temperature. [Since
writing the above we have been informed by M. Moissan that titanium,
heated to 800°, burns brilliantly in a current of nitrogen. It might there-
fore be used with advantage to remove nitrogen from air, inasmuch as we
have found that it does not combine with argon. — April, 1895.]
(d), (e) Lithium at a dull red heat absorbs nitrogen ||, but the difficulty
of obtaining the metal in quantity precludes its application. On the other
* Annales de Chimie, (3), LIT. p. 82.
t Schutzenberger, Comptes Eendus, LXXXIX. 644.
J Annalen der Chemie u. Pharmacie, ci. 360.
§ Annalen der Chemie w. Pharmacie, LXXIII. 34.
|| Ouvrard, Comptes Eendus, cxiv. 120.
140 ARGON, A NEW CONSTITUENT OF THE ATMOSPHERE. [214
hand, strontium and barium, prepared by electrolysing solutions of their
chlorides in contact with mercury, and subsequently removing the mercury
by distillation, are said by Maquenne* to absorb nitrogen with readiness.
Although we have not tried these metals for removing nitrogen, still our
experience with their amalgams has led us to doubt their efficacy, for it is
extremely difficult to free them from mercury by distillation, and the product
is a fused ingot, exposing very little surface to the action of the gas. The
process might, however, be worth a trial.
Barium is the efficient absorbent for nitrogen when a mixture of barium
carbonate and carbon is ignited in a current of nitrogen, yielding cyanide.
Experiments have shown, however, that the formation of cyanides takes
place much more readily and abundantly at a high temperature, a tempe-
rature not easily reached with laboratory appliances. Should the process
ever come to be worked on a large scale, the gas rejected by the barium will
undoubtedly prove a most convenient source of argon.
(/) Nitride of magnesium was prepared by Deville and Caron (loc. tit.)
during the distillation of impure magnesium. It has been more carefully
investigated by Briegleb and Geuther-f, who obtained it by igniting metallic
magnesium in a current of nitrogen. It forms an orange-brown, friable
substance, very porous, and it is easily produced at a bright red heat. When
magnesium, preferably in the form of thin turnings, is heated in a combustion
tube in a current of nitrogen, the tube is attacked superficially, a coating
of magnesium silicide being formed. As the temperature rises to bright
redness, the magnesium begins to glow brightly, and combustion takes place,
beginning at that end of the tube through which the gas is introduced.
The combustion proceeds regularly, the glow extending down the tube, until
all the metal has united with nitrogen. The heat developed by the combi-
nation is considerable, and the glass softens; but by careful attention and
regulation of the rate of the current, the tube lasts out an operation. A
piece of combustion tubing of the usual length for organic analysis packed
tightly with magnesium turnings, and containing about 30 grams., absorbs
between seven and eight litres of nitrogen. It is essential that oxygen be
excluded from the tube, otherwise a fusible substance is produced, possibly
nitrate, which blocks the tube. With the precaution of excluding oxygen,
the nitride is loose and porous, and can easily be removed from the tube with
a rod ; but it is not possible to use a tube twice, for the glass is generally
softened and deformed.
(<7) Nitride of aluminium has been investigated by Mallet J. He ob-
tained it in crystals by heating the metal to whiteness in a carbon crucible.
* Ouvrard, Comptes Bendus, cxiv. 25, and 220.
t Annalen der Chemie u. Pharmacie, cxxm. 228.
$ Journ. Chem. Soc. 1876, Vol. u. p. 349.
1895] ARGON, A NEW CONSTITUENT OF THE ATMOSPHERE. 141
But aluminium shows no tendency to unite with nitrogen at a red heat,
and cannot be used as an absorbent for the gas.
(k) Gerresheim * states that he has induced combination between nitrogen
and mercury; but the affinity between these elements is of the slightest, for
the compound is explosive.
(i) In addition to these, metallic manganese in a finely divided state has
been shown to absorb nitrogen at a not very elevated temperature, forming
a nitride of the formula Mn5N2f.
(j) [A mixture of nitrogen with hydrogen, standing over acid, is absorbed
at a fair rate under the influence of electric sparks. But with an apparatus
such as that shown in Fig. 1, the efficiency is but a fraction (perhaps ^) of
that obtainable when oxygen is substituted for hydrogen and alkali for acid.
—April, 1895.]
4. Early Experiments on sparking Nitrogen with Oxygen in presence of
Alkali. '
In our earliest attempts to isolate the suspected gas by the method of
Cavendish, we used a Ruhmkorff coil of medium size actuated by a battery
of five Grove cells. The gases were contained in a test-tube A, Fig. 1,
standing over a large quantity of weak alkali B, and the current was con-
veyed in wires insulated by U-shaped glass tubes CC passing through the
liquid round the mouth of the test-tube. The inner platinum ends DD of
the wires were sealed into the glass insulating tubes, but reliance was not
placed upon these sealings. In order to secure tightness in spite of cracks,
mercury was placed in the bends. This disposition of the electrodes compli-
cates the apparatus somewhat and entails the use of a large depth of liquid
in order to render possible the withdrawal of the tubes, but it has the great
advantage of dispensing with sealing electrodes of platinum into the prin-
cipal vessel, which might give way and cause the loss of the experiment at
the most inconvenient moment. With the given battery and coil a some-
what short spark, or arc, of about 5 millims. was found to be more favourable
than a longer one. When the mixed gases were in the right proportion, the
rate of absorption was about 30 cub. centims. per hour, or 30 times as fast
as Cavendish could work with the electrical machine of his day.
To take an example, one experiment of this kind started with 50 cub.
centims. of air. To this, oxygen was gradually added until, oxygen being in
excess, there was no perceptible contraction during an hour's sparking. The
remaining gas was then transferred at the pneumatic trough to a small
* Annalen der Chemie u. Pharmacie, cxcv. 373.
t 0. Prehlinger, Monatsh.f. Chemie, xv. 391.
142
ARGON, A NEW CONSTITUENT OF THE ATMOSPHERE.
[214
measuring vessel, sealed by mercury, in which the volume was found to be
TO cub. centim. On treatment with alkaline pyrogallate, the gas shrank
to "32 cub. centim. That this small residue could not be nitrogen was
argued from the fact that it had withstood the prolonged action of the
spark, although mixed with oxygen in nearly the most favourable proportion.
Fig. 1.
The residue was then transferred to the test-tube with an addition of
another 50 cub. centims. of air, and the whole worked up with oxygen as
before. The residue was now 2'2 cub. centims., and, after removal of oxygen,
'76 cub. centim.
1895] ARGON, A NEW CONSTITUENT OF THE ATMOSPHERE. 143
Although it seemed almost impossible that these residues could be either
nitrogen or hydrogen, some anxiety was not unnatural, seeing that the final
sparking took place under somewhat abnormal conditions. The space was
very restricted, and the temperature (and with it the proportion of aqueous
vapour) was unduly high. But any doubts that were felt upon this score
were removed by comparison experiments in which the whole quantity of
air operated on was very small. Thus, when a mixture of 5 cub. centims. of
air with 7 cub. centims. of oxygen was sparked for one hour and a quarter,
the residue was '47 cub. centim., and, after removal of oxygen, '06 cub. centim.
Several repetitions having given similar results, it became clear that the final
residue did not depend upon anything that might happen when sparks passed
through a greatly reduced volume, but was in proportion to the amount of air
operated upon.
No satisfactory examination of the residue which refused to be oxidised
could be made without the accumulation of a larger quantity. This, however,
was difficult of attainment at the time in question. The gas seemed to rebel
against the law of addition. It was thought that the cause probably lay in
the solubility of the gas in water, a suspicion since confirmed. At length,
however, a sufficiency was collected to allow of sparking in a specially con-
structed tube, when a comparison with the air spectrum taken under similar
conditions proved that, at any rate, the gas was not nitrogen. At first
scarcely a trace of the principal nitrogen lines could be seen, but after
standing over water for an hour or two these lines became apparent.
[The apparatus shown in Fig. 1 has proved to be convenient for the puri-
fication of small quantities of argon, and for determinations of the amount of
argon present in various samples of gas, e.g., in the gases expelled from
solution in water. To set it in action an alternating current is much to be
preferred to a battery and break. At the Royal Institution the primary
of a small RuhmkorfF was fed from the 100- volt alternating current supply,
controlled by two large incandescent lamps in series with the coil. With this
arrangement the voltage at the terminals of the secondary, available for
starting the sparks, was about 2000, and could be raised to 4000 by plugging
out one of the lamps. With both lamps in use the rate of absorption of
mixed gases was 80 cub. centims. per hour, and this was about as much as
could well be carried out in a test-tube. Even with this amount of power it
was found better to abandon the sealings at D. No inconvenience arises from
the open ends, if the tubes are wide enough to ensure the liberation of any
gas included over the mercury when they are sunk below the liquid.
The power actually expended upon the coil is very small. When the
apparatus is at work the current taken is only 2 "4 amperes. As regards
the voltage, by far the greater part is consumed in the lamps. The efficient
voltage at the terminals of the primary coil is best found indirectly. Thus, if
144
ARGON, A NEW CONSTITUENT OF THE ATMOSPHERE.
[214
A be the current in amperes, V the total voltage, V1 the voltage at the
terminals of the coil, F2 that at the terminals of the lamps, the watts
used are*
In the present case a Cardew voltmeter gave F= 90£, F2 = 88; and
in the formula may be neglected. Thus,
= 2*4 x 2'5 = 6'0 approximately.
The work consumed by the coil when the sparks are passing is, thus, less
than y^ of a horse-power ; but, in designing an apparatus, it must further be
remembered that in order to maintain the arc, a pretty high voltage is
required at the terminals of the secondary when no current is passing in it. —
April, 1895.]
5. Early Experiments on Withdrawal of Nitrogen from Air by
means of Red-hot Magnesium.
It having been proved that nitrogen, at a bright red heat, was easily
absorbed by magnesium, best in the form of turnings, an attempt was success-
fully made to remove that gas from the residue left after eliminating oxygen
from air by means of red-hot copper.
Fig. 2.
The preliminary experiment was made in the following manner: —
A combustion tube, A, was filled with magnesium turnings, packed tightly
by pushing them in with a rod. This tube was connected with a second
piece of combustion tubing, B, by means of thick-walled india-rubber tubing,
* Ayrton and Sumpner, Proc. Roy. Soc. Vol. XLIX. p. 427, 1891.
1895] ARGON, A NEW CONSTITUENT OF THE ATMOSPHERE. 145
carefully wired ; B contained copper oxide, and, in its turn, was connected
with the tube CD, one-half of which contained soda-lime previously ignited to
expel moisture, while the other half was filled with phosphoric anhydride.
E is a measuring vessel, and F is a gas-holder containing "atmospheric
nitrogen."
In beginning an experiment, the tubes were heated with long-flame
burners, and pumped empty; a little hydrogen was formed by the action of
the moisture on the metallic magnesium ; it was oxidised by the copper oxide
and absorbed by the phosphoric pentoxide. A gauge attached to the
Sprengel's pump, connected with the apparatus, showed when a vacuum
had been reached. A quantity of nitrogen was then measured in E, and
admitted into contact with the red-hot magnesium. Absorption took place,
rapidly at first and then slowly, as shown by the gauge on the Sprengel's
pump. A fresh quantity was then measured and admitted, and these
operations were repeated until no more could be absorbed. The system of
tubes was then pumped empty by means' of the Sprengel's pump, and the
gas was collected. The magnesium tube was then detached and replaced
by another. The unabsorbed gas was returned to the measuring-tube by a
device shown in the figure (G) and the absorption recommenced. After 1094
cub. centims. of gas had been thus treated, there was left about 50 cub.
centims. of gas, which resisted rapid absorption. It still contained nitrogen,
however, judging by the diminution of volume which it experienced when
allowed to stand in contact with red-hot magnesium. Its density was,
nevertheless, determined by weighing a small bulb of about 40 cub. centims.
capacity, first with air, and afterwards with the gas. The data are these : —
grm.
(a) Weight of bulb and air — that of glass counterpoise . . 0'8094
„ „ alone — that of glass counterpoise . . . 07588
0-0506
(6) Weight of bulb and gas — that of glass counterpoise . . 0'8108
„ „ alone — that of glass counterpoise . . . 0'7588
gas 0-0520
Taking as the weight of a litre of air, 1'29347 grms., the mean of the
latest results, and of oxygen (=16) 1'42961 grms.*, the density of the
residual gas is 14'88.
This result was encouraging, although weighted with the unavoidable
error attaching to the weighing of a very small amount. Still the fact
remains that the supposed nitrogen was heavier than air. It would hardly
have been possible to make a mistake of 2'7 milligrams.
* For note see foot of p. 146.
R. IV. 10
146
ARGON, A NEW CONSTITUENT OF THE ATMOSPHERE.
[214
It is right here to place on record the fact that this first experiment
was to a great extent carried out by Mr Percy Williams, to whose skill
in manipulation and great care its success is due, and to whom we desire
here to express our thanks.
Experiments were now begun on a larger scale, the apparatus employed
being shown in Figs. 3 and 4.
Fig. 3.
A and B are large glass gas-holders of about 10 litres capacity. C is an
arrangement by which gas could be introduced at will into the gas-holder A,
either by means of an india-rubber tube slipped over the open end of the
U-tube, or, as shown in the figure, from a test-tube. The tube D was half
* The results on which this and the subsequent calculations are based are as follows (the
weights are those of 1 litre) : —
Air
Oxygen
Nitrogen
Hydrogen
Regnault
1-29349
1-43011
1-25647
0-08988
Von Jolly
1-29383
1-42971
1-25819
Leduc
1-29330
1-42910
1-25709
0-08985
Rayleigh
1-29327
1-42952
1-25718
0-09001
Regnault's numbers have an approximate correction applied to them by Crafts. The mean of
these numbers is taken, that of Regnault for nitrogen being omitted, as there is reason to believe
that his specimen was contaminated with hydrogen.
Air
Oxygen
Nitrogen
Hydrogen
1-29347
1-42961
1-25749
0-08991
This ratio gives for air the composition by volume —
Oxygen 20-91 per cent.
Nitrogen 79-09
a result verified by experiment.
It is, of course, to be understood that these densities of nitrogen refer to atmospheric nitrogen,
that is, to air from which oxygen, water vapour carbon dioxide, and ammonia have been removed.
1895] AEGON, A NEW CONSTITUENT OF THE ATMOSPHERE. 147
filled with soda-lime (a), half with phosphoric anhydride (6). Similarly, the
tube E, which was kept at a red heat by means of the long-flame burner, was
filled half with very porous copper (a), reduced from dusty oxide by heating
in hydrogen, half with copper oxide in a granular form (6). The next tube,
F, contained granular soda-lime, while G contained magnesium turnings, also
heated to bright redness by means of a long-flame burner. H contained
phosphoric anhydride, and / soda-lime. All joints were sealed, excepting
those connecting the hard-glass tubes E and G to the tubes next them.
The gas-holder A having been filled with nitrogen, prepared by passing air
over red-hot copper, and introduced at 0, the gas was slowly passed through
the system of tubes into the gas-holder B, and back again. The magnesium
in the tube G having then ceased to absorb was quickly removed and
replaced by a fresh tube. This tube was of course full of air, and before the
tube G was heated, the air was carried back from B towards A by passing a
little nitrogen from right to left. The oxygen in the air was removed by the
metallic copper, and the nitrogen passed into the gas-holder A, to be returned
in the opposite direction to B.
Fig. 4.
In the course of about ten days most of the nitrogen had been absorbed.
The magnesium was not always completely exhausted; usually the nitride
presented the appearance of a blackish-yellow mass, easily shaken out of the
tube. It is needless to say that the tube was always somewhat attacked,
becoming black with a coating of magnesium silicide. The nitride of mag-
nesium, whether blackish or orange, if left for a few hours exposed to moist
air, was completely converted into white, dusty hydroxide, and during
exposure it gave off a strong odour of ammonia. If kept in a stoppered
bottle, however, it was quite stable.
It was then necessary, in order to continue the absorption, to carry on
operations on a smaller scale, with precautions to exclude atmospheric air as
completely as possible. There was at this stage a residue of 1500 cub.
centims.
10—2
148 ARGON, A NEW CONSTITUENT OF THE ATMOSPHERE. [214
The apparatus was therefore altered to that shown in Fig. 4, so as to make
it possible to withdraw all the gas out of the gas-holder A.
The left-hand exit led to the Sprengel's pump ; the compartment (a) of
the drying-tube B was filled with soda-lime, and (b) with phosphoric anhydride.
G is a tube into which the gas could be drawn from the gas-holder A. The
stop-cock, as shown, allows gas to pass through the horizontal tubes, and does
not communicate with A ; but a vertical groove allows it to be placed in com-
munication either with the gas-holder, or with the apparatus to the right.
The compartment (a) of the second drying-tube D contained soda-lime, and
(b) phosphoric anhydride. The tube D communicated with a hard-glass tube
E, heated over a long-flame burner ; it was partly filled with metallic copper,
and partly with copper oxide. This tube, as well as the tube F filled with
magnesium turnings, was connected to the drying-tube with india-rubber.
The gas then entered G, a graduated reservoir, and the arrangement H
permitted the removal or introduction of gas from or into the apparatus. The
gas was gradually transferred from the gas-holder to the tube C, and passed
backwards and forwards over the red-hot magnesium until only about 200 cub.
centims. were left. It was necessary to change the magnesium tube, which
was made of smaller size than formerly, several times during the operation.
This was done by turning out the long-flame burners and pumping off all gas
in the horizontal tubes by means of the Sprengel's pump. This gas was
carefully collected. The magnesium tube was then exchanged for a fresh
one, and after air had been exhausted from the apparatus, nitrogen was intro-
duced from the reservoir. Any gas evolved from the magnesium (and
apparently there was always a trace of hydrogen, either occluded by the
magnesium, or produced by the action of aqueous vapour on the metal) was
oxidised by the copper oxide. Had oxygen been present, it would have been
absorbed by the metallic copper, but the copper preserved its red appearance
without alteration, whereas a little copper oxide was reduced during the
series of operations. The gas, which had been removed by pumping, was
reintroduced at H, and the absorption continued.
The volume of the gas was thus, as has been said, reduced to about 200
cub. centims. It would have been advisable to take exact measurements, but,
unfortunately, some of the original nitrogen had been lost through leakage ;
and a natural anxiety to see if there was any unknown gas led to pushing on
operations as quickly as possible.
The density of the gas was next determined. The bulb or globe in which
the gas was weighed was sealed to a two-way stop-cock, and the weight of
distilled and air-free water filling it at 17'15° was 162'654 grms., correspond-
ing to a capacity of 162'843 cub. centims. The shrinkage on removing air
completely was 0'0212 cub. centim. Its weight, when empty, should therefore
be increased by the weight of that volume of air, which may be taken as
1895] ARGON, A NEW CONSTITUENT OF THE ATMOSPHERE. 149
O000026 grm. This correction, however, is perhaps hardly worth applying in
the present case.
The counterpoise was an exactly similar bulb of equal capacity, and
weighing about 0*2 grm. heavier than the empty globe. The balance was a
very sensitive one by Oertling, which easily registered one-tenth of a milligrm.
By the process of swinging, one-hundredth of a milligrm. could be determined
with fair accuracy.
In weighing the empty globe, 0'2 grm. was placed on the same pan as that
which hung from the end of the beam to which it was suspended, and the final
weight was adjusted by means of a rider, or by small weights on the other
pan. This process practically leads to weighing by substitution of gas for
weights. The bulb was always handled with gloves, to avoid moisture or
grease from the fingers.
Three experiments, of which it is unnecessary to give details, were made
to test the degree of accuracy with which a gas could be. weighed, the gas
being dried air, freed from carbon dioxide. The mean result gave for the
weight of one litre of air at 0° and 760 millims. pressure, 1*2935 grm.
Regnault found 1 '29340, a correction having been applied by Crafts to allow
for the estimated alteration of volume caused by the contraction of his
vacuous bulb. The mean result of determinations by several observers is
1-29347 ; while one of us found 1 "29327.
The globe was then filled with the carefully dried gas.
Temperature, 18'80°. Pressure, 759*3 millims.
Weight of 162-843 cub. centims. of gas 0'21897 grm.
Weight of 1 litre gas at 0° and 760 millims 1-4386
Density, that of air compared with O, = 16, being 14'476 16'100 grms.
It is evident from these numbers that the dense constituent of the air was
being concentrated. As a check, the bulb was pumped empty and again
weighed ; its weight was 0'21903 grm. This makes the density 16105.
It appeared advisable to continue to absorb nitrogen from this gas. The
first tube of magnesium removed a considerable quantity of gas ; the nitride
was converted into ammonium chloride, and the sample contained 6 6 '30 per
cent, of chlorine, showing, as has before been remarked, that if any of the
heavier constituent of the atmosphere had been absorbed, it formed no basic
compound with hydrogen. The second tube of magnesium was hardly
attacked; most of the magnesium had melted, and formed a layer at the
lower part of the tube. That which was still left in the body of the tube was
black on the surface, but had evidently not been much attacked. The
ammonium chloride which it yielded weighed only 0'0035 grm.
150 ARGON, A NEW CONSTITUENT OF THE ATMOSPHERE. [214
The density of the remaining gas was then determined. But as its
volume was only a little over 100 cub. centims., the bulb, the capacity of
which was 162 cub. centims., had to be filled at reduced pressure. This was
easily done by replacing the pear-shaped reservoir of the mercury gas-holder
by a straight tube, and noting the level of the mercury in the gas-holder and
in the tube which served as a mercury reservoir against a graduated mirror-
scale by help of a cathetometer at the moment of closing the stop-cock of the
density bulb.
The details of the experiment are these : —
Temperature, 19'12° C. Barometric pressure, 749'8 millims. (corr.).
Difference read on gas-holder and tube, 225'25 millims. (corr.).
Actual pressure, 524'55 millims.
Weight of 162-843 cub. centims. of gas .... 017913 grm.
Weight of 1 litre at 0° and 760 millims. pressure . T7054
Density 19'086 grms.
This gas is accordingly at least 19 times as heavy as hydrogen.
A portion of the gas was then mixed Avith oxygen, and submitted to a
rapid discharge of sparks for four hours in presence of caustic potash. It
contracted, and on absorbing the excess of oxygen with pyrogallate of
potassium the contraction amounted to 15'4 per cent, of the original volume.
The question then arises, if the gas contain 15'4 per cent, of nitrogen, of
density 14'014, and 84'6 per cent, of other gas, and if the density of the
mixture were 19'086, what would be the density of the other gas ? Calcula-
tion leads to the number 20'0.
A vacuum-tube was filled with a specimen of the gas of density 19'086,
and it could not be doubted that it contained nitrogen, the bands of which
were distinctly visible. It was probable, therefore, that the true density of
the pure gas lay not far from 20 times that of hydrogen. At the same time
many lines were seen which could not be recognised as belonging to the
spectrum of any known substance.
Such were the preliminary experiments made with the aid of magnesium
to separate from atmospheric nitrogen its dense constituent. The methods
adopted in preparing large quantities will be subsequently described.
6. Proof of the Presence of Argon in Air, by means of Atmolysis.
It has already (§ 2) been suggested that if " atmospheric nitrogen "
contains two gases of different densities, it should be possible to obtain direct
evidence of the fact by the method of atmolysis. The present section contains
an account of carefully conducted experiments directed to this end.
1895] ARGON, A NEW CONSTITUENT OF THE ATMOSPHERE. 151
The atmolyser was prepared (after Graham) by combining a number of
" churchwarden " tobacco pipes. At first twelve pipes were used in three
groups, each group including four pipes connected in series. The three
groups were then connected in parallel, and placed in a large glass tube closed
in such a way that a partial vacuum could be maintained in the space outside
the pipes by a water-pump. One end of the combination of pipes was open
to the atmosphere, or rather was connected with the interior of an open bottle
containing sticks of caustic alkali, the object being mainly to dry the air.
The other end of the combination was connected to a bottle aspirator,
initially full of water, and so arranged as to draw about two per cent, of the
air which entered the other end of the pipes. The gas collected was thus a
very small proportion of that which leaked through the pores of the pipes,
and should be relatively rich in the heavier constituents of the atmosphere.
The flow of water from the aspirator could not be maintained very constant,
but the rate of two per cent, was never much exceeded. The necessary four
litres took about sixteen hours to collect.
The air thus obtained was treated exactly as ordinary air had been treated
in determinations of the density of atmospheric nitrogen. Oxygen was re-
moved by red-hot copper followed by cupric oxide, ammonia by sulphuric
acid, carbonic anhydride and moisture by potash and phosphoric anhydride.
The following are the results : —
Globe empty July 10, 14 2'81789
Globe full September 15 (twelve pipes) . . "50286
Weight of gas 2-31503
Ordinary atmospheric nitrogen 2'31016
Difference + '00487
Globe empty September 17 2'81345
Globe full September 18 (twelve pipes) . . '50191
Weight of gas 2-31154
Ordinary atmospheric nitrogen 2'31016
Difference + '00138
Globe empty September 21 2'82320
Globe full September 20 (twelve pipes) . . '51031
Weight of gas 2-31289
Ordinary atmospheric nitrogen 2'31016
Difference . . + '00273
152 ARGON, A NEW CONSTITUENT OF THE ATMOSPHERE. [214
Globe empty September 21, October 30 . . 2*82306
Globe full September 22 (twelve pipes) . . '51140
Weight of gas 2'31166
Ordinary atmospheric nitrogen 2'31016
Difference . + '00150
The mean excess of the four determinations is '00262 gram., or if we omit
the first, which depended upon a vacuum weighing of two months old,
•00187 gram.
The gas from prepared air was thus in every case denser than from
unprepared air, and to an extent much beyond the possible errors of experi-
ment. The excess was, however, less than had been expected, and it was
thought that the arrangement of the pipes could be improved. The final
delivery of gas from each of the groups in parallel being so small in comparison
with the whole streams concerned, it seemed possible that each group was not
contributing its proper share, and even that there might be a flow in the
wrong direction at the delivery end of one or two of them. To meet this
objection, the arrangement in parallel had to be abandoned, and for the
remaining experiments eight pipes were connected in simple series. The
porous surface in operation was thus reduced, but this was partly compensated
for by an improved vacuum. Two experiments were made under the new
conditions : —
Globe empty, October 30, November 5 . . 2'82313
Globe full, November 3 (eight pipes) . . . '50930
Weight of gas 2'31383
Ordinary atmospheric nitrogen 2'31016
Difference + '00367
Globe empty, November 5, 8 2'82355
Globe full, November 6 (eight pipes) . . . '51011
Weight of gas 2'31344
Ordinary atmospheric nitrogen 2'31016
Difference + '00328
The excess being larger than before is doubtless due to the greater
efficiency of the atmolysing apparatus. It should be mentioned that the
above recorded experiments include all that have been tried, and the con-
clusion seems inevitable that " atmospheric nitrogen " is a mixture and not a
simple body.
It was hoped that the concentration of the heavier constituent would be
sufficient to facilitate its preparation in a pure state by the use of prepared
1895] ARGON, A NEW CONSTITUENT OF THE ATMOSPHERE. 153
air in substitution for ordinary air in the oxygen apparatus. The advance of
3£ mg. on the 11 mg., by which atmospheric nitrogen is heavier than chemical
nitrogen, is indeed not to be despised, and the use of prepared air would be
convenient if the diffusion apparatus could be set up on a large scale and
be made thoroughly self-acting.
7. Negative Experiments to prove that Argon is not derived from Nitrogen or
from Chemical Sources.
Although the evidence of the existence of argon in the atmosphere,
derived from the comparison of densities of atmospheric and chemical nitrogen
and from the diffusion experiments (§ 6), appeared overwhelming, we have
thought it undesirable to shrink from any labour that would tend to complete
the verification. With this object in view, an experiment was undertaken
and carried to a conclusion on November 13, in which 3 litres of chemical
nitrogen, prepared from ammonium nitrite, were treated with oxygen in
precisely the manner in which atmospheric nitrogen had been found to yield
a residue of argon. In the course of operations an accident occurred, by which
no gas could have been lost, but of such a nature that from 100 to 200 cub.
centims. of air must have entered the working vessel. The gas remaining at
the close of the large scale operations was worked up as usual with battery
and coil until the spectrum showed only slight traces of the nitrogen lines.
When cold, the residue measured 4 cub. centims. This was transferred, and
after treatment with alkaline pyrogallate to remove oxygen, measured 3*3 cub.
centims. If atmospheric nitrogen had been employed, the final residue should
have been about 30 cub. centims. Of the 3'3 cub. centims. actually left, a
part is accounted for by the accident alluded to, and the result of the
experiment is to show that argon is not formed by sparking a mixture of
oxygen and chemical nitrogen.
In a second experiment of the same kind 5660 cub. centims. of nitrogen
from ammonium nitrite were treated with oxygen in the large apparatus
(Fig. 7,§ 8). The final residue was 3'5 cub. centims.; and, as evidenced by the
spectrum, it consisted mainly of argon.
The source of the residual argon is to be found in the water used for the
manipulation of the large quantities of gas (6 litres of nitrogen and 11 litres
of oxygen) employed. Unfortunately the gases had been collected by allowing
them to bubble up into aspirators charged with ordinary water, and they were
displaced by ordinary water. In order to obtain information with respect to
the contamination that may be acquired in this way, a parallel experiment
was tried with carbonic anhydride. Eleven litres of the gas, prepared from
marble and hydrochloric acid with ordinary precautions for the exclusion of
air, were collected exactly as oxygen was commonly collected. It was then
154 ARGON, A NEW CONSTITUENT OF THE ATMOSPHERE. [214
transferred by displacement with water to a gas pipette charged with a
solution containing 100 grins, of caustic soda. The residue which refused
absorption measured as much as 110 cub. centims. In another experiment
where the water employed had been partially de-aerated, the residue left
amounted to 71 cub. centims., of which 26 cub. centims. were oxygen. The
quantities of dissolved gases thus extracted from water during the collection
of oxygen and nitrogen suffice to explain the residual argon of the negative
experiments.
It may perhaps be objected that the impurity was contained in the
carbonic anhydride itself as it issued from the generating vessel, and was
not derived from the water in the gas-holder ; and indeed there seems to be
a general impression that it is difficult to obtain carbonic anhydride in a state
of purity. To test this question, 18 litres of the gas, made in the same
generator and from the same materials, were passed directly into the absorp-
tion pipette. Under these conditions, the residue was only 6£ cub. centims.,
corresponding to 4 cub. centims. from 11 litres. The quantity of gas employed
was determined by decomposing the resulting sodium carbonate with hydro-
chloric acid, allowance being made for a little carbonic anhydride contained
in the soda as taken from the stock bottle. It will be seen that there is no
difficulty in reducing the impurity to ^^th, even when india-rubber connec-
tions are freely used, and no extraordinary precautions are taken. The large
amount of impurity found in the gas when collected over water must therefore
have been extracted from the water.
A similar set of experiments was carried out with magnesium. The
nitrogen, of which three litres were used, was prepared by the action of
bleaching-powder on ammonium chloride. It was circulated in the usual
apparatus over red-hot magnesium, until its volume had been reduced to
about 100 cub. centims. An equal volume of hydrogen was then added, owing
to the impossibility of circulating a vacuum. The circulation then proceeded
until all absorption had apparently stopped. The remaining gas was then
passed over red-hot copper oxide into the Sprengel's pump, and collected. As
it appeared still to contain hydrogen, which had escaped oxidation, owing to
its great rarefaction, it was passed over copper oxide for a second and a third
time. As there was still a residue, measuring 12'5 cub. centims., the gas was
left in contact with red-hot magnesium for several hours, and then pumped
out; its volume was then 4'5 cub. centims. Absorption was, however, still
proceeding, when the experiment terminated, for at a low pressure, the rate is
exceedingly slow. This gas, after being sparked with oxygen contracted to
3'0 cub. centims., and on examination was seen to consist mainly of argon.
The amount of residue obtainable from three litres of atmospheric nitrogen
should have amounted to a large multiple of this quantity.
1895] ARGON, A NEW CONSTITUENT OF THE ATMOSPHERE. 155
In another experiment, 15 litres of nitrogen prepared from a mixture of
ammonium chloride and sodium nitrite by warming in a flask (some nitrogen
having first been drawn off by a vacuum-pump, in order to expel all air from
the flask and from the contained liquid) were collected over water in a large
gas-holder. The nitrogen was not bubbled through the water, but was
admitted from above, while the water escaped below. This nitrogen was
absorbed by red-hot magnesium, contained in tubes heated in a combustion-
furnace. The unabsorbed gas was circulated over red-hot magnesium in a
special small apparatus, by which its volume was reduced to 15 cub. centims.
As it was impracticable further to reduce the volume by means of magnesium,
the residual 15 cub. centims. were transferred to a tube, mixed with oxygen,
and submitted to sparking over caustic soda. The residue after absorption of
oxygen, which undoubtedly consisted of pure argon, amounted to 3'5 cub.
centims. This is one-fortieth of the quantity which would have been obtained
from atmospheric nitrogen, and its presence can be accounted for, we venture
to think, first from the water in the gas-holder, which had not been freed from
dissolved gas by boiling in vacuo (it has already been shown that a consider-
able gain may ensue from this source), and second, from leakage of air
which accidentally took place, owing to the breaking of a tube. The leakage
may have amounted to 200 cub. centims., but it could not be accurately
ascertained. Quantitative negative experiments of this nature are exceedingly
difficult, and require a long time to carry them to a successful conclusion.
8. Reparation of Argon on a Large Scale.
To separate nitrogen from "atmospheric nitrogen" on a large scale, by
help of magnesium, several devices were tried. It is not necessary to describe
them all in detail. Suffice it to say that an attempt was made to cause a
store of " atmospheric nitrogen " to circulate by means of a fan, driven by a
water-motor. The difficulty encountered here was leakage at the bearing of
the fan, and the introduced air produced a cake which blocked the tube on
coming into contact with the magnesium. It might have been possible to
remove oxygen by metallic copper; but instead of thus complicating the
apparatus, a water-injector was made use of to induce circulation. Here also
it is unnecessary to enter into details. For, though the plan worked well,
and although about 120 litres of " atmospheric nitrogen" were absorbed, the
yield of argon was not large, about GOO cub. centims. having been collected.
This loss was subsequently discovered to be due partially, at least, to the rela-
tively high solubility of argon in water. In order to propel the gas over
magnesium, through a long combustion-tube packed with turnings, a consider-
able water-pressure, involving a large flow of water, was necessary. The gas
was brought into intimate contact with this water, and presuming that several
thousand litres of water ran through the injector, it is obvious that a not
156 ARGON, A NEW CONSTITUENT OF THE ATMOSPHERE. [214
inconsiderable amount of argon must have been dissolved. Its proportion
was increasing at each circulation, and consequently its partial pressure also
increased. Hence, towards the end of the operation, at least, there is every
reason to believe that a serious loss had occurred.
It was next attempted to pass " atmospheric nitrogen " from a gas-holder
first through a combustion tube of the usual length packed with metallic
copper reduced from the oxide; then through a small U-tube containing a
little water, which was intended as an index of the rate of flow; the gas was
then dried by passage through tubes filled with soda-lime and phosphoric
anhydride ; and it next passed through a long iron tube (gas-pipe) packed
with magnesium turnings, and heated to bright redness in a second com-
bustion-furnace.
After the iron tube followed a second small U-tube containing water,
intended to indicate the rate at which the argon escaped into a small gas-
holder placed to receive it. The nitrogen was absorbed rapidly, and argon
entered the small gas-holder. But there was reason to suspect that the iron
tube is permeable by argon at a red heat. The first tube-full allowed very
little argon to pass. After it had been removed and replaced by a second, the
same thing was noticed. The first tube was difficult to clean ; the nitride of
magnesium forms a cake on the interior of the tube, and it was very difficult
to remove it ; moreover this rendered the filling of the tube very troublesome,
inasmuch as its interior was so rough that the magnesium turnings could only
with difficulty be forced down. However, the permeability to argon, if such
be the case, appeared to have decreased. The iron tube was coated internally
with a skin of magnesium nitride, which appeared to diminish its permeability
to argon. After all the magnesium in the tube had been converted into
nitride (and this was easily known, because a bright glow proceeded gradually
from one end of the tube to the other) the argon remaining in the iron tube
was " washed " out by a current of nitrogen ; so that, after a number of opera-
tions, the small gas-holder contained a mixture of argon with a considerable
quantity of nitrogen.
On the whole, the use of iron tubes is not to be recommended, owing to
the difficulty in cleaning them, and the possible loss through their permeability
to argon. There is no such risk of loss with glass tubes, but each operation
requires a new tube, and the cost of the glass is considerable if much nitrogen
is to be absorbed. Tubes of porcelain were tried; but the glaze in the
interior is destroyed by the action of the red-hot magnesium, and the tubes
crack on cooling.
By these processes 157 litres of " atmospheric nitrogen " were reduced in
volume to about 245 litres in all of a mixture of nitrogen and argon. This
1895] ARGON, A NEW CONSTITUENT OF THE ATMOSPHERE.
157
mixture was afterwards circulated over red-hot magnesium, in order to remove
the last portion of nitrogen.
Fig. 5.
As the apparatus employed for this purpose proved very convenient, a full
description of its construction is here given. A diagram is shown in Fig. 5,
which sufficiently explains the arrangement of the apparatus. A is the
circulator. It consists of a sort of Sprengel's pump (a) to which a supply of
mercury is admitted from a small reservoir (6). This mercury is delivered into
a gas-separator (c), and the mercury overflows into the reservoir (d). When
its level rises, so that it blocks the tube (/), it ascends in pellets or pistons
into (e), a reservoir which is connected through (g) with a water-pump. The
mercury falls into (6), and again passes down the Sprengel tube (a). No
attention is, therefore, required, for the apparatus works quite automatically.
This form of apparatus was employed several years ago by Dr Collie.
The gas is drawn from the gas-holder B, and passes through a tube C,
which is heated to redness by a long-flame burner, and which contains in one
half metallic copper, and in the other half copper oxide. This precaution is
taken in order to remove any oxygen which may possibly be present, and also
any hydrogen or hydrocarbon. In practice, it was never found that the
copper became oxidised, or the oxide reduced. It is, however, useful to guard
against any possible contamination. The gas next traversed a drying-tube D,
the anterior portion containing ignited soda-lime, and the posterior portion
phosphoric anhydride. From this it passed a reservoir, Z)', from which it
could be transferred, when all absorption had ceased, into the small gas-holder.
158 ARGON, A NEW CONSTITUENT OF THE ATMOSPHERE. [214
It then passed through E, a piece of combustion-tube, drawn out at both ends,
filled with magnesium turnings, and heated by a long-flame burner to redness.
Passing through a small bulb, provided with electrodes, it again entered the
fall-tube.
After the magnesium tube E had done its work, the stop-cocks were all
closed, and the gas was turned down, so that the burners might cool. The
mixture of argon and nitrogen remaining in the system of tubes was pumped
out by a Sprengel's pump through F, collected in a large test-tube, and
reintroduced into the gas-holder B through the side-tube G, which requires
no description. The magnesium tube was then replaced by a fresh one ; the
system of tubes was exhausted of air ; argon and nitrogen were admitted from
the gas-holder B ; the copper-oxide tube and the magnesium tube were again
heated ; and the operation was repeated until absorption ceased. It was easy
to decide when this point had been reached, by making use of the graduated
cylinder H, from which water entered the gas-holder B. It was found
advisable to keep all the water employed in these operations, for it had become
saturated with argon. If gas was withdrawn from the gas-holder, its place
was taken by this saturated water.
The absorption of nitrogen proceeds very slowly towards the end of the
operation, and the diminution in volume of the gas is not greater than 4 or 5
cub. centims. per hour. It is, therefore, somewhat difficult to judge of the
end-point, as will be seen when experiments on the density of this gas are
described. The magnesium tube, towards the end of the operations, was
made so hot that the metal was melted in the lower part of the tube, and
sublimed in the upper part. The argon and residual nitrogen had, therefore,
been thoroughly mixed with gaseous magnesium during its passage through
the tube E.
To avoid possible contamination with air in the Sprengel's pump, the last
portion of gas collected from the system of tubes was not re-admitted to the
gas-holder B, but was separately stored.
The crude argon was collected in two operations. First, the quantity
made by absorption by magnesium in glass tubes with the water-pump
circulator was purified. Later, after a second supply had been prepared by
absorption in iron tubes, the mixture of argon and nitrogen was united with
the first quantity and circulated by means of the mercury circulator, in the
gas-holder J5. Attention will be drawn to the particular sample of gas
employed in describing further experiments made with the argon.
By means of magnesium, about 7 litres of nitrogen can be absorbed in an
hour. The changing of the tubes of magnesium, however, takes some time ;
consequently, the largest amount absorbed in one day was nearly 30 litres.
1895] ARGON, A NEW CONSTITUENT OF THE ATMOSPHERE. 159
At a later date a quantitative experiment was carried out on a large scale,
the amount of argon from 100 litres of " atmospheric " nitrogen, measured at
20°, having been absorbed by magnesium, and the resulting argon measured
at 12°. During the process of absorbing nitrogen in the combustion-furnace,
however, one tube cracked, and it is estimated that about 4 litres of nitrogen
escaped before the crack was noticed. With this deduction, and assuming
that the nitrogen had been measured at 12°, 93'4 litres of atmospheric
nitrogen were taken. The magnesium required for absorption weighed
409 grms. The amount required by theory should have been 285 grms.; but
it must be remembered that in many cases the magnesium was by no means
wholly converted into nitride. The first operation yielded about 3 litres of a
mixture of nitrogen and argon, which was purified in the circulating apparatus.
The total residue, after absorption of the nitrogen, amounted to 921 cub.
centims. The yield is therefore 0'986 per cent.
At first no doubt the nitrogen gains a little argon from the water over
which it stands. But, later, when the argon forms the greater portion of the
gaseous mixture, its solubility in water must materially decrease its volume.
It is difficult to estimate the loss from this cause. The gas-holder, from
which the final circulation took place, held three litres of water. Taking the
solubility of argon as 4 per cent., this would mean a loss of about 120 cub.
centims. If this is not an over-estimate, the yield of argon would be
increased to 1040 cub. centims., or I'll per cent. The truth probably lies
between these two estimates.
It may be concluded, with probability, that the argon forms approximately
1 per cent, of the " atmospheric " nitrogen.
The principal objection to the oxygen method of isolating argon, as
hitherto described, is the extreme slowness of the operation. An absorption
of 30 cub. centims. of mixed gas means the removal of but 12 cub. centims. of
nitrogen. At this rate 8 hours are required for the isolation of 1 cub. centim.
of argon, supposed to be present in the proportion of 1 per cent.
In extending the scale of operations we had the great advantage of the
advice of Mr Crookes, who a short time ago called attention to the flame
rising from platinum terminals, which convey a high tension alternating
electric discharge, and pointed out its dependence upon combustion of the
nitrogen and oxygen of the air*. Mr Crookes was kind enough to arrange an
impromptu demonstration at his own house with a small alternating current
plant, in which it appeared that the absorption of mixed gas was at the rate
of 500 cub. centims. per hour, or nearly 20 times as fast as with the battery.
* Chemical News, Vol. LXV. p. 301, 1892.
160 ARGON, A NEW CONSTITUENT OF THE ATMOSPHERE. [214
The arrangement is similar to that first described by Spottiswoode*. The
primary of a Ruhmkorff coil is connected directly with the alternator, no
break or condenser being required ; so that, in fact, the coil acts simply
as a high potential transformer. When the arc is established the platinum
terminals may be separated much beyond the initial striking distance.
The plant with which the large scale operations have been made consists
of a De Meritens alternator, kindly lent by Professor J. J. Thomson, and a gas
engine. As transformer, one of Swinburne's hedgehog pattern has been
employed with success, but the ratio of transformation (24 : 1) is scarcely
sufficient. A higher potential, although, perhaps, not more efficient, is more
convenient. The striking distance is greater, and the arc is not so liable to
go out. Accordingly most of the work to be described has been performed
with transformers of the Ruhmkorff type.
The apparatus has been varied greatly, and it cannot be regarded as
having even yet assumed a final form. But it will give a sufficient idea of
the method if we describe an experiment in which a tolerably good account
was kept of the air and oxygen employed. The working vessel was a glass
flask, A (Fig. 6), of about 1500 cub. centims. capacity, and stood, neck down-
wards, over a large jar of alkali, B. As in the small scale experiments, the
leading-in wires were insulated by glass tubes, DD, suitably bent and carried
through the liquid up the neck. For the greater part of the length iron wires
were employed, but the internal extremities, EE, were of platinum, doubled
upon itself at the terminals from which the discharge escaped. The glass
protecting tubes must be carried up for some distance above the internal level
of the liquid, but it is desirable that the arc itself should not be much raised
above that level. A general idea of the disposition of the electrodes will be
obtained from Fig. 6. To ensure gas tightness the bends were occupied by
mercury. A tube, C, for the supply or withdrawal of gas was carried in the
same way through the neck.
The Ruhmkorff employed in this operation was one of medium size.
When the mixture was rightly proportioned and the arc of full length, the
rate of absorption was about 700 cub. centims. per hour. A good deal of time
is lost in starting, for, especially when there is soda on the platinums, the arc
is liable to go out if lengthened prematurely. After seven days the total
quantity of air let in amounted to 7925 cub. centims., and of oxygen (prepared
from chlorate of potash) 9137 cub. centims. On the eighth and ninth days
oxygen alone was added, of which about 500 cub. centims. was consumed,
while there remained about 700 cub. centims. in the flask. Hence the pro-
portion in which the air and oxygen combined was as 70 : 96. On the eighth
day there was about three hours' work, and the absorption slackened off to
* " A Mode of Exciting an Induction-coil," Phil. Mag. Vol. vm. p. 390, 1879.
1895]
ARGON, A NEW CONSTITUENT OF THE ATMOSPHERE.
161
about one quarter of the previous rate. On the ninth day (September 8) the
rate fell off still more, and after three hours' work became very slow. The
progress towards removal of nitrogen was examined from time to time with
the spectroscope, the points being approximated and connected with a small
Fig. 6.
Leyden jar. At this stage the yellow nitrogen line was faint, but plainly
visible. After about four hours' more work, the yellow line had disappeared,
and for two hours there had been no visible contraction. It will be seen that
the removal of the last part of the nitrogen was very slow, mainly on account
of the large excess of oxygen present.
11
162 ARGON, A NEW CONSTITUENT OF THE ATMOSPHERE. [214
The final treatment of the residual 700 cub. centims. of gas was on the
model of the small scale operations already described (§ 4). By means of a
pipette the gas was gradually transferred to a large test-tube standing over
alkali. Under the influence of sparks (from battery and coil) passing all the
while, the superfluous oxygen was consumed with hydrogen fed in slowly
from a voltameter. If the nitrogen had been completely removed, and if
there were no unknown ingredient in the atmosphere, the volume under this
treatment should have diminished without limit. But the contraction stopped
at a volume of 65 cub. centims., and the volume was taken backwards and
forwards through this as a minimum by alternate treatment with oxygen and
hydrogen added in small quantities, with prolonged intervals of sparking.
Whether the oxygen or the hydrogen were in excess could be determined at
any moment by a glance at the spectrum. At the minimum volume the gas
was certainly not hydrogen or oxygen. Was it nitrogen ? On this point the
testimony of the spectroscope was equally decisive. No trace of the yellow
nitrogen line could be seen even with a wide slit and under the most favour-
able conditions.
When the gas stood for some days over water, the nitrogen line again
asserted itself, and many hours of sparking with a little oxygen were required
again to get rid of it. As it was important to know what proportions of
nitrogen could be made visible in this way, a little air was added to gas that
had been sparked for some time subsequently to the disappearance of nitrogen
in its spectrum. It was found that about 1^ per cent, was clearly, and about
3 per cent, was conspicuously, visible. About the same numbers apply to the
visibility of nitrogen in oxygen when sparked under these conditions, that is,
at atmospheric pressure, and with a jar in connection with the secondary
terminals.
When we attempt to increase the rate of absorption by the use of a
more powerful electric arc, further experimental difficulties present them-
selves. In the arrangement already described, giving an absorption of 700
cub. centims. per hour, the upper part of the flask becomes very hot. With a
more powerful arc the heat rises to such a point that the flask is filled with
steam and the operation comes to a standstill.
It is necessary to keep the vessel cool by either the external or internal
application of liquid to the upper surface upon which the hot gases from the
arc impinge. One way of effecting this is to cause a small fountain of alkali
to impinge on the top of the flask, so as to wash the whole of the upper
surface. This plan is very effective, but it is open to the objection that a break-
down would be disastrous, and it would involve special arrangements to avoid
losing the argon by solution in the large quantity of alkali required. It is
simpler in many respects to keep the vessel cool by immersing it in a large
body of water, and the inverted flask arrangement (Fig. 6) has been applied in
1895]
ARGON, A NEW CONSTITUENT OF THE ATMOSPHERE.
163
this manner. But, on the whole, it appears to be preferable to limit the
application of the cooling water to the upper part of the external surface,
building up for this purpose a suitable wall of sheet lead cemented round the
glass. The most convenient apparatus for large-scale operations that has
hitherto been tried is shown in the accompanying figure (Fig. 7).
Fig. 7.
Scale
The vessel A is a large globe of about 6 litres capacity, intended for
demonstrating the combustion of phosphorus in oxygen gas, and stands in an
inclined position. It is about half filled with a solution of caustic soda. The
neck is fitted with a rubber stopper, B, provided with four perforations. Two
of these are fitted with tubes, (7, D, suitable for the supply or withdrawal of
gas or liquid. The other two allow the passage of the stout glass tubes, E, F,
which contain the electrodes. For greater security against leakage, the
interior of these tubes is charged with water, held in place by small corks,
and the outer ends are cemented up. The electrodes are formed of stout iron
wires terminated by thick platinums, G, H, triply folded together, and welded
at the ends. The lead walls required to enclose the cooling water are partially
shown at I. For greater security the india-rubber cork is also drowned in
water, held in place with the aid of sheet-lead. The lower part of the globe
is occupied by about 3 litres of a 5 per cent, solution of caustic soda, the
solution rising to within about half-an-inch of the platinum terminals. With
this apparatus an absorption of 3 litres of mixed gas per hour can be
attained, — about 3000 times the rate at which Cavendish could work.
11—2
164 ARGON, A NEW CONSTITUENT OF THE ATMOSPHERE. [214
When it is desired to stop operations, the feed of air (or of chemical
nitrogen in blank experiments) is cut off, oxygen alone being supplied as long
as any visible absorption occurs. Thus at the close the gas space is occupied
by argon and oxygen with such nitrogen as cannot readily be taken up in a
condition of so great dilution. The oxygen, being too much for convenient
treatment with hydrogen, was usually absorbed with copper and ammonia,
and the residual gas was then worked over again as already described in
an apparatus constructed upon a smaller scale.
It is worthy of notice that with the removal of the nitrogen, the arc-
discharge from the dynamo changes greatly in appearance, bridging over more
directly and in a narrower band from one platinum to the other, and assuming
a beautiful sky-blue colour, instead of the greenish hue apparent so long as
oxidation of nitrogen is in progress.
In all the large-scale experiments, an attempt was made to keep a reckon-
ing of the air and oxygen employed, in the hope of obtaining data as to the
proportional volume of argon in air, but various accidents too often interfered.
In one successful experiment (January, 1895), specially undertaken for the
sake of measurement, the total air employed was 9250 cub. centims., and the
oxygen consumed, manipulated with the aid of partially de-aerated water,
amounted to 10,820 cub. centims. The oxygen contained in the air would be
1942 cub. centims. ; so that the quantities of " atmospheric nitrogen " and of
total oxygen which enter into combination would be 7308 cub. centims., and
12,762 cub. centims. respectively. This corresponds to N + T75 0 — the oxygen
being decidedly in excess of the proportion required to form nitrous acid —
2HN02, or H2O + N2 + 3 O. The argon ultimately found on absorption of the
excess of oxygen was 75'0 cub. centims., reduced to conditions similar to those
under which the air was measured, or a little more than 1 per cent, of the
"atmospheric nitrogen" used. It is probable, however, that some of the
argon was lost by solution during the protracted operations required in order
to get quit of the last traces of nitrogen.
[In recent operations at the Royal Institution, where a public supply of
alternating current at 100 volts is available, the scale of the apparatus has
been still further increased.
The capacity of the working vessel is 20 litres, of which about one half is
occupied by a strong solution of caustic soda. The platinum terminals are
very massive, and the flame rising from them is prevented from impinging
directly upon the glass by a plate of platinum held over it and supported by
a wire which passes through the rubber cork. In the electrical arrangements
we have had the advantage of Mr Swinburne's advice. The transformers are
two of the " hedgehog " pattern, the thick wires being connected in parallel
and the thin wires in series. In order to control the current taken when the
1895] ARGON, A NEW CONSTITUENT OF THE ATMOSPHERE. 165
arc is short or the platinums actually in contact, a choking-coil, provided with
a movable core of fine iron wires, is inserted in the thick wire circuit. In
normal working the current taken from the mains is about 22 amperes, so
that some 2£ h. p. is consumed. At the same time the actual voltage at the
platinum terminals is 1500. When the discharge ceases, the voltage at the
platinum rises to 3000*, which is the force actually available for re-starting
the discharge if momentarily stopped.
With this discharge, the rate of absorption of mixed gases is about 7 litres
per hour. When the argon has accumulated to a considerable extent, the
rate falls off, and after several days' work, about 6 litres per hour becomes the
maximum. In commencing operations it is advisable to introduce, first, the
oxygen necessary to combine with the already included air, after which the
feed of mixed gases should consist of about 11 parts of oxygen to 9 parts of
air. The mixed gases may be contained in a large gas-holder, and then, the
feed being automatic, very little attention is required. When it is desired to
determine the rate of absorption, auxiliary gas-holders of glass, graduated into
litres, are called into play. If the rate is unsatisfactory, a determination may
be made of the proportion of oxygen in the working vessel, and the necessary
gas, air, or oxygen, as the case may be, introduced directly.
In re-starting the arc after a period of intermission, it is desirable to cut
off the connection with the principal gas-holder. The gas (about two litres in
amount) ejected from the working vessel by the expansion is then retained in
the auxiliary holder, and no argon finds its way further back. The connection
between the working vessel and the auxiliary holder should be made without
india-rubber, which is liable to be attacked by the ozonized gases.
The apparatus has been kept in operation lor fourteen hours continuously,
and there should be no difficulty in working day and night. An electric
signal could easily be arranged to give notice of the extinction of the arc,
which sometimes occurs unexpectedly; or an automatic device for re-striking
the arc could be contrived. — April, 1895.]
9. Density of Argon prepared by means of Oxygen.
A first estimate of the density of argon prepared by the oxygen method
was founded upon the data recorded already respecting the volume present in
air, on the assumption that the accurately known densities of " atmospheric "
and of chemical nitrogen differ on account of the presence of argon in the
former, and that during the treatment with oxygen nothing is oxidised except
nitrogen. Thus, if
* A still higher voltage on open circuit would be preferable.
166 ARGON, A NEW CONSTITUENT OF THE ATMOSPHERE. [214
D = density of chemical nitrogen,
jy= „ atmospheric nitrogen,
d = „ argon,
a = proportional volume of argon in atmospheric nitrogen,
the law of mixtures gives
d = D+(D'-D)/a.
In this formula D' — D and a are both small, but they are known with fair
accuracy. From the data already given for the experiment of September 8th
65
0-79 x 7925
= 0-0104;
whence, if on an arbitrary scale of reckoning D = 2*2990, IX = 2'3102, we find
d = 3-378. Thus if N2 be 14, or O2 be 16, the density of argon is 20'6.
Again, from the January experiment,
whence, if N = 14, the density of argon is 20'6, as before. There can be little
doubt, however, that these numbers are too high, the true value of a being
greater than is supposed in the above calculations.
A direct determination by weighing is desirable, but hitherto it has not
been feasible to collect by this means sufficient to fill the large globe (§1)
employed for other gases. A mixture of about 400 cub. centims. of argon with
pure oxygen, however, gave the weight 2'7315, 0-1045 in excess of the weight
of oxygen, viz., 2'6270. Thus, if a. be the ratio of the volume of argon to the
whole volume, the number for argon will be
2-6270 + 0-1045/a.
The value of a, being involved only in the excess of weight above that of
oxygen, does not require to be known very accurately. Sufficiently concordant
analyses by two methods gave a = 0'1845 ; whence, for the weight of the gas
we get 3*193 ; so that if O = 16, the density of the gas would be 19'45. An
allowance for residual nitrogen, still visible in the gas before admixture of
oxygen, raises this number to 197, which may be taken as the density of pure
argon resulting from this determination*.
* [The proportion of nitrogen (4 or 5 per cent, of the volume) was estimated from the
appearance of the nitrogen lines in the spectrum, these being somewhat more easily visible than
when 3 per cent, of nitrogen was introduced into pure argon (§ 8). — April, 1895.]
1895] ARGON, A NEW CONSTITUENT OF THE ATMOSPHERE. 167
10. Density of Argon prepared by means of Magnesium,*.
It has already been stated that the density of the residual gas from the
first and preliminary attempt to separate oxygen and nitrogen from air by
means of magnesium was 19'086, and allowing for contraction on sparking
with oxygen the density is calculable as 20'01. The following determinations
of density were also made : —
(a) After absorption in glass tubes, the water circulator having been used,
and subsequent circulation by means of mercury circulator until rate of con-
traction had become slow, 162-843 cub. centims., measured at 757*7 millims.
(corr.) pressure, and 16-81° C., weighed 0'2683 grm. Hence,
Weight of 1 litre at 0° and 760 millims 1-7543 grms.
Density compared with hydrogen (O = 16) . . . 19*63 „
This gas was again circulated over red-hot magnesium for two days.
Before circulation it contained nitrogen as was evident from its spectrum;
after circulating, nitrogen appeared to be absent, and absorption had com-
pletely stopped. The density was again determined.
(6) 162-843 cub. centims., measured at 745'4 millims. (corr.) pressure, and
17-25° C., weighed 0'2735 grm. Hence,
Weight of 1 litre at 0° and 760 millims 1-8206 grms.
Density compared with hydrogen (O = 16) . . . 20*38 „
Several portions of this gas, having been withdrawn for various purposes,
were somewhat contaminated with air, owing to leakage, passage through the
pump, &c. All these portions were united in the gas-holder with the main
stock, and circulated for eight hours, during the last three of which no
contraction occurred. The gas removed from the system of tubes by the
mercury-pump was not restored to the gas-holder, but kept separate.
(c) 162-843 cub. centims., measured at 758'1 millims. (corr.) pressure, and
17-09° C., weighed 0'27705 grm. Hence,
Weight of 1 litre at 0° and 760 millims 1-8124 grms.
Density compared with hydrogen (O = 16) . . . 20'28
The contents of the gas-holder were subsequently increased by a mixture
of nitrogen and argon from 37 litres of atmospheric nitrogen, and after
circulating, density was determined. The absorption was however not com-
plete.
(d) 162'843 cub. centims., measured at 767'6 millims. (corr.) pressure, and
16-31° C., weighed 0-2703 grm. Hence,
* See Addendum, p. 184.
168 ARGON, A NEW CONSTITUENT OF THE ATMOSPHERE. [214
Weight of 1 litre at 0° and 760 millims 1-742 grms.
Density compared with hydrogen (0 = 16) . . . 19'49 „
The gas was further circulated, until all absorption had ceased. This took
about six hours. Density was again determined.
(e) 162-843 cub. centims., measured at 767'7 millims. (corr.) pressure, and
15-00° C., weighed 0'2773 grm. Hence,
Weight of 1 litre at 0° and 760 millims 1-7784 grms.
Density compared with hydrogen (O = 16) . . . 19*90 „
(/) A second determination was carried out, without further circulation.
162*843 cub. centims., measured at 769*0 millims. (corr.) pressure, and
16-00° C., weighed 0'2757 grm. Hence,
Weight of 1 litre at 0° and 760 millims 1*7713 grms.
Density compared with hydrogen (O = 16) . . . 19*82 „
(g) After various experiments had been made with the same sample of
gas, it was again circulated until all absorption ceased. A vacuum-tube was
filled with it, and showed no trace of nitrogen.
The density was again determined : —
162*843 cub. centims., measured at 750 millims. (corr.) pressure, and at
15*62° C., weighed 0*26915 grm.
Weight of 1 litre at 0° and 760 millims 1*7707 grms.
Density compared with hydrogen (O = 16) . . . 19*82 „
These comprise all the determinations of density made. It should be
stated that there was some uncertainty discovered later about the weight of
the vacuous globe in (6) and (c). Rejecting these weighings, the mean of (e),
(/), and (g) is 19*88. The density may be taken as 19*9, with approximate
accuracy.
It is better to leave these results without comment at this point, and to
return to them later.
11. Spectrum of Argon.
Vacuum tubes were filled with argon prepared by means of magnesium at
various stages in this work, and an examination of these tubes has been
undertaken by Mr Crookes, to whom we wish to express our cordial thanks
for his kindness in affording us helpful information with regard to its
spectrum. The first tube was filled with the early preparation of density 19'09,
which obviously contained some nitrogen. A photograph of the spectrum was
taken, and compared with a photograph of the spectrum of nitrogen, and it
was at once evident that a spectrum different from that of nitrogen had
been registered.
1895] ARGON, A NEW CONSTITUENT OF THE ATMOSPHERE. 169
Since that time many other samples have been examined.
The spectrum of argon, seen in a vacuum tube of about 3 millims. pressure,
consists of a great number of lines, distributed over almost the whole visible
field. Two lines are specially characteristic; they are less refrangible than
the red lines of hydrogen or lithium, and serve well to identify the gas when
examined in this way. Mr Crookes, who gives a full account of the spectrum
in a separate communication, has kindly furnished us with the accurate
wave-lengths of these lines as well as of some others next to be described ;
they are respectively 696'56 and 705*64 x lO"8 millim.
Besides these red lines, a bright yellow line, more refrangible than the
sodium line, occurs at 603*84. A group of five bright green lines occurs next,
besides a number of less intensity. Of this group of five, the second, which is
perhaps the most brilliant, has the wave-length 561*00. There is next a blue,
or blue- violet, line of wave-length 470'2 and last, in the less easily visible part
of the spectrum, there are five strong violet lines, of which the fourth, which
is the most brilliant, has the wave-length 420-0.
Unfortunately, the red lines, which are not to be mistaken for those of
any other substance, are only to be seen at atmospheric pressure when a very
powerful jar-discharge is passed through argon. The spectrum, seen under
these conditions, has been examined by Professor Schuster. The most
characteristic lines are perhaps those in the neighbourhood of F, and are very
easily seen if there be not too much nitrogen, in spite of the presence of some
oxygen and water- vapour. The approximate wave-lengths are : —
487-91 .... Strong.
(486-07) . . . . F.
484*71 .... Not quite so strong.
480-52 .... Strong.
476-50]
473-53 > . . . . Fairly strong characteristic triplet.
472-56 j
It is necessary to anticipate Mr Crookes's communication, and to state
that when the current is passed from the induction-coil in one direction,
that end of the capillary tube next the positive pole appears of a redder, and
that next the negative of a bluer hue. There are, in effect, two spectra,
which Mr Crookes has succeeded in separating to a considerable extent.
Mr E. C. C. Baly *, who has noticed a similar phenomenon, attributes it to
the presence of two gases. The conclusion would follow that what we have
termed " argon " is in reality a mixture of two gases which have as yet not
been separated. This conclusion, if true, is of great importance, and experi-
* Proc. Phys. Soc. 1893, p. 147. He says: "When an electric current is passed through a
mixture of two gases, one is separated from the other, and appears in the negative glow."
170 ARGON, A NEW CONSTITUENT OF THE ATMOSPHERE. [214
merits are now in progress to test it by the use of other physical methods.
The full bearing of this possibility will appear later.
A comparison was made of the spectrum seen in a vacuum tube with the
spectrum in a "plenum" tube, i.e., one filled at atmospheric pressure. Both
spectra were thrown into a field at the same time. It was evident that they
were identical, although the relative strengths of the lines were not always
the same. The seventeen most striking lines were absolutely coincident.
The presence of a small quantity of nitrogen interferes greatly with the
argon spectrum. But we have found that in a tube with platinum electrodes,
after the discharge has been passed for four hours, the spectrum of nitrogen
disappears, and the argon spectrum manifests itself in full purity. A specially
constructed tube, with magnesium electrodes, which we hoped would yield
good results, removed all traces of nitrogen it is true, but hydrogen was
evolved from the magnesium, and showed its characteristic lines very
strongly. However, these are easily identified. The gas evolved on heating
magnesium in vacuo, as proved by a separate experiment, consists entirely of
hydrogen.
Mr Crookes has proved the identity of the chief lines of the spectrum of
gas separated from air-nitrogen by aid of magnesium with that remaining
after sparking air-nitrogen with oxygen, in presence of caustic soda solution.
Professor Schuster has also found the principal lines identical in the
spectra of the two gases, when taken from the jar-discharge at atmospheric
pressure.
12. Solubility of Argon in Water.
The tendency of the gas to disappear when manipulated over water in
small quantities having suggested that it might be more than usually soluble
in that liquid, special experiments were tried to determine the degree of
solubility.
The most satisfactory measures relating to the gas isolated by means of
oxygen were those of September 28. The sample contained a trace of
oxygen, and (as judged by the spectrum) a residue of about 2 per cent, of
nitrogen. The procedure and the calculations followed pretty closely the
course marked out by Bunsen*, and it is scarcely necessary to record the
details. The quantity of gas operated upon was about 4 cub. centims., of
which about 1£ cub. centims. were absorbed. The final result for the
solubility was 3'94 per 100 of water at 12° C., about 2£ times that of nitrogen.
Similar results have been obtained with argon prepared by means of mag-
nesium. At a temperature of 13'9°, 131 arbitrary measures of water absorbed
* Gasometry, p. 141.
1895]
ARGON, A NEW CONSTITUENT OF THE ATMOSPHERE.
171
5*3 of argon. This corresponds to a solubility in distilled water, previously
freed from dissolved gas by boiling in vacuo for a quarter of an hour, and
admitted to the tube containing argon without contact with air, of 4'05 cub.
centims. of argon per 100 of water.
The fact that the gas is more soluble than nitrogen would lead us to
expect it in increased proportion in the dissolved gases of rain water.
Experiment has confirmed this anticipation. Some difficulty was at first
experienced in collecting a sufficiency for the weighings in the large globe of
nearly 2 litres capacity. Attempts at extraction by means of a Topler pump
without heat were not very successful. It was necessary to operate upon
large quantities of water, and then the pressure of the liquid itself acted as an
obstacle to the liberation of gas from all except the upper layers. Tapping
the vessel with a stick of wood promotes the liberation of gas in a remarkable
manner, but to make this method effective, some means of circulating the
water would have to be introduced.
Fig. 8.
The extraction of the gases by heat proved to be more manageable.
Although a large quantity of water has to be brought to or near 100° C., a
prolonged boiling is not necessary, as it is not a question of collecting the
whole of the gas contained in the water. The apparatus employed, which
worked very well after a little experience, will be understood from the
accompanying figure. The boiler A was constructed from an old oil-can, and
was heated by an ordinary ring Bunsen burner. For the supply and removal
of water, two co-axial tubes of thin brass, and more than four feet in length,
172 ARGON, A NEW CONSTITUENT OF THE ATMOSPHERE. [214
were applied upon the regenerative principle. The outgoing water flowed in
the inner tube BG, continued from C to D by a prolongation of composition
tubing. The inflowing water from a rain-water cistern was delivered into a
glass tube at E, and passed through a brass connecting tube FG into the
narrow annular space between the two principal tubes GH. The neck of the
can was fitted with an india-rubber cork and delivery-tube, by means of which
the gases were collected in the ordinary way. Any carbonic anhydride was
removed by alkali before passage into the glass aspirating bottles used as
gas-holders.
The convenient working of this apparatus depends very much upon the
maintenance of a suitable relation between the heat and the supply of water.
It is desirable that the water in the can should actually boil, but without a
great development of steam ; otherwise not only is there a waste of heat, and
thus a smaller yield of gas, but the inverted flask used for the collection of the
gas becomes inconveniently hot and charged with steam. It was found
desirable to guard against this by the application of a slow stream of water to
the external surface of the flask. When the supply of water is once adjusted,
nearly half a litre of gas per hour can be collected with very little attention.
The gas, of which about four litres are required for each operation, was
treated with red-hot copper, cupric oxide, sulphuric acid, potash, and finally
phosphoric anhydride, exactly as atmospheric nitrogen was treated in former
weighings. The weights found, corresponding to those recorded in § 1, were
on two occasions 2'3221 and 2'3227, showing an excess of 24 milligrms. above
the weight of true nitrogen. Since the corresponding excess for atmospheric
nitrogen is 11 milligrms., we conclude that the water- nitrogen is relatively
twice as rich in argon.
Unless some still better process can be found, it may be desirable to
collect the gases ejected from boilers, or from large supply pipes which run
over an elevation, with a view to the preparation of argon upon a large scale.
The above experiments relate to rain water. As regards spring water, it
is known that many thermal springs emit considerable quantities of gas,
hitherto regarded as nitrogen. The question early occurred to us as to what
proportion, if any, of the new gas was contained therein. A notable example
of a nitrogen spring is that at Bath, examined by Daubeny in 1833. With
the permission of the authorities of Bath, Dr Arthur Richardson was kind
enough to collect for us about 10 litres of the gases discharged from the
King's Spring. A rough analysis on reception showed that it contained
scarcely any oxygen and but little carbonic anhydride. Two determinations
of density were made, the gas being treated in all respects as air, prepared
by diffusion and unprepared, were treated for the isolation of atmospheric
nitrogen. The results were : —
1895]
ARGON, A NEW CONSTITUENT OF THE ATMOSPHERE.
October 29 2-30513
November 7 . 2*30532
173
Mean 2-30522
The weight of the " nitrogen " from the Bath gas is thus about half-way
between that of chemical and " atmospheric " nitrogen, suggesting that the
proportion of argon is less than in air, instead of greater, as had been
expected.
13. Behaviour at Low Temperatures.
A single experiment was made with an early sample of gas, of density
19'1, which certainly contained a considerable amount of nitrogen. On
compressing it in a pressure apparatus to between 80 and 100 atmospheres
pressure, and cooling to — 90° by means of boiling nitrous oxide, no appear-
ance of liquefaction could be observed. As the critical pressure was not
likely to be so high as the pressure to which it had been exposed, the
non-liquefaction was ascribed to insufficient cooling.
VAPOUR-PRESSURES.
Temperature
Pressure
Temperature
Pressure
Temperature
Pressure
- 186-9
740-5 millims.
- 136°'2
27'3 atms.
- 129°4
35 '8 atms.
- 139-1
23-7 atms.
-135-1
29-0 „
- 128-6
38-0 „
- 138-3
25-3 „
- 134-4
29-8 „
-121-0
50-6 „
Density
Gas
Critical
tempera-
ture
Critical
pressure
Boiling-
point
Freezing-
point
Freezing
pressure
Density
of gas
of liquid
at
boiling-
Colour
of
liquid
point
atms.
millims.
-
Hydrogen, H2
Below
20-0
?°
?°
?
1
?
Colour-
- 220-0°
less
Nitrogen, N2 .
- 146-0
35-0
- 194-4
-214-0
60
14
0-885
5)
Carbon mon- )
oxide, CO... ]
- 139-5
35-5
- 190-0
- 207-0
100
14
?
n
Argon, Aj ...
-121-0
50-6
- 186-9
- 189-6
?
19-9
About
„
1-5
Oxygen, 02 ...
-118-8
50-8
-182-7
?
1
16
1-124
Bluish
Nitric oxide, )
NO (
- 93-5
71-2
- 153-6
- 167-0
138
15
?
Colour-
Methane, CH4
- 81-8
54-9
- 164-0
- 185-8
80
8
0-415
less
n
174 ARGON, A NEW CONSTITUENT OF THE ATMOSPHERE. [214
This supposition turned out to be correct. For, on sending a sample to
Professor Olszewski, the author of most of the accurate measurements of the
constants of gases at low temperatures, he was kind enough to submit it to
examination. His results are published elsewhere; but, for convenience of
reference, his tables, showing vapour-pressures, and giving a comparison
between the constants of argon and those of other gases, are here reproduced.
14. The ratio of the Specific Heats of Argon*.
In order to decide regarding the elementary or compound nature of argon,
experiments were made on the velocity of sound in it. It will be remem-
bered that from the velocity of sound, the ratio of the specific heat at
constant pressure to that at constant volume can be deduced by means of
the equation
where n is the frequency, \ is the wave-length of sound, v its velocity, e the
isothermal elasticity, d the density, (1 + at) the temperature-correction, Gp
the specific heat at constant pressure, and Gv that at constant volume. In
comparing two gases at the same temperature, each of which obeys Boyle's
law with sufficient approximation and in using the same sound, many of these
factors disappear, and the ratio of specific heats of one gas may be deduced
from that of the other, if known, by the simple proportion
\*d : \'*d' : : 1'408 : x,
where for example \ and d refer to air, of which the ratio is 1*408, according
to the mean of observations by Rb'ntgen (T4053), Wiillner (T4053), Kayser
(1-4106), and Jamin and Richard (T41).
The apparatus employed, although in principle the same as that usually
employed, differed somewhat from the ordinary pattern, inasmuch as the tube
was a narrow one, of 2 millims. bore, and the vibrator consisted of a glass rod,
sealed into one end of the tube, so that about 15 centims. projected outside
the tube, while 15 centims. was contained in the tube. By rubbing the
projecting part longitudinally with a rag wet with alcohol, vibrations of
exceedingly high pitch of the gas contained in the tube took place, causing
waves which registered their nodes by the usual device of lycopodium powder.
The temperature was that of the atmosphere and varied little from 17'5° ;
the pressure was also atmospheric, and varied only one millim. during the
experiments. Much of the success of these experiments depends on so
adjusting the length of the tube as to secure a good echo, else the wave-
heaps are indistinct. But this is easily secured by attaching to its open end
* See Addendum, p. 185.
1895]
AEGON, A NEW CONSTITUENT OF THE ATMOSPHERE.
175
a piece of thick-walled india-rubber tubing, which can be closed by a clip at
a spot which is found experimentally to produce good heaps at the nodes.
The accuracy of this instrument has frequently been tested ; but fresh
experiments were made with air, carbon dioxide, and hydrogen, so as to make
certain that reasonably reliable results were obtainable. Of these an account
is here given.
Number of observations
Half-wave-length
Gas in tube
Eatio^
I.
II.
I.
II.
Air
3
2
19-60
19-59
1-408 Assumed
C02
3
15-11
1-276 Found
H2
3
...
73-6
1-376 Found
To compare these results with those of previous observers, the following
numbers were obtained for carbon dioxide : — Cazin, 1*291 ; Rontgen, 1*305 ;
De Lucchi, 1*292; Miiller, 1*265; Wiillner, 1-311; Dulong, 1'339 ; Masson,
1-274; Regnault, 1*268; Amagat, 1'299; and Jamin and Richard, 1*29. It
appears just to reject Dulong's number, which deviates so markedly from the
rest ; the mean of those remaining is 1*288, which is in sufficient agreement
with that given above. For the ratio of the specific heats of hydrogen, we
have:— Cazin, 1*410; Rontgen, 1'385 ; Dulong, 1'407 ; Masson, 1'401 ; Reg-
nault, 1'400 ; and Jamin and Richard, 1'410. The mean of these numbers
is 1'402. This number appears to differ considerably from the one given
above. But it must be noted, first, that the wave-length which should have
been found is 74*5, a number differing but little from that actually found ;
second, that the waves were long and that the nodes were somewhat difficult
to place exactly; and third, that the atomic weight of hydrogen has been
taken as unity, whereas it is more likely to be I'Ol, if oxygen, as was done,
be taken as 16. The atomic weight 1*01 raises the found value of the ratio
to 1*399, a number differing but little from the mean value found by other
observers.
Having thus established the trustworthiness of the method, we proceed to
describe our experiments with argon.
Five series of measurements were made with the sample of gas of density
19*82. It will be remembered that a previous determination with the same
gas gave as its density 19*90. The mean of these two numbers was therefore
taken as correct, viz., 19*86.
The individual measurements are :
176
ARGON, A NEW CONSTITUENT OF THE ATMOSPHERE.
[214
I.
II.
III.
IV.
V.
Mean
18-16
18-14
18-02
18-04
18-03
millims.
18-08
for the half-wave-length. Calculating the ratio of the specific heats, the
number 1*644 is obtained.
The narrowness of the tube employed in these experiments might per-
haps raise a doubt regarding the accuracy of the measurements, for it is
conceivable that in so narrow a tube the viscosity of the gas might affect the
results. We therefore repeated the experiments, using a tube of 8 millims.
internal diameter.
The mean of eleven readings with air, at 18°, gave a half- wave-length of
34'62 millims. With argon in the same tube, and at the same temperature,
the half-wave-length was, as a mean of six concordant readings, 31'64 millims.
The density of this sample of argon, which had been transferred from a water
gas-holder to a mercury gas-holder, was 19'82 ; and there is some reason to
suspect the presence of a trace of air, for it had been standing for some time.
The result, however, substantially proves that the ratio previously found
was correct. In the wide tube, Gp : Cv :: T61 : 1. Hence the conclusion
must be accepted that the ratio of specific heats is practically T66 : 1.
It will be noticed that this is the theoretical ratio for a monatomic gas,
that is, a gas in which all energy imparted to it at constant volume is ex-
pended in effecting translational motion. The only other gas of which the
ratio of specific heats has been found to fulfil this condition is mercury at a
high temperature*. The extreme importance of these observations will be
discussed later.
15. Attempts to induce Chemical Combination.
A great number of attempts were made to induce chemical combination
with the argon obtained by use of magnesium, but without any positive
result. In such a case as this, however, it is necessary to chronicle negative
results, if for no other reason but that of justifying its name, "argon." These
will be detailed in order.
(a) Oxygen in Presence of Caustic Alkali. — This need not be further
discussed here ; the method of preparing argon is based on its inactivity
under such conditions.
* Kundt and Warburg, Pogg. Ann. 157, p. 353, 1876.
1895] ARGON, A NEW CONSTITUENT OF THE ATMOSPHERE. 177
(6) Hydrogen. — It has been mentioned that, in order to free argon from
excess of oxygen, hydrogen was admitted, and sparks passed to cause combi-
nation of hydrogen and oxygen. Here again caustic alkali was present, and
argon appeared to be unaffected.
A separate experiment was, however, made in absence of water, though
no special pains was taken to dry the mixture of gases. The argon was
admitted up to half an atmosphere pressure into a bulb, through whose sides
passed platinum wires, carrying pointed poles of gas-carbon. Hydrogen was
then admitted until atmospheric pressure had been attained. Sparks were
then passed for four hours by means of a large induction coil, actuated by
four storage cells. The gas was confined in a bulb closed by two stop-cocks,
and a small V-tube with bulbs was interposed, to act as a gauge, so that if
expansion or contraction had taken place, the escape or entry of gas would be
observable. The apparatus, after the passage of sparks, was allowed to cool
to the temperature of the atmosphere, and, on opening the stop-cock, the
level of water in the V-tube remained unaltered. It may therefore be con-
cluded that, in all probability, no combination has occurred ; or, that if it has,
it was attended with no change of volume.
(c) Chlorine. — Exactly similar experiments were performed with dry,
and afterwards with moist, chlorine. The chlorine had been stored over strong
sulphuric acid for the first experiment, and came in contact with dry argon.
Three hours sparking produced no change of volume. A drop of water was
admitted into the bulb. After four hours sparking, the volume of the gas,
after cooling, was diminished by about ^ cub. centim., due probably to the
solution of a little chlorine in the small quantity of water present.
(d) Phosphorus. — A piece of combustion-tubing, closed at one end, con-
taining at the closed end a small piece of phosphorus, was sealed to the
mercury reservoir containing argon ; connected to the same reservoir was a
mercury gauge and a Sprengel's pump. After removing all air from the
tubes, argon was admitted to a pressure of 600 millims. The middle portion
of the combustion-tube was then heated to bright redness, and the phosphorus
was distilled slowly from back to front, so that its vapour should come into
contact with argon at a red heat. When the gas was hot, the level of the
gauge altered ; but, on cooling, it returned to its original level, showing that
no contraction had taken place. The experiment was repeated several times,
the phosphorus being distilled through the red-hot tube from open to closed
end, and vice versa. In each case, on cooling, no change of pressure was
remarked. Hence it may be concluded that phosphorus at a red heat is
without action on argon. It may be remarked parenthetically that no gaseous
compound of phosphorus is known, which does not possess a volume different
from the sum of those of its constituents. That no solid compound was
R. TV. 12
178 ARGON, A NEW CONSTITUENT OF THE ATMOSPHERE. [214
formed is sufficiently proved by the absence of contraction. The phosphorus
was largely converted into the red modification during the experiment.
(e) Sulphur. — An exactly similar experiment was performed with sulphur,
again with negative results. It may therefore be concluded that sulphur and
argon are without action on each other at a red heat. And again, no gaseous
compound of sulphur is known in which the volume of the compound is equal
to the sum of those of its constituents.
(f) Tellurium. — As this element has a great tendency to unite with
heavy metals, it was thought worth while to try its action. In this, and in
the experiments to be described, a different form was given to the apparatus.
The gas was circulated over the reagent employed, a tube containing it being
placed in the circuit. The gas was dried by passage over soda-lime and
phosphoric anhydride ; it then passed over the tellurium or other reagent,
then through drying tubes, and then back to the gas-holder. That combina-
tion did not occur was shown by the unchanged volume of gas in the gas-
holder ; and it was possible, by means of the graduated cylinder which ad-
mitted water to the gas-holder, to judge of as small an absorption as half
a cubic centimetre. The tellurium distilled readily in the gas, giving the
usual yellow vapours ; and it condensed, quite unchanged, as a black subli-
mate. The volume of the gas, when all was cold, was unaltered.
(g) Sodium. — A piece of sodium, weighing about half a gramme, was
heated in argon. It attacked the glass of the combustion tube, which it
blackened, owing to liberation of silicon ; but it distilled over in drops into
the cold part of the tube. Again no change of volume occurred, nor was the
surface of the distilled sodium tarnished; it was brilliant, as it is when sodium
is distilled in vacuo. It may probably also be concluded from this experiment
that silicon, even while being liberated, is without action on argon.
The action of compounds was then tried ; those chosen were such as lead
to oxides or sulphides. Inasmuch as the platinum-metals, which are among
the most inert of elements, are attacked by fused caustic soda, its action was
investigated.
(h) Fused and Red-hot Caustic Soda. — The soda was prepared from
sodium, in an iron boat, by adding drops of water cautiously to a lump of the
metal. When action had ceased, the soda was melted, and the boat intro-
duced into a piece of combustion-tube placed in the circuit. After three
hours circulation no contraction had occurred. Hence caustic soda has no
action on argon.
(i) Soda-lime at a red heat. — Thinking that the want of porosity of fused
caustic soda might have hindered absorption, a precisely similar experiment
was carried out with soda-lime, a mixture which can be heated to bright
redness without fusion. Again no result took place after three hours heating.
1895] ARGON, A NEW CONSTITUENT OF THE ATMOSPHERE. 179
(j) Fused Potassium Nitrate was tried under the impression that oxygen
plus a base might act where oxygen alone failed. The nitrate was fused, and
kept at a bright red heat for two hours, but again without any diminution in
volume of the argon.
(k) Sodium Peroxide. — Yet another attempt was made to induce combi-
nation with oxygen and a base, by heating sodium peroxide to redness in a
current of argon for over an hour, but also without effect. It is to be noticed
that metals of the platinum group would have entered into combination
under such treatment.
(1) Persulphides of Sodium and Calcium. — Soda-lime was heated to
redness in an open crucible, and some sulphur was added to the red-hot mass,
the lid of the crucible being then put on. Combination ensued, with forma-
tion of polysulphides of sodium and calcium. This product was heated to
redness for three hours in a brisk current of argon, again with negative result.
Again, metals of the platinum group would have combined under such treat-
ment.
(ni) Some argon was shaken in a tube with nitro-hydrochloric acid.
On addition of potash, so as to neutralise the acid, and to absorb the free
chlorine and nitrosyl chloride, the volume of the gas was barely altered. The
slight alteration was evidently due to solubility in the aqueous liquid, and it
may be concluded that no chemical action took place.
(n) Bromine-water was also without effect. The bromine vapour was
removed with potash.
(o) A mixture of potassium permanganate and hydrochloric acid, involv-
ing the presence of nascent chlorine, had no action, for on absorbing chlorine
by means of potash, no alteration in volume had occurred.
(p) Argon is not absorbed by platinum black. A current was passed
over a pure specimen of this substance; as usual, however, it contained
occluded oxygen. There was no absorption in the cold. At 100° no action
took place ; and on heating to redness, by which the black was changed to
sponge, still no evidence of absorption was noticed. In all these experiments,
absorption of half a cubic centimetre of argon could have at once been
detected.
We do not claim to have exhausted the possible reagents. But this much
is certain, that the gas deserves the name " argon," for it is a most astonish-
ingly indifferent body, inasmuch as it is unattacked by elements of very
opposite character, ranging from sodium and magnesium on the one hand, to
oxygen, chlorine, and sulphur on the other. It will be interesting to see if
fluorine also is without action, but for the present that experiment must be
postponed, on account of difficulties of manipulation.
12—2
180 ARGON, A NEW CONSTITUENT OF THE ATMOSPHERE. [214
It will also be necessary to try whether the inability of argon to combine
at ordinary or at high temperatures is due to the instability of its possible
compounds, except when cold. Mercury vapour at 800° would present a
similar instance of passive behaviour.
16. General Conclusions.
It remains, finally, to discuss the probable nature of the gas or gases
which we have succeeded in separating from atmospheric air, and which has
been provisionally named argon.
That argon is present in the atmosphere, and is not manufactured during
the process of separation is amply proved by many lines of evidence. First,
atmospheric nitrogen has a high density, while chemical nitrogen is lighter.
That chemical nitrogen is a uniform substance is proved by the identity of
properties of samples prepared by several different processes, and from several
different compounds. It follows, therefore, that the cause of the high density
of atmospheric nitrogen is due to the admixture with heavier gas. If that
gas possesses the density of 20 compared with hydrogen as unity, atmospheric
nitrogen should contain of it approximately 1 per cent. This is found to be
the case, for on causing the nitrogen of the atmosphere to combine with
oxygen in presence of alkali, the residue amounted to about 1 per cent. ; and
on removing nitrogen with magnesium the result is similar.
Second : This gas has been concentrated in the atmosphere by diffusion.
It is true that it cannot be freed from oxygen and nitrogen by diffusion, but
the process of diffusion increases relatively to nitrogen the amount of argon
in that portion which does not pass through the porous walls. That this is
the case is proved by the increase of density of that mixture of argon and
nitrogen.
Third : On removing nitrogen from " atmospheric nitrogen " by means of
magnesium, the density of the residue increases proportionately to the concen-
tration of the heavier constituent.
Fourth : As the solubility of argon in water is relatively high, it is to be
expected that the density of the mixture of argon and nitrogen, pumped out
of water along with oxygen should, after removal of the oxygen, exceed that
of " atmospheric nitrogen." Experiment has shown that the density is con-
siderably increased.
Fifth: It is in the highest degree improbable that two processes, so
different from each other, should each manufacture the same product. The
explanation is simple if it be granted that these processes merely eliminate
nitrogen from " atmospheric nitrogen."
1895] ARGON, A NEW CONSTITUENT OF THE ATMOSPHERE. 181
Sixth : If the newly discovered gas were not in the atmosphere, the dis-
crepancies in the density of " chemical " and " atmospheric " nitrogen would
remain unexplained.
Seventh : It has been shown that pure nitrogen, prepared from its com-
pounds, leaves a negligible residue when caused to enter into combination
with oxygen or with magnesium.
There are other lines of argument which suggest themselves; but we
think that it will be acknowledged that those given above are sufficient to
establish the existence of argon in the atmosphere.
It is practically certain that the argon prepared by means of electric
sparking with oxygen is identical with argon prepared by means of magne-
sium. The samples have in common : —
First : Spectra which have been found by Mr Crookes, Professor Schuster,
and ourselves to be practically identical.
Second : They have approximately the same density. The density of
argon, prepared by means of magnesium, was 19'9 ; that of argon, from spark-
ing with oxygen, about 197 ; these numbers are practically identical.
Third : Their solubility in water is the same.
That argon is an element, or a mixture of elements, may be inferred from
the observations of § 14. For Clausius has shown that if K be the energy of
translatory motion of the molecules of a gas, and H their whole kinetic energy,
then
K 3(CP-CV)
H~ ~^CV '
Cp and Cv denoting as usual the specific heat at constant pressure and at
constant volume respectively. Hence, if, as for mercury vapour and for argon
(§ 14), the ratio of specific heats Gp : Cv be If, it follows that K = H, or that
the whole kinetic energy of the gas is accounted for by the translatory motion
of its molecules. In the case of mercury the absence of interatomic energy
is regarded as proof of the monatomic character of the vapour, and the
conclusion holds equally good for argon.
The only alternative is to suppose that if argon molecules are di- or poly-
atomic, the atoms acquire no relative motion, even of rotation, a conclusion
improbable in itself and one postulating the sphericity of such complex groups
of atoms.
Now a monatomic gas can be only an element, or a mixture of elements ;
and hence it follows that argon is not of a compound nature.
According to Avogadro, equal volumes of gases at the same temperature
and pressure contain equal numbers of molecules. The molecule of hydrogen
182 ARGON, A NEW CONSTITUENT OF THE ATMOSPHERE. [214
gas, the density of which is taken as unity, is supposed to consist of two
atoms. Its molecular weight is therefore 2. Argon is approximately 20
times as heavy as hydrogen, that is, its molecular weight is 20 times as great
as that of hydrogen, or 40. But its molecule is monatomic, hence its atomic
weight, or, if it be a mixture, the mean of the atomic weights of the elements
in that mixture, taken for the proportion in which they are present, must
be 40.
This conclusion rests on the assumption that all the molecules of argon
are monatomic. The result of the first experiment is, however, so nearly that
required by theory, that there is room for only a small number of molecules
of a different character. A study of the expansion of argon by heat is pro-
posed, and would doubtless throw light upon this question.
There is evidence both for and against the hypothesis that argon is a
mixture : for, owing to Mr Crookes' observations of the dual character of its
spectrum ; against, because of Professor Olszewski's statement that it has a
definite melting-point, a definite boiling-point, and a definite critical tem-
perature and pressure ; and because on compressing the gas in presence of its
liquid, pressure remains sensibly constant until all gas has condensed to
liquid. The latter experiments are the well-known criteria of a pure sub-
stance; the former is not known with certainty to be characteristic of a
mixture. The conclusions which follow are, however, so startling, that in our
future experimental work we shall endeavour to decide the question by other
means.
For the present, however, the balance of evidence seems to point to sim-
plicity. We have, therefore, to discuss the relations to other elements of an
element of atomic weight 40. We inclined for long to the view that argon
was possibly one, or more than one, of the elements which might be expected
to follow fluorine in the periodic classification of the elements — elements
which should have an atomic weight between 19, that of fluorine, and 23,
that of sodium. But this view is apparently put out of court by the discovery
of the monatomic nature of its molecules.
The series of elements possessing atomic weights near 40 are : —
Chlorine 35'5
Potassium 39'1
Calcium 40'0
Scandium 44'0
There can be no doubt that potassium, calcium, and scandium follow
legitimately their predecessors in the vertical columns, lithium, beryllium, and
boron, and that they are in almost certain relation with rubidium, strontium,
and (but not so certainly) yttrium. If argon be a single element, then there
1895] ARGON, A NEW CONSTITUENT OF THE ATMOSPHERE. 183
is reason to doubt whether the periodic classification of the elements is com-
plete ; whether, in fact, elements may not exist which cannot be fitted among
those of which it is composed. On the other hand, if argon be a mixture of
two elements, they might find place in the eighth group, one after chlorine
and one after bromine. Assuming 37 (the approximate mean between the
atomic weights of chlorine and potassium) to be the atomic weight of the
lighter element, and 40 the mean atomic weight found, and supposing that
the second element has an atomic weight between those of bromine, 80, and
rubidium, 85*5, viz. 82, the mixture should consist of 93'3 per cent, of the
lighter, and 6'7 per cent, of the heavier element. But it appears improbable
that such a high percentage as 6'7 of a heavier element should have escaped
detection during liquefaction.
If the atomic weight of the lighter element were 38, instead of 37, how-
ever, the proportion of heavier element would be considerably reduced. Still,
it is difficult to account for its not having been detected, if present.
If it be supposed that argon belongs to the eighth group, then its proper-
ties would fit fairly well with what might be anticipated. For the series,
which contains
SiIV, Pni*ndv S™VI, and ClItovn
might be expected to end with an element of monatomic molecules, of no
valency, i.e. incapable of forming a compound, or if forming one, being an
octad ; and it would form a possible transition to potassium, with its mono-
valence, on the other hand. Such conceptions are, however, of a speculative
nature ; yet they may be perhaps excused, if they in any way lead to experi-
ments which tend to throw more light on the anomalies of this curious
element.
In conclusion, it need excite no astonishment that argon is so indifferent
to reagents. For mercury, although a monatomic element, forms compounds
which are by no means stable at a high temperature in the gaseous state ;
and attempts to produce compounds of argon may be likened to attempts to
cause combination between mercury gas at 800° and other elements. As for
the physical condition of argon, that of a gas, we possess no knowledge why
carbon, with its low atomic weight, should be a solid, while nitrogen is a gas,
except in so far as we ascribe molecular complexity to the former and com-
parative molecular simplicity to the latter. Argon, with its comparatively
low density and its molecular simplicity, might well be expected to rank
among the gases. And its inertness, which has suggested its name, suffi-
ciently explains why it has not previously been discovered as a constituent of
compound bodies.
We would suggest for this element, assuming provisionally that it is not
a mixture, the symbol A.
184
ARGON, A NEW CONSTITUENT OF THE ATMOSPHERE.
[214
We have to record our thanks to Messrs Gordon, Kellas, and Matthews,
and especially to Mr Percy Williams, for their assistance in the prosecution
of this research.
ADDENDUM (by Professor W. Ramsay).
March 20, 1895.
Further determinations of the density of argon prepared by means of
magnesium have been made. In each case the argon was circulated over
magnesium for at least two hours after all absorption of nitrogen had stopped,
as well as over red-hot copper, copper oxide, soda-lime, and phosphoric anhy-
dride. The gas also passed out of the mercury gas-holder through phosphoric
anhydride into the weighing globe. The results are in complete accordance
with previous determinations of density ; and for convenience of reference the
former numbers are included in the table which follows.
DENSITY OF ARGON.
Date
Volume
Tempera-
ture
Pressure
Weight
Weight of
1 litre at
0° and 760
millims.
Density
(0 = 16)
(1) Nov. 26
cub. centims.
162-843
15°00
millims.
767-7
grm.
0-2773
1-7784
19-904
(2) „ 27
162-843
16-00
769-0
0-2757
1-7717
19-823
(3) Dec. 22
162-843
15-62
750-1
0-26915
1-7704
19-816
(4) Feb. 16
162-843
13-45
771-1
0-2818
1-7834
19-959
(5) „ 19
162-843
14-47
768-2
0-2789
1-7842
19-969
(6) „ 24
162-843
17-85
764-4
0-2738
1-7810
19-932
The general mean is 19*900; or if Nos. (2) and (3) be rejected as sus-
piciously low, the mean of the remaining four determinations is 19'941. The
molecular weight may therefore be taken as 39*9 without appreciable error.
The value of R in the gas-equation R=pv/T has also been determined
between — 89° and + 248°. For this purpose, a gas-thermometer was filled
with argon, and a direct comparison was made with a similar thermometer
filled with hydrogen.
The method of using such a hydrogen-thermometer has already been
described by Ramsay and Shields*. For the lowest temperature, the ther-
mometer bulbs were immersed in boiling nitrous oxide; for atmospheric
temperature, in running water ; for temperatures near 100° in steam, and for
the remaining temperatures, in the vapours of chlorobenzene, aniline, and
quinolene.
* Trans. Chem. Soc. Vol. 63, pp. 835, 836. It is to be noticed that the value of R is not
involved in using the hydrogen-thermometer ; its constancy alone is postulated.
1895] ARGON, A NEW CONSTITUENT OF THE ATMOSPHERE.
The results are collected in the following tables : —
HYDROGEN THERMOMETER.
185
Temperature
Pressure
Volume (corr.)
E
13-04
millims.
763-6
1-00036
2-6705
99-84
992-6
1-00280
2-6697
130-62
1073-8
1-00364
2-6701
185-46
1218-5
1-00518
2-6716
248-66
1385-1
1-00703
2-6737
-87-92
497-3
0-99756
2-6804
The value of R is thus practically constant, and this affords a proof that-
the four last temperatures have been estimated with considerable accuracy.
ARGON THERMOMETER.
Temperature
Pressure
Volume (corr.)
R
Series I
14-15
millims.
701-7
1-000396
2-4446
14-27
699-7
1-000401
2-4366
14-40
702-6
1-000404
2-4462
19-96
906-5
1-00280
2-4379
100-06
904-8
1-00280
2-4322
-87-92
455-6
0-99756
2-4556
By mischance, air leaked into the bulb ; it was therefore refilled.
Series II. ...
130-58
1060-0
1-0037
2-6363
185-46
1200-3
1-0052
2-6317
A bubble of argon leaked into the bulb, and the value of R increased.
Series III....
12-05
760-9
1-00034
2-6698
12-61
761-3
1-00034
2-6728
248-66
1384-0
1-0070
2-6717
248-66
1376-9
1-0070
2-6580
- 87-92
495-7
0-99756
2-6718
It may be concluded from these numbers, that argon undergoes no mole-
cular change between — 88° and + 250°.
Further determinations of the wave-length of sound in argon have been
made, the wider tube having been used. In every case the argon was as
186
ARGON, A NEW CONSTITUENT OF THE ATMOSPHERE.
[214
carefully purified as possible. In experiment (3) too much lycopodium dust
was present in the tube ; that is perhaps the cause of the low result. For
completeness' sake, the original result in the narrow tube has also been given.
Date
Density
Half- wave-length
Temperature
Ratio
In air
In argon
Air
Argon
Dec 6
19-92
19-96
19-97
19-94
19-59
33-73
34-10
34-23
18-08
31-00
31-31
31-68
17°5
67
7-22
11-20
17°5
6-5
8-64
11-49
1-644
1-641
1-629
1-659
Feb 15
„ 20
Mar 19
The general mean of these numbers is 1*643; if (3) be rejected, it is 1*648.
In the last experiment every precaution was taken. The half-wave-length
in air is the mean of 11 readings, the highest of which was 34*67 and the
lowest 34-00. They run :—
34-67 ; 34-06 ; 34-27 ; 34'39 ; 34*00 ; 34*00 ; 34*13 ; 34-20 ; 34*20 ; 34*33 ; 34-33.
11*25°; 11-00°; 10*80°; 10'8°; 10*0° ; 11*0° ; 11*3° ; 11*4° ; 11-4°; 11*6° ; 11*6°.
With argon the mean is also that of 11 readings, of which the highest is
31-83, and the lowest, 31*5. They are :—
31*5 ;31-5 ; 31*66 ; 31*55 ; 31*83 ; 31*77 ; 31-81 ; 31*83 ; 31*83 ; 31-50; 31'66.
11*8° ; 11-8° ; 11-20° ; 11*40° ; 11*60° ; 11*40° ; 11-40°; 11-4° ; 11*5° ; 11*5° ; 11*4°.
If the atomic weight of argon is identical with its molecular weight, it
must closely approximate to 39*9. But if there were some molecules of A2
present, mixed with a much larger number of molecules of A]; then the
atomic weight would be correspondingly reduced. Taking an imaginary case,
the question may be put: — What percentage of molecules of A2 would raise
the density of A1 from 19*0 to 19*9 ? A density of 19*0 would imply an
atomic weight of 38*0, and argon would fall into the gap between chlorine
and potassium. Calculation shows that in 10,000 molecules, 474 molecules
of A.J would have this result, the remaining 9526 molecules being those of Ax.
Now if molecules of A^ be present, it is reasonable to suppose that their
number would be increased by lowering the temperature, and diminished by
heating the gas. A larger change of density should ensue on lowering than
on raising the temperature, however, as on the above supposition, there is
not a large proportion of molecules of A2 present.
But it must be acknowledged that the constancy of the found value of R
is not favourable to this supposition.
1895] ARGON, A NEW CONSTITUENT OF THE ATMOSPHERE. 187
A similar calculation is possible for the ratio of specific heats. Assuming
the gas to contain 5 per cent, of molecules of A2, and 95 per cent, of mole-
cules of AI the value of 7, the ratio of specific heats, would be 1*648. All
that can be said on this point is, that the found ratio approximates to this
number ; but whether the results are to be trusted to indicate a unit in the
second decimal appears to me doubtful.
The question must therefore for the present remain open.
ADDENDUM.
April 9.
It appears worth while to chronicle an experiment of which an accident
prevented the completion. It may be legitimately asked, Does magnesium
not absorb any argon, or any part of what we term argon ? To decide this
question, about 500 grms. of magnesium nitride, mixed with metallic mag-
nesium which had remained unacted on, during extraction of nitrogen from
" air-nitrogen," was placed in a flask, to which a reservoir full of dilute hydro-
chloric acid was connected. The flask was coupled with a tube full of red-hot
copper oxide, intended to oxidise the hydrogen which would be evolved by
the action of the hydrochloric acid on the metallic magnesium. To the end
of the copper oxide tube a gas-holder was attached, so as to collect any
evolved gas ; and the system was attached to a vacuum-pump, in order to
exhaust the apparatus before commencing the experiment, as well as to
collect all gas which should be evolved, and remain in the flask.
On admitting hydrochloric acid to the flask of magnesium nitride a violent
reaction took place, and fumes of ammonium chloride passed into the tube of
copper oxide. These gave, of course, free nitrogen. This had not been fore-
seen ; it would have been well to retain these fumes by plugs of glass-wool.
The result of the experiment was that about 200 cub. centims. of gas were
collected. After sparking with oxygen in presence of caustic soda, the volume
was reduced to 3 cub. centims. of a gas which appeared to be argon.
215.
ARGON.
[Royal Institution Proceedings, xiv. pp. 524—538, Ap. 1895.]
IT is some three or four years since I had the honour of lecturing here one
Friday evening upon the densities of oxygen and hydrogen gases, and upon
the conclusions that might be drawn from the results. It is not necessary,
therefore, that I should trouble you to-night with any detail as to the method
by which gases can be accurately weighed. I must take that as known,
merely mentioning that it is substantially the same as is used by all investi-
gators nowadays, and introduced more than fifty years ago by Regnault. It
was not until after that lecture that I turned my attention to nitrogen ; and
in the first instance I employed a method of preparing the gas which originated
with Mr Vernon Harcourt, of Oxford. In this method the oxygen of ordinary
atmospheric air is got rid of with the aid of ammonia. Air is bubbled through
liquid ammonia, and then passed through a red-hot tube. In its passage the
oxygen of the air combines with the hydrogen of the ammonia, all the oxygen
being in that way burnt up and converted into water. The excess of ammonia
is subsequently absorbed with acid, and the water by ordinary desiccating
agents. That method is very convenient ; and, when I had obtained a few
concordant results by means of it, I thought that the work was complete, and
that the weight of nitrogen was satisfactorily determined. But then I
reflected that it is always advisable to employ more than one method, and
that the method that I had used — Mr Vernon Harcourt's method — was not
that which had been used by any of those who had preceded me in weighing
nitrogen. The usual method consists in absorbing the oxygen of air by means
of red-hot copper ; and I thought that I ought at least to give that method a
trial, fully expecting to obtain forthwith a value in harmony with that already
afforded by the ammonia method. The result, however, proved otherwise.
The gas obtained by the copper method, as I may call it, proved to be one-
1895] ARGON. 189
thousandth part heavier than that obtained by the ammonia method ; and, on
repetition, that difference was only brought out more clearly. This was about
three years ago. In order, if possible, to get further light upon a discrepancy
which puzzled me very much, and which, at that time, I regarded only with
disgust and impatience, I published a letter in Nature* inviting criticisms
from chemists who might be interested in such questions. I obtained various
useful suggestions, but none going to the root of the matter. Several persons
who wrote to me privately were inclined to think that the explanation was to
be sought in a partial dissociation of the nitrogen derived from ammonia.
For, before going further, I ought to explain that, in the nitrogen obtained by
the ammonia method, some — about a seventh part — is derived from the
ammonia, the larger part, however, being derived as usual from the atmosphere.
If the chemically derived nitrogen were partly dissociated into its component
atoms, then the lightness of the gas so prepared would be explained.
The next step in the enquiry was, if possible, to exaggerate the discrepancy.
One's instinct at first is to try to get rid of a discrepancy, but I believe that
experience shows such an endeavour to be a mistake. What one ought to do
is to magnify a small discrepancy with a view to finding out the explanation ;
and, as it appeared in the present case that the root of the discrepancy lay in
the fact that part of the nitrogen prepared by the ammonia method was
nitrogen out of ammonia, although the greater part remained of common
origin in both cases, the application of the principle suggested a trial of the
weight of nitrogen obtained wholly from ammonia. This could easily be done
by substituting pure oxygen for atmospheric air in the ammonia method, so
that the whole, instead of only a part, of the nitrogen collected should be
derived from the ammonia itself. The discrepancy was at once magnified
some five times. The nitrogen so obtained from ammonia proved to be about
one-half per cent, lighter than nitrogen obtained in the ordinary way from the
atmosphere, and which I may call for brevity " atmospheric " nitrogen.
That result stood out pretty sharply from the first; but it was necessary
to confirm it by comparison with nitrogen chemically derived in other ways.
The table before you gives a summary of such results, the numbers being the
weights in grams actually contained under standard conditions in the globe
employed.
ATMOSPHERIC NITROGEN.
By hot copper (1892) 2'3103
By hot iron (1893) 2'3100
By ferrous hydrate (1894) 2'3102
Mean 2'3102
[Vol. iv. p. 1.]
190 ARGON. [215
CHEMICAL NITROGEN.
From nitric oxide 2'3001
From nitrous oxide 2'2990
From ammonium nitrite purified at a red heat . . . 2'2987
From urea 2'2985
From ammonium nitrite purified in the cold .... 2'2987
Mean 2'2990
The difference is about 11 milligrams, or about one-half per cent.; and it
was sufficient to prove conclusively that the two kinds of nitrogen — the
chemically derived nitrogen and the atmospheric nitrogen — differed in weight,
and therefore, of course, in quality, for some reason hitherto unknown.
I need not spend time in explaining the various precautions that were
necessary in order to establish surely that conclusion. One had to be on one's
guard against impurities, especially against the presence of hydrogen, which
might seriously lighten any gas in which it was contained. I believe, however,
that the precautions taken were sufficient to exclude all questions of that
sort, and the result, which I published about this time last year*, stood
sharply out, that the nitrogen obtained from chemical sources was different
from the nitrogen obtained from the air.
Well, that difference, admitting it to be established, was sufficient to show
that some hitherto unknown gas is involved in the matter. It might be that
the new gas was dissociated nitrogen, contained in that which was too light,
the chemical nitrogen — and at first that was the explanation to which I
leaned ; but certain experiments went a long way to discourage such a suppo-
sition. In the first place, chemical evidence — and in this matter I am greatly
dependent upon the kindness of chemical friends — tends to show that, even if
ordinary nitrogen could be dissociated at all into its component atoms, such
atoms would not be likely to enjoy any very long continued existence. Even
ozone goes slowly back to the more normal state of oxygen; and it was
thought that dissociated nitrogen would have even a greater tendency to
revert to the normal condition. The experiment suggested by that remark
was as follows — to keep chemical nitrogen — the too light nitrogen which
might be supposed to contain dissociated molecules — for a good while, and to
examine whether it changed in density. Of course it would be useless to
shut up gas in a globe and weigh it, and then, after an interval, to weigh it
again, for there would be no opportunity for any change of weight to occur,
even although the gas within the globe had undergone some chemical
alteration. It is necessary to re-establish the standard conditions of tempera-
ture and pressure which are always understood when we speak of filling a
* [Vol. iv. p. 104.]
1895] ARGON. 191
globe with gas, for I need hardly say that filling a globe with gas is but a
figure of speech. Everything depends upon the temperature and pressure at
which you work. However, that obvious point being borne in mind, it was
proved by experiment that the gas did not change in weight by standing for
eight months — a result tending to show that the abnormal lightness was not
the consequence of dissociation.
Further experiments were tried upon the action of the silent electric dis-
charge— both upon the atmospheric nitrogen and upon the chemically derived
nitrogen — but neither of them seemed to be sensibly affected by such
treatment; so that, altogether, the balance of evidence seemed to incline
against the hypothesis of abnormal lightness in the chemically derived
nitrogen being due to dissociation, and to suggest strongly, as almost the only
possible alternative, that there must be in atmospheric nitrogen some con-
stituent heavier than true nitrogen.
At that point the question arose, What was the evidence that all the so-
called nitrogen of the atmosphere was of one quality ? And I remember — I
think it was about this time last year, or a little earlier — putting the question
to my colleague, Professor Dewar. His answer was that he doubted whether
anything material had been done upon the matter since the time of Cavendish,
and that I had better refer to Cavendish's original paper. That advice I
quickly followed, and I was rather surprised to find that Cavendish had him-
self put this question quite as sharply as I could put it. Translated from the
old-fashioned phraseology connected with the theory of phlogiston, his question
was whether the inert ingredient of the air is really all of one kind ; whether
all the nitrogen of the air is really the same as the nitrogen of nitre.
Cavendish not only asked himself this question, but he endeavoured to answer
it by an appeal to experiment.
I should like to show you Cavendish's experiment in something like its
original form. He inverted a U-tube filled with mercury, the legs standing in
two separate mercury cups. He then passed up, so as to stand above the
mercury, a mixture of nitrogen, or of air, and oxygen; and he caused an
electric current from a frictional electrical machine like the one I have before
me to pass from the mercury in the one leg to the mercury in the other,
giving sparks across the intervening column of air. • I do not propose to use a
frictional machine to-night, but I will substitute for it one giving electricity
of the same quality of the construction introduced by Mr Wimshurst, of which
we have a fine specimen in the Institution. It stands just outside the door
of the theatre, and will supply an electric current along insulated wires, lead-
ing to the mercury cups ; and, if we are successful, we shall cause sparks to
pass through the small length of air included above the columns of mercury.
There they are ; and after a little time you will notice that the mercury rises,
indicating that the gas is sensibly absorbed under the influence of the sparks
192 ARGON. [215
and of a piece of potash floating on the mercury. It was by that means that
Cavendish established his great discovery of the nature of the inert ingredient
in the atmosphere, which we now call nitrogen; and, as I have said, Cavendish
himself proposed the question, as distinctly as we can do, Is this inert
ingredient all of one kind ? and he proceeded to test that question. He
found, after days and weeks of protracted experiment, that, for the most part,
the nitrogen of the atmosphere was absorbed in this manner, and converted
into nitrous acid ; but that there was a small residue remaining after pro-
longed treatment with sparks, and a final absorption of the residual oxygen.
That residue amounted to about T^ part of the nitrogen taken; and Cavendish
draws the conclusion that, if there be more than one inert ingredient in the
atmosphere, at any rate the second ingredient is not contained to a greater
extent than T^y part.
I must not wait too long over the experiment. Mr Gordon tells me that
a certain amount of contraction has already occurred ; and if we project the U
upon the screen, we shall be able to verify the fact. It is only a question of
time for the greater part of the gas to be taken up, as we have proved by
preliminary experiments.
In what I have to say from this point onwards, I must be understood as
speaking as much on behalf of Professor Ramsay as for myself. At the first,
the work which we did was to a certain extent independent. Afterwards we
worked in concert, and all that we have published in our joint names must be
regarded as being equally the work of both of us. But, of course, Professor
Ramsay must not be held responsible for any chemical blunder into which I
may stumble to-night.
By his work and by mine the heavier ingredient in atmospheric nitrogen
which was the origin of the discrepancy in the densities has been isolated, and
we have given it the name of " argon." For this purpose we may use the
original method of Cavendish, with the advantages of modern appliances. We
can procure more powerful electric sparks than any which Cavendish could
command by the use of the ordinary Ruhmkorff coil stimulated by a battery
of Grove cells; and it is possible so to obtain evidence of the existence of
argon. The oxidation of nitrogen by that method goes on pretty quickly. If
you put some ordinary air, or, better still, a mixture of air and oxygen, in a tube
in which electric sparks are made to pass for a certain time, then in looking
through the tube, you observe the well-known reddish-orange fumes of the
oxides of nitrogen. I will not take up time in going through the experiment,
but will merely exhibit a tube already prepared (image on screen).
One can work more efficiently by employing the alternate currents from
dynamo machines which are now at our command. In this Institution we
have the advantage of a public supply; and if I pass alternate currents
1895] AKGON. 193
originating in Deptford through this Ruhmkorff coil, which acts as what is
now called a " high potential transformer," and allow sparks from the secondary
to pass in an inverted test tube between platinum points, we shall be able to
show in a comparatively short time a pretty rapid absorption of the gases.
The electric current is led into the working chamber through bent glass tubes
containing mercury, and provided at their inner extremities with platinum
points. In this arrangement we avoid the risk, which would otherwise be
serious, of a fracture just when we least desired it. I now start the sparks by
switching on the Ruhmkorff to the alternate current supply ; and, if you will
take note of the level of the liquid representing the quantity of mixed gases
included, I think you will see after, perhaps, a quarter of an hour that the
liquid has very appreciably risen, owing to the union of the nitrogen and the
oxygen gases under the influence of the electrical discharge, and subsequent
absorption of the resulting compound by the alkaline liquid with which the gas
space is enclosed.
By means of this little apparatus, which is very convenient for operations
upon a moderate scale, such as analyses of " nitrogen " for the amount of argon
that it may contain, we are able to get an absorption of about 80 cubic centi-
metres per hour, or about 4 inches along this test tube, when all is going well.
In order, however, to effect the isolation of argon on any considerable 'scale
by means of the oxygen method, we must employ an apparatus still more
enlarged. The isolation of argon requires the removal of nitrogen, and, indeed,
of very large quantities of nitrogen, for, as it appears, the proportion of argon
contained in atmospheric nitrogen is only about 1 per cent., so that for every
litre of argon that you wish to get you must eat up some hundred litres of
nitrogen. That, however, can be done upon an adequate scale by calling to
our aid the powerful electric discharge now obtainable by means of the
alternate current supply and high potential transformers.
In what I have done upon this subject I have had the advantage of the
advice of Mr Crookes, who some years ago drew special attention to the
electric discharge or flame, and showed that many of its properties depended
upon the fact that it had the power of causing, upon a very considerable scale,
a combination of the nitrogen and the oxygen of the air in which it was
made.
I had first thought of showing in the lecture room the actual apparatus
which I have employed for the concentration of argon ; but the difficulty is
that, as the apparatus has to be used, the working parts are almost invisible,
and I came to the conclusion that it would really be more instructive as well
as more convenient to show the parts isolated, a very little effort of imagina-
tion being then all that is required in order to reconstruct in the mind the
actual arrangements employed.
R. iv. 13
194 ARGON. [215
First, as to the electric arc or flame itself. We have here a transformer
made by Pike and Harris. It is not the one that I have used in practice ;
but it is convenient for certain purposes, and it can be connected by means of
a switch with the alternate currents of 100 volts furnished by the Supply
Company. The platinum terminals that you see here are modelled exactly
upon the plan of those which have been employed in practice. I may say a
word or two on the question of mounting. The terminals require to be very
massive on account of the heat evolved. In this case they consist of platinum
wire doubled upon itself six times. The platinums are continued by iron
wires going through glass tubes, and attached at the ends to the copper leads.
For better security, the tubes themselves are stopped at the lower ends with
corks and charged with water, the advantage being that, when the whole
arrangement is fitted by means of an indiarubber stopper into a closed vessel,
you have a witness that, as long as the water remains in position, no leak can
have occurred through the insulating tubes conveying the electrodes.
Now, if we switch on the current and approximate the points sufficiently,
we get the electric flame. There you have it. It is, at present, showing a
certain amount of soda. That in time would burn off. After the arc has once
been struck, the platinums can be separated ; and then you have two tongues
of fire ascending almost independently of one another, but meeting above.
Under the influence of such a flame, the oxygen and the nitrogen of the air
combine at a reasonable rate, and in this way the nitrogen is got rid of. It is
now only a question of boxing up the gas in a closed space, where the argon
concentrated by the combustion of the nitrogen can be collected. But there
are difficulties to be encountered here. One cannot well use anything but a
glass vessel. There is hardly any metal available that will withstand the
action of strong caustic alkali and of the nitrous fumes resulting from the
flame. One is practically limited to glass. The glass vessel employed is a
large flask with a single neck, about half full of caustic alkali. The electrodes
are carried through the neck by means of an indiarubber bung provided also
with tubes for leading in the gas. The electric flame is situated at a distance
of only about half an inch above the caustic alkali. In that way an efficient
circulation is established ; the hot gases as they rise from the flame strike the
top, and then as they come round again in the course of the circulation they
•pass sufficiently close to the caustic alkali to ensure an adequate removal of
the nitrous fumes.
There is another point to be mentioned. It is necessary to keep the
vessel cool; otherwise the heat would soon rise to such a point that there
would be excessive generation of steam, and then the operation would come to
a standstill. In order to meet this difficulty the upper part of the vessel is
provided with a water-jacket, in which a circulation can be established. No
doubt the glass is severely treated, but it seems to stand it in a fairly amiable
1895] ARGON. 195
By means of an arrangement of this kind, taking nearly three horse-power
from the electric supply, it is possible to consume nitrogen at a reasonable
rate. The transformers actually used are the "Hedgehog" transformers of
Mr Swinburne, intended to transform from 100 volts to 2400 volts. By
Mr Swinburne's advice I have used two such, the fine wires being in series so
as to accumulate the electrical potential and the thick wires in parallel. The
rate at which the mixed gases are absorbed is about seven litres per hour ; and
the apparatus, when once fairly started, works very well as a rule, going for
many hours without attention. At times the arc has a trick of going out, and
it then requires to be restarted *by approximating the platinums. We have
already worked 14 hours on end, and by the aid of one or two automatic
appliances it would, I think, be possible to continue operations day and
night.
The gases, air and oxygen in about equal proportions, are mixed in a large
gas-holder, and are fed in automatically as required. The argon gradually
accumulates ; and when it is desired to stop operations the supply of nitrogen
is cut off, and only pure oxygen allowed admittance. In this way the remain-
ing nitrogen is consumed, so that, finally, the working vessel is charged with
a mixture of argon and oxygen only, from which the oxygen is removed by
ordinary well-known chemical methods. I may mention that at the close of
the operation, when the nitrogen is all gone, the arc changes its appearance,
and becomes of a brilliant blue colour.
I have said enough about this method, and I must now pass on to the
alternative method which has been very successful in Professor Ramsay's
hands — that of absorbing nitrogen by means of red-hot magnesium. By the
kindness of Professor Ramsay and Mr Matthews, his assistant, we have here
the full scale apparatus before us almost exactly as they use it. On the
left there is a reservoir of nitrogen derived from air by the simple removal
of oxygen. The gas is then dried. Here it is bubbled through sulphuric
acid. It then passes through a long tube made of hard glass and charged
with magnesium in the form of thin turnings. During the passage of the gas
over the magnesium at a bright red heat, the nitrogen is absorbed in a great
degree, and the gas which finally passes through is immensely richer in argon
than that which first enters the hot tube. At the present time you see a
tolerably rapid bubbling on the left, indicative of the flow of atmospheric
nitrogen into the combustion furnace ; whereas, on the right, the outflow is
very much slower. Care must be taken to prevent the heat rising to such a
point as to soften the glass. The concentrated argon is collected in a second
gas-holder, and afterwards submitted to further treatment. The apparatus
employed by Professor Ramsay in the subsequent treatment is exhibited
in the diagram, and is very effective for its purpose ; but I am afraid that
the details of it would not readily be followed from any explanation that
13—2
196 ARGON. [215
I could give in the time at my disposal. The principle consists in the
circulation of the mixture of nitrogen and argon over hot magnesium, the gas
being made to pass round and round until the nitrogen is effectively removed
from it. At the end that operation, as in the case of the oxygen method,
proceeds somewhat slowly. When the greater part of the nitrogen is gone,
the remainder seems to be unwilling to follow, and it requires somewhat pro-
tracted treatment in order to be sure that the nitrogen has wholly disappeared.
When I say " wholly disappeared," that, perhaps, would be too much to say in
any case. What we can say is that the spectrum test is adequate to show the
presence, or at any rate to show the addition,' of about one-and-a-half per cent,
of nitrogen to argon as pure as we can get it ; so that it is fair to argue that
any nitrogen at that stage remaining in the argon is only a small fraction of
one-and-a-half per cent.
I should have liked at this point to be able to give advice as to which of
the two methods — the oxygen method or the magnesium method — is the
easier and the more to be recommended ; but I confess that I am quite at a
loss to do so. One difficulty in the comparison arises from the fact that they
have been in different hands. As far as I can estimate, the quantities of
nitrogen eaten up in a given time are not very different. In that respect,
perhaps, the magnesium method has some advantage ; but, on the other hand,
it may be said that the magnesium process requires a much closer supervision,
so that, perhaps, fourteen hours of the oxygen method may not unfairly
compare with eight hours or so of the magnesium method. In practice a
great deal would depend upon whether in any particular laboratory alternate
currents are available from a public supply. If the alternate currents are at
hand, I think it may probably be the case that the oxygen method is the
easier ; but, otherwise, the magnesium method would, probably, be preferred,
especially by chemists who are familiar with operations conducted in red-hot
tubes.
I have here another experiment illustrative of the reaction between
magnesium and nitrogen. Two- rods of that metal are suitably mounted in
an atmosphere of nitrogen, so arranged that we can bring them into contact
and cause an electric arc to form between them. Under the action of the
heat of the electric arc the nitrogen will combine with the magnesium ; and
if we had time to carry out the experiment we could demonstrate a rapid
absorption of nitrogen by this method. When the experiment was first tried,
I had hoped that it might be possible, by the aid of electricity, to start the
action so effectively that the magnesium would continue to burn independently
under its own developed heat in the atmosphere of nitrogen. Possibly, on a
larger scale, something of this sort might succeed, but I bring it forward here
only as an illustration. We turn on the electric current, and bring the
magnesiums together. You see a brilliant green light, indicating the vaporisa-
1895] ARGON. 197
tion of the magnesium. Under the influence of the heat the magnesium
burns, and there is collected in the glass vessel a certain amount of brownish-
looking powder which consists mainly of the nitride of magnesium. Of course,
if there is any oxygen present it has the preference, and the ordinary white
oxide of magnesium is formed.
The gas thus isolated is proved to be inert by the very fact of its
isolation. It refuses to combine under circumstances in which nitrogen, itself
always considered very inert, does combine — both in the case of the oxygen
treatment and in the case of the magnesium treatment ; and these facts are,
perhaps, almost enough to justify the name which we have suggested for it.
But, in addition to this, it has been proved to be inert under a considerable
variety of other conditions such as might have been expected to tempt it into
combination. I will not recapitulate all the experiments which have been
tried, almost entirely by Professor Ramsay, to induce the gas to combine.
Hitherto, in our hands, it has not done so ; and I may mention that recently,
since the publication of the abstract of our paper read before the Royal
Society, argon has been submitted to the action of titanium at a red heat,
titanium being a metal having a great affinity for nitrogen, and that argon
has resisted the temptation to which nitrogen succumbs. We never have
asserted, and we do not now assert, that argon can under no circumstances be
got to combine. That would, indeed, be a rash assertion for any one to
venture upon ; and only within the last few weeks there has been a most
interesting announcement by M. Berthelot, of Paris, that, under the action of
the silent electric discharge, argon can be absorbed when treated in contact
with the vapour of benzine. Such a statement, coming from so great an
authority, commands our attention; and if we accept the conclusion, as I
suppose we must do, it will follow that argon has, under those circumstances,
combined.
Argon is rather freely soluble in water. That is a thing that troubled us
at first in trying to isolate the gas ; because, when one was dealing with very
small quantities, it seemed to be always disappearing. In trying to accumulate
it we made no progress. After a sufficient quantity had been prepared, special
experiments were made on the solubility of argon in water. It has been
found that argon, prepared both by the magnesium method and by the oxygen
method, has about the same solubility in water as oxygen— some two-and-a-
half times the solubility of nitrogen. This suggests, what has been verified by
experiment, that the dissolved gases of water should contain a larger propor-
tion of argon than does atmospheric nitrogen. I have here an apparatus of a
somewhat rough description, which I have employed in experiments of this
kind. The boiler employed consists of an old oil-can. The water is supplied to
it and drawn from it by coaxial tubes of metal. The incoming cold water flows
through the outer annulus between the two tubes. The outgoing hot water
198 ARGON. [215
passes through the inner tube, which ends in the interior of the vessel at a
higher level. By means of this arrangement the heat of the water which has
done its work is passed on to the incoming water not yet in operation, and in
that way a limited amount of heat is made to bring up to the boil a very
much larger quantity of water than would otherwise be possible, the greater
part of the dissolved gases being liberated at the same time. These are
collected in the ordinary way. What you see in this flask is dissolved air
collected out of water in the course of the last three or four hours. Such gas,
when treated as if it were atmospheric nitrogen, that is to say after removal
of the oxygen and minor impurities, is found to be decidedly heavier than
atmospheric nitrogen to such an extent as to indicate that the proportion of
argon contained is about double. It is obvious, therefore, that the dissolved
gases of water form a convenient source of argon, by which some of the labour
of separation from air is obviated. During the last few weeks I have been
supplied from Manchester by Mr Macdougall, who has interested himself in
this matter, with a quantity of dissolved gases obtained from the condensing
water of his steam engine.
As to the spectrum, we have been indebted from the first to Mr Crookes,
and he has been good enough to-night to bring some tubes which he will
operate, and which will show you at all events the light of the electric
discharge in argon. I cannot show you the spectrum of argon, for unfortunately
the amount of light from a vacuum tube is not sufficient for the projection of
its spectrum. Under some circumstances the light is red, and under other
circumstances it is blue. Of course when these lights are examined with the
spectroscope — and they have been examined by Mr Crookes with great care —
the differences in the colour of the light translate themselves into different
groups of spectrum lines. We have before us Mr Crookes' map, showing the
two spectra upon a very large scale. The upper is the spectrum of the blue
light ; the lower is the spectrum of the red light ; and it will be seen that
they differ very greatly. Some lines are common to both ; but a great many
lines are seen only in the red, and others are seen only in the blue. It is
astonishing to notice what trifling changes in the conditions of the discharge
bring about such extensive alterations in the spectrum.
One question of great importance upon which the spectrum throws light
is, Is the argon derived by the oxygen method really the same as the argon
derived by the magnesium method ? By Mr Crookes' kindness I have had an
opportunity of examining the spectra of the two gases side by side, and such
examination as I could make revealed no difference whatever in the two
spectra, from which, I suppose, we may conclude either that the gases are
absolutely the same, or, if they are not the same, that at any rate the
ingredients by which they differ cannot be present in more than a small
proportion in either of them.
1895]
ARGON.
199
My own observations upon the spectrum have been made principally at
atmospheric pressure. In the ordinary process of sparking, the pressure is
atmospheric ; and, if we wish to look at the spectrum, we have nothing more
to do than to include a jar in the circuit, and to put a direct-vision prism to
the eye. At my request, Professor Schuster examined some tubes containing
argon at atmospheric pressure prepared by the oxygen method, .and I have
here a diagram of a characteristic group. He also placed upon the sketch
some of the lines of zinc, which were very convenient as directing one exactly
where to look. See figure.
43
44
45
46
47
48
5000
Argon
-^Red
Zinc
1
Hydrogen
Within the last few days, Mr Crookes has charged a radiometer with
argon. When held in the light from the electric lamp, the vanes revolve
rapidly. Argon is anomalous in many respects, but not, you see, in this.
Next, as to the density of argon. Professor Ramsay has made numerous
and careful observations upon the density of the gas prepared by the mag-
nesium method, and he finds a density of about 19'9 as compared with
hydrogen. Equally satisfactory observations upon the gas derived by the
oxygen method have not yet been made*, but there is no reason to suppose
that the density is different, such numbers as 19'7 having been obtained.
One of the most interesting matters in connection with argon, however, is
what is known as the ratio of the specific heats. I must not stay to elaborate
the questions involved, but it will be known to many who hear me that the
velocity of sound in a gas depends upon the ratio of two specific heats — the
specific heat of the gas measured at constant pressure, and the specific heat
measured at constant volume. If we know the density of a gas, and also the
velocity of sound in it, we are in a position to infer this ratio of specific heats ;
and by means of this method, Professor Ramsay has determined the ratio in
the case of argon, arriving at the very remarkable result that the ratio of
* [See Proc. Roy. Soc. Vol. LIX. p. 198, 1896.]
200 ARGON. [215
specific heats is represented by the number T65, approaching very closely to
the theoretical limit, 1'67. The number 1/67 would indicate that the gas has
no energy except energy of translation of its molecules. If there is any other
energy than that, it would show itself by this number dropping below T67.
Ordinary gases, oxygen, nitrogen, hydrogen, &c., do drop below, giving the
number 1*4. Other gases drop lower still. If the ratio of specific heats is
1*65, practically T67, we may infer that the whole energy of motion is trans-
lational ; and from that it would seem to follow by arguments which, however,
I must not stop to elaborate, that the gas must be of the kind called by
chemists monatomic.
I had intended to say something of the operation of determining the ratio
of specific heats, but time will not allow. The result is, no doubt, very
awkward. Indeed, I have seen some indications that the anomalous properties
of argon are brought as a kind of accusation against us. But we had the very
best intentions in the matter. The facts were too much for us ; and all that
we can do now is to apologise for ourselves and for the gas.
Several questions may be asked, upon which I should like to say a word or
two, if you will allow me to detain you a little longer. The first question (I do
not know whether I need ask it) is, Have we got hold of a new gas at all ?
I had thought that that might be passed over, but only this morning I read in
a technical journal the suggestion that argon was our old friend nitrous oxide.
Nitrous oxide has roughly the density of argon ; but that, so far as I can see,
is the only point of resemblance between them.
Well, supposing that there is a new gas, which I will not stop to discuss,
because I think that the spectrum alone would be enough to prove it, the
next question that may be asked is, Is it in the atmosphere ? This matter
naturally engaged our earnest attention at an early stage of the enquiry. I
will only indicate in a few words the arguments which seem to us to show
that the answer must be in the affirmative.
In the first place, if argon be not in the atmosphere, the original
discrepancy of densities which formed the starting-point of the investigation
remains unexplained, and the discovery of the new gas has been made upon a
false clue. Passing over that, we have the evidence from the blank experi-
ments, in which nitrogen originally derived from chemical sources is treated
either with oxygen or with magnesium, exactly as atmospheric nitrogen is
treated. If we use atmospheric nitrogen, we get a certain proportion of argon,
about 1 per cent. If we treat chemical nitrogen in the same way we get, I
will not say absolutely nothing, but a mere fraction of what we should get had
atmospheric nitrogen been the subject. You may ask, Why do we get any
fraction at all from chemical nitrogen ? It is not difficult to explain the small
residue, because in the manipulation of the gases large quantities of water are
1895] ARGON. 201
used ; and, as I have already explained, water dissolves argon somewhat freely.
In the processes of manipulation some of the argon will come out of solution,
and it remains after all the nitrogen has been consumed.
Another wholly distinct argument is founded upon the method of diffusion
introduced by Graham. Graham showed that if you pass gas along porous
tubes you alter the composition, if the gas is a mixture. The lighter con-
stituents go more readily through the pores than do the heavier ones. The
experiment takes this form. A number of tobacco pipes — eight in the actual
arrangement — are joined together in series with indiarubber junctions, and
they are put in a space in which a vacuum can be made, so that the space
outside the porous pipes is vacuous or approximately so. Through the pipes
ordinary air is led. One end may be regarded as open to the atmosphere.
The other end is connected with an aspirator so arranged that the gas collected
is only some 2 per cent, of that which leaks through the porosities. The case
is like that of an Australian river drying up almost to nothing in the course
of its flow. Well, if we treat air in that way, collecting only the small residue
which is less willing than the remainder to penetrate the porous walls, and
then prepare " nitrogen " from it by removal of oxygen and moisture, we
obtain a gas heavier than atmospheric nitrogen, a result which proves that the
ordinary nitrogen of the atmosphere is not a simple body, but is capable of
being divided into parts by so simple an agent as the tobacco pipe.
If it be admitted that the gas is in the atmosphere, the further question
arises as to its nature.
At this point I would wish to say a word of explanation. Neither in our
original announcement at Oxford, nor at any time since, until the 31st of
January, did we utter a word suggesting that argon was an element ; and it
was only after the experiments upon the specific heats that we thought that
we had sufficient to go upon in order to make any such suggestion in public.
I will not insist that that observation is absolutely conclusive. It is certainly
strong evidence. But the subject is difficult, and one that has given rise to
some difference of opinion among physicists. At any rate this property dis-
tinguishes argon very sharply from all the ordinary gases.
One question which occurred to us at the earliest stage of the enquiry, as
soon as we knew that the density was not very different from 21, was the
question of whether, possibly, argon could be a more condensed form of
nitrogen, denoted chemically by the symbol N3. There seem to be several
difficulties in the way of this supposition. Would such a constitution be
consistent with the ratio of specific heats (1'65) ? That seems extremely
doubtful. Another question is, Can the density be really as high as 21, the
number required on the supposition of N3 ? As to this matter, Professor
Ramsay has repeated his measurements of density, and he finds that he cannot
202 ARGON. [215
get even so high as 20. To suppose that the density of argon is really 21,
and that it appears to be 20 in consequence of nitrogen still mixed with it,
would be to suppose a contamination with nitrogen oiit of all proportion to
what is probable. It would mean some 14 per cent, of nitrogen, whereas it
seems that from one-and-a-half to two per cent, is easily enough detected by
the spectroscope. Another question that may be asked is, Would N8 require
so much cooling to condense it as argon requires ?
There is one other matter on which I would like to say a word — the
question as to what N3 would be like if we had it. There seems to be a
great discrepancy of opinions. Some high authorities, among whom must be
included, I see, the celebrated Mendeleef, consider that N3 would be an
exceptionally stable body; but most of the chemists with whom I have
consulted are of opinion that N3 would be explosive, or, at any rate, absolutely
unstable. That is a question which may be left for the future to decide. We
must not attempt to put these matters too positively. The balance of evidence
still seems to be against the supposition that argon is N3, but for my part I
do not wish to dogmatise.
A few weeks ago we had an eloquent lecture from Professor Riicker on the
life and work of the illustrious Helmholtz. It will be known to many that
during the last few months of his life Helmholtz lay prostrate in a semi-
paralysed condition, forgetful of many things, but still retaining a keen
interest in science. Some little while after his death we had a letter from
his widow, in which she described how interested he had been in our
preliminary announcement at Oxford upon this subject, and how he desired
the account of it to be read to him over again. He added the remark, " I
always thought that there must be something more in the atmosphere."
216.
ON THE STABILITY OR INSTABILITY OF CERTAIN
FLUID MOTIONS. III.*
[Proceedings of the London Mathematical Society, xxvu. pp. 5 — 12, 1895.]
THE steady motions in question are those in which the velocity is parallel
to a fixed line (#), and such that U is a function of y only. In the disturbed
motion U + u, v, the infinitely small quantities u, v are supposed to be periodic
functions of x, proportional to eikx, and, as dependent upon the time, to be
proportional to eint, where n is a constant, real or imaginary. Under these
circumstances the equation determining v is
The vorticity (Z) of the steady motion is ^dU/dy. If throughout any layer Z
be constant, d*U/dy* vanishes, and, whenever n + kU does not also vanish,
d*v/dy*-k*v = Q, (2)
or v = Aeky + Be~ky (3)
If there are several layers in each of which Z is constant, the various solutions
of the form (3) are to be fitted together, the arbitrary constants being so
chosen as to satisfy certain boundary conditions. The first of these conditions
is evidently
A« = 0 (4)f
The second may be obtained by integrating (1) across the boundary. Thus
(- U\ &(dv}-&(—} -0 C5)
\k+ )' \dy)~ \dy)'1
* The two earlier papers upon this subject are to be found in Proc. Lond. Math. Soc. Vol. xi.
p. 57, 1880 [Vol. i. p. 474]; Vol. xix. p. 67, 1887 [Vol. in. p. 17]. The fluid is supposed to be
destitute of viscosity.
t [A being the symbol of finite differences.]
204 ON THE STABILITY OR INSTABILITY [216
At a fixed wall v = 0.
Equation (2) secures that the vorticity shall remain constant in each layer,
and equation (3) that there shall be no slipping at the surface of transition.
Equations (2) and (3) together may be regarded as expressing the continuity
of the motion at the surface between the layers.
In the first of the papers above referred to, I have applied equation (1) to
prove that, if d?Ujdy2 be of one sign throughout the whole interval between
two fixed walls, n can have no imaginary part. It is true that, if n+kU
vanishes anywhere, the expression for d^vjdy^ — J^v in (1) becomes infinite,
unless indeed v = 0 at the place in question; and Lord Kelvin* considers that
the "disturbing infinity" thus introduced vitiates the proof of stability. To
this criticism it may be replied f that, " if n be complex, there is no disturbing
infinity, and that, therefore, the argument does not fail, regarded as one for
excluding complex values of n. What happens when n has a real value,
such that n + k U vanishes at an interior point, is a subject for further
consideration."
In embarking upon this it will be convenient to take first the case of (2),
(3), (4), (5), where the vorticity of the steady motion is uniform through
layers of finite thickness. Any general conclusions arrived at in this way
should at least throw light upon the extreme case where the number of the
layers is infinitely great, and their thickness is infinitely small.
Starting from the first wall at y = 0, let the surfaces between the layers
occur at y = yi, y = y*, &c., and let the values of U at these places be Ul,
Uz, &c. In conformity with (4) and with the condition that v = 0, when
y = 0, we may take in the first layer
y = vt = MI sinh ky ; (6)
in the second layer
v = v2 = vl + Mz sinh k(y- y,}; (7)
in the third layer
v = v3 = v2 + M3 sinh k (y — y2) ; (8)
and so onj.
If the second fixed wall be in the rth layer at y = y', then
M1 sinh ky'-+ 3fs-sinh k (y' — y^) + . . . + Mr sinh k (y — yr-^) = 0. . . .(9)
We have still to express the conditions (5) at the various surfaces of transition.
At the first surface
v = Ml sirih % , A (dv / dy) = kM2;
* Phil. Mag. Vol. xxiv. p. 275, 1887.
t PMl. Mag. Vol. xxxiv. p. 66, 1892. [Vol. in. p. 580.]
J This is the process followed iu the second of the papers cited, with a slight difference of
notation.
1895] OF CERTAIN FLUID MOTIONS. 205
at the second surface
v = 3/! sink ki/2 + Mz sinh k (y2 — y^), A (dv / dy) = kM3 ;
and so on. If we denote the values of A (dll/dy) at the various surfaces by
Ax, A2, &c., the conditions may be written
(n + k tfj) M 2 - Aj . Ml sinh kyl = 0
(n + k U2) M3 - A2 . [Ml sinh ky.2 + M,2 sinh k (yz - yt}} = 0 [ . . . .(10)
The r — 1 equations (10) together with (9) suffice to determine n, and the
r — l ratios Ml: Mz : M3: ... : Mr. The determinantal equation in n is of
degree r — 1 , the number of the surfaces of transition ; and corresponding to
each root there is an expression for v, definite except as regards a constant
multiplier.
It is important to note that the disturbances thus expressed are such as
leave the vorticity unaltered in the interior of every layer ; that they relate, in
fact, merely to waves upon the surfaces of transition. The additional vorticity
due to the disturbance is proportional to d*v/dy* — k2v, and is equated to zero
in (2). If we wish to consider the most general disturbance possible, we must
provide for an arbitrary vorticity at every point.
The nature of the normal modes of disturbance not yet considered will be
apparent from a comparison between (1) and (2). Even though d-U [dy* = 0,
the latter does not follow from the former, unless it be assumed that n + k U
is finite. Wherever n + kU vanishes, that is, at the places where the wave
velocity is equal to the stream velocity, (1) is satisfied, even though (2) be
violated. Thus any value of — kU to be found anywhere in the fluid is an
admissible value of n, and the corresponding normal function (v) is obtained
by allowing the arbitrary constants in (3) to be discontinuous at this place as
well as at the surfaces of transition, subject of course to the condition that v
itself shall be continuous. The new arbitrary constant thus disposable allows
all the conditions to be satisfied with the value of n already prescribed.
The equations (9), (10) already found suffice for the present purpose if we
introduce a fictitious surface of transition at the place in question. Suppose,
for example, that A3 = 0 in the third of equations (10). It will follow either
that MI = 0, or that n + kU3 = 0. In the first alternative the constants A and
B are continuous, and all local peculiarity disappears. The second alternative
is the one with which we are now concerned. The equations suffice, as usual,
to determine n (equal to —kU,), as well as the ratios of the M's which give
the form of the normal function. The mode of disturbance is such that a new
vorticity is introduced at the place, or rather at the plane in question. In
one sense this is the only new vorticity ; but the waves upon the surfaces of
transition involve changes of vorticity as regards given positions in space,
though not as regards given portions of fluid.
206 ON THE STABILITY OR INSTABILITY [216
We have now to consider what occurs at a second place in the fluid where
the velocity happens to be the same as at the first place. The second place
may be either within a layer of originally uniform vorticity or upon a surface
of transition. In the first case nothing very special presents itself. If there
be no new vorticity at the second place, the value of v is definite as usual, save
as to an arbitrary multiplier. But, consistently with the given value of n, there
may be new vorticity at the second as well as at the first place, and then the
complete value of v for the given n may be regarded as composed of two parts,
each proportional to one of the new vorticities, and each affected by an
arbitrary multiplier.
If the second place lie upon a surface of transition, we have a state of
things corresponding to the " disturbing infinity " in (1). In the above
example, where A3 = 0, n + k U3 = 0, we have now further to suppose that Ul ,
the velocity at the first surface of transition, coincides with U3. From the
first of equations (10), since n + kUl = 0, while Ax and sinh ky^ are finite, we
see that Ml must vanish. Hence v = 0 throughout the entire layer from the
wall y = 0 to y = yr. The remainder of the motion from y = yt to y — y1 is to
be determined as usual.
From the fact that v = 0, we might be tempted to infer that the surface in
question behaves like a fixed wall. But a closer examination shows that the
inference would be unwarranted. In order to understand this it may be well
to investigate the relation between v and the displacement of the surface,
supposed also to be proportional to eint . e***. Thus, if the equation of the
surface be
F = y_heint+** = 0! ........................... (11)
the condition to be satisfied is*
so that -ih(n + kU1)+v = Q ........................... (13)
is the required relation. Using this, we see from the first of equations (10)
that h does not vanish, but is given by
The propagation of a wave at the same velocity as that at which the fluid
moves does not entail the existence of a finite velocity v.
That v vanishes at a surface of transition where n + kU=0 follows in
general from (5), seeing that the value of A (dU/dy) is finite. That region of
* Lamb's Hydrodynamics, § 10.
1895] OF CERTAIN FLUID MOTIONS. 207
the fluid, bounded by this surface and one of the fixed walls, which does not
include the added vorticity, will in. general remain undisturbed, but there may
be exceptions when one of the values of n proper to this region (regarded as
bounded by fixed walls) happens to coincide with that prescribed. It does
not appear that the infinity which enters when n + kU=0 disturbs any
general conclusions as to the conditions of stability, or even seriously modifies
the character of the solutions themselves.
When d?U/dy* is finite, we must fall back upon equation (1). The
character of the disturbing infinity at a place (say, y = 0) where n + kU
vanishes would be most satisfactorily investigated by means of the complete
solution of some particular case. It is, however, sufficient to examine the
form of solution in the neighbourhood of y = 0, and for this purpose the
differential equation may be simplified. Thus, when y is small, n + kU may
be regarded as proportional to y, and d2U/dy2 as approximately constant. In
comparison with the large term, kzv may be neglected, and it suffices to
consider
= 0, ........................... (15)
a known constant multiplying y being omitted for brevity. This falls under
the head of Riccati's equation
d*v/dy2 + y*v = Q, .......................... (16)
of which the solution* is in general (m fractional)
v = Jy{AJm(Z) + BJ_m(Z)}> ..................... (17)
where m = I f(p + 2), £=2ra#1/!™ ................... (18)
When, as in the present case, m is integral, J-m(%) is to be replaced by
the function of the second kind Ym(j;). The general solution of (15) is
accordingly
(19)
In passing through zero y changes sign, and with it the character of the
functions. If we regard (19) as applicable on the positive side, then on the
negative side we may write
v = </y{CJl(2Jy) + DY1(Wy)}, .................. (20)
the arguments of the functions in (20) being pure imaginaries.
The functions Ji(z), Fi(*) are given by
* Lommel, Studien tiber die BesscVschen Functionen, § 31, Leipzig, 1868; Gray and Mathews'
Bessel s Functions, p. 233, 1895.
208 ON THE STABILITY OR INSTABILITY [216
where Sm= 1 +£ + £+ ... + 1/m ...................... (23)
When y is small, (19) gives
» = 4 (y - iy2} + B {i (l - y + iy) - log (2 Vy) (y - *y) + y$ - iytf,} ;• • -(24)
so that ultimately
v = ^B, dv/dy = A-±B\ogy, d2v/dya- = - A -%By-\ ...(25)
v remaining finite in any case.
We will now show that any value of — kU is an admissible value of n in
(1). The place where n+kU=Q is taken as origin of y; and in the first
instance we will suppose that n + k U vanishes nowhere else. In the immediate
neighbourhood of y = 0, the solutions applicable on the two sides are (19), (20),
and they are subject to the condition that v shall be continuous. Hence, by
(25), B — D, leaving three constants arbitrary. The manner in which the
functions start from y = 0 being thus ascertained, their further progress is
subject to the original equation (1), which completely defines them when the
three arbitraries are known. In the present case two relations are given by
the conditions to be satisfied at the fixed walls or other boundaries of the
fluid, and thus is determined the entire form of v, save as to a constant
multiplier. If, as must usually be the case, B and D are finite, there is
infinite vorticity at the origin, but this is no more than occurs even when
d*U/dif is zero throughout the region surrounding the origin.
Any other places at which n + k U = 0 may be treated in a similar manner,
and the most general solution will contain as many arbitrary constants as
there are places of infinite vorticity. But the vorticity need not be infinite
merely because n + k U = 0 ; and, in fact, a particular solution may be
obtained with only one infinite vorticity. At any other of the critical places,
such, for example, as we may now suppose the origin to be, B and D may
vanish, so that
0 = 0.
From this discussion it would seem that the infinities which present them-
selves when n + kU=-Q do not seriously interfere with the application of the
general theory, so long as the square of the disturbance from steady motion
is neglected. The value of conclusions relating only to infinitely small
disturbances is another question.
1895] OF CERTAIN FLUID MOTIONS. 209
When regard is paid to viscosity, the difficulties are of course much
increased. In the particular case where the original vorticity is uniform, the
problem of small disturbances has been solved by Lord Kelvin*, who shows
that the motion is stable by the aid of a special solution not proportional to a
simple exponential function of the time. If we retain the supposition of the
present paper that the disturbance as a function of the time is proportional to
eint, we obtain an equation [(52) in Lord Kelvin's paper] which has been
discussed by Stokes f. From his results it appears that it is not possible to
find a solution applicable to an unlimited fluid which shall be periodic with
respect to x, and remain finite when y = + oo , and this whether n be real or
complex. The cause of the failure would appear to lie in the fact, indicated
by Lord Kelvin's solution, that the stability is ultimately of a higher order
than can be expressed by any simple exponential function of the time.
[Addendum, January, 1896.— It may be well to emphasise more fully that
the solutions of this paper only profess to apply in the limit, when the dis-
turbances are infinitely small The constant factor which represents the scale
of the disturbance must be imagined to be so small that the actual disturbance
nowhere rises to such a magnitude as to interfere with the approximations upon
which (1) is founded. For example, in (25), although dv/dy is infinite at
y = 0 relatively to its value at other places, it must still be regarded as
infinitely small throughout in comparison with the quantities which define the
steady motion.]
* Phil. Mag. Vol. xxiv. p. 191, 1887.
t Camb. Phil. Trans. Vol. x. p. 105, 1857.
217.
ON THE PROPAGATION OF WAVES UPON THE PLANE
SURFACE SEPARATING TWO PORTIONS OF FLUID OF
DIFFERENT VORTICITIES.
[Proceedings of the London Mathematical Society, xxvu. pp. 13 — 18, 1895.]
IN former papers* I have considered the problem of the motion in two
dimensions of in viscid incompressible fluid between two parallel walls. In the
case where the steady motion is such that in each half of the layer included
between the walls the vorticity is constant, it appeared that the motion is
stable, small displacements of the surface separating the two vorticities being
propagated as waves of constant amplitude. More particularly, if the velocity
of the steady motion increase uniformly from zero at the walls to the value U
in the middle stratum, a disturbance proportional to ei(nt+kx) requires that
n + kU= U/b.teohkb, (1)
where 26 is the distance between the walls. The wave-length is 2ir/k, and
the fact that n is real indicates that the disturbance is stable.
Discussions upon the difficult question of the nature of the instability
manifested by fluids in their flow through pipes of moderate bore seemed to
make it desirable to push the investigation of the disturbance from some
simple case of steady motion so far at least as to include the squares of the
small quantities.
In the present paper the problem chosen for the purpose is that above
referred to, simplified by excluding the fixed walls, or, what comes to the
same thing, by supposing them removed to a distance very great in comparison
with the wave-length of the disturbance. We suppose, then, that in the
steady motion the surface of separation coincides with y = 0, that when y is
positive the vorticity is + to, and that when y is negative the vorticity is — CD.
* " On the Stability or Instability of certain Fluid Motions," Proc. Lond. Math. Soc. Vol. xi.
p. 57, 1880 [Vol. i. p. 474]; Vol. xix. p. 67, 1887 [Vol. in. p. 17].
1895] ON THE PROPAGATION OF WAVES UPON A PLANE SURFACE. 211
In the disturbed motion the surface separating the two vorticities is displaced,
so that its equation becomes y = h cos x, k being put equal to unity for the
sake of brevity.
In virtue of the incompressibility, the component velocities, denoted as
usual by u and v, are connected with a stream-function -^ by the relations
(2)
The vorticity is represented by £V2i/r, which is accordingly equal to + «.
During the steady motion of the upper fluid, we have
^ = a + (3y + nf ............................... (3)
In consequence of the disturbance ^ deviates from the value given by (3) ;
but, since, by a known theorem, the vorticity remains throughout equal to to,
the addition to i/r must satisfy V2i|r = 0. The additional terms must also
satisfy the condition of being periodic in period 2?r; and thus we obtain
altogether as the expression for \jr during the disturbed motion
i/r = a + fty + my2 + e~v (A1 cos x + B^ sin x)
+ e-2^,, cos 2# + £2 sin 2#) + ..., .................. (4)
positive exponents being excluded by the condition to be satisfied when
y = + oo . Similarly in the lower fluid
-»/r' = a + fly — my* + ey (-4/ cos x + B± sin x)
+ e*y (A3f cos 2x + B2' sin 2a?) + ...................... (5)
From these values of ^r, i/r' the velocities u, v at any point are deducible
by (2).
We have still to satisfy the conditions at the surface of separation
y = h cosx .................................. (6)
It is necessary that u and v, as given by i/r and ty', should there be continuous,
any sliding of the one body of fluid upon the other being equivalent to a
vortex-sheet, and therefore excluded by the conditions of the problem. Thus
at the surface we must have
d(^-^')/dx = 0, d(^-^')/dy = 0 ................ (7)
For the purposes of the first approximation, where only the first power of h is
retained, y may be put equal to zero in the exponential terms so soon as the
differentiations have been performed. Equations (7) give accordingly
— sin x (Al — A/) + cos x (B1 — B^)
- 2 sin 2x(A2-A,')+ 2 cos 2x(B2-B*)- ............ =0,
/3 - £' + 4>a>h cos x - cos x (A^ + A^} - sin x (B^ + 1?,')
2') -2 sin 2x(B2 + B2')- ............ = 0;
14—2
212 ON THE PROPAGATION OF WAVES UPON THE PLANE SURFACE [217
from which it appears that to this approximation all the coefficients with
suffixes higher than unity must vanish. Also
Thus ty = a + fiy + (oy2 + 2(0he-ycosa;, ..................... (8)
^r' — a' + fiy — ay* + 2wA ev cos x, ..................... (9)
are the values of ^ determined in accordance with (6) and the other prescribed
conditions. From (8) or (9), we find as the values of u and v at the surface
u = ft, v=2a)hsina;, ........................... (10)
applicable when the form of the surface is that given by (6), at the moment,
we may suppose, when t = 0.
By means of (10) it is possible to determine the form and position of the
surface of separation at time dt, and thus to trace out its transformation. In
the present case it will be simplest merely to verify that the propagation of
the form (6) with a certain velocity ( V) satisfies all the conditions. If
F(x,y,t) = y-hcoa(x- Vt) = 0 .................. (11)
be the equation of the surface, the condition to be satisfied* is
Here, when t = 0,
dF dF dF
-j- = — Vh sin x, -j- = ft sin #, =-
dt dx dy
so that (12) becomes, with use of (10),
showing that (11) continues to represent the surface of separation at time dt,
provided that
(13)
Accordingly, if (13) be satisfied, equation (11) suffices to represent the
changes in the surface of separation for any length of time, or, in other words,
the disturbance is propagated as a simple wave.
From (8) it appears that /3 represents the velocity in the steady motion
when y = 0, and the result is in accordance with (1), where tanh kb = l. The
disturbance may be supposed to be got rid of by the introduction of a flexible
lamina at the surface of separation. If, by forces applied to it, the lamina be
straightened out so as to coincide with y = 0, and be held there at rest, the
steady motion is recovered.
* Lamb's Hydrodynamics, § 10.
1895] SEPARATING TWO PORTIONS OF FLUID OF DIFFERENT VORTlClTlES. 213
In proceeding to further approximations, in which higher powers of h are
retained, it appears either from the equations, or immediately from the
symmetries involved, that all the B's vanish, so that cosines only occur in (4)
and (5), that
AI = AI, A3 = A3, A5 = Aa, 6LC. ',
A2' = — A2, A4' = - A4, &c.;
and further that ft' = (3. Equations (4) and (5) may thus be written
•^r = a + @y 4- a>?/2 + Ate~v cos x + A2e~^ cos 2a? + A,er^ cos 3#+ . . ., . . .(14)
.......... (15)
A! is of order h, A2 of order A2, A3 of order h3, and so on. If we are content to
neglect h6, we may stop at A5; and we find as the equations necessary in order
to secure the continuity of u and v at the surface (6)
^ (2 + T + n) - « + ^ (2i + T) - 34> r -
2^2 (2 + 2A2) =
From these equations the values of the constants may be determined by
successive approximations. Thus, if we retain terms of the order A,2, A, , A4 , &c.,
vanish and
This is the second approximation. The fifth approximation gives
(18,19,20)
, l = , .
which values are to be substituted in (14), (15).
The next step is the investigation of the values of u, v at the surface (6).
They are most conveniently expressed as
214 ON THE PROPAGATION OF WAVES UPON A PLANE SURFACE. [217
We get, correct as far as A5,
(21)
(22)
the terms containing cos 4# in (21), and sin 5x in (22), vanishing to this order.
If we substitute these values in (12), we obtain
h sin x {- V + £ + 2o> + £o>A2 - ^&>A4} + ^o>A5 sin 3x = 0. . . .(23)
So far, then, as terms in h*, the surface of separation (6) is propagated as a
simple wave with velocity given by
V = j3 + 2co + $coh2; ........................... (24)
but, if terms in A6 are retained, a change of form manifests itself, corresponding
to the term in wh5 sin 3# outstanding in (23).
Hitherto the wave-length has been supposed to be 2?r, but, if we now take
it to be 27T/&, (24) becomes
V = /3 + 2(y / k . (1 + £W), .................... (25)
as is evident by " dimensions." The velocity of propagation is that of the
flow of the fluid in the steady motion at the place where
ky = l + ±kW ............................... (26)
So far as the present investigation can reach, there is no sign of the
amplitude of a wave tending spontaneously to increase.
218.
ON SOME PHYSICAL PROPERTIES OF ARGON AND HELIUM*.
[Proceedings of the Royal Society, LIX. pp. 198—208, Jan. 1896.]
Density of Argon.
IN our original paper-f- are described determinations by Prof. Ramsay, of
the density of argon prepared with the aid of magnesium. The volume
actually weighed was 163c.c.,and the adopted mean result was 19'941, referred
to O2= 16. At that time a satisfactory conclusion as to the density of argon
prepared by the oxygen method of Cavendish had not been reached, although
a preliminary result (197) obtained from a mixture of argon and oxygen J
went far to show that the densities of the gases prepared by the two methods
were the same. In order further to test the identity of the gases, it was
thought desirable to pursue the question of density ; and I determined, as
the event proved, somewhat rashly, to attempt large scale weighings of pure
argon with the globe of 1800 c.c. capacity employed in former weighings
of gases 1 1 which could be obtained in quantity.
The accumulation of the 3 litres of argon, required for convenient working,
involved the absorption of some 300 litres of nitrogen, or about 800 litres of
the mixture with oxygen. This was effected at the Royal Institution with
the apparatus already described §, and which is capable of absorbing the
mixture at the rate of about 7 litres per hour. The operations extended
themselves over nearly three weeks, after which the residual gases amounting
to about 10 litres, still containing oxygen with a considerable quantity of
* [Some of the results here given were announced before the British Association at the Ipswich
meeting. See Report, Sept. 13, 1895.]
t Eayleigh and Bamsay, Phil. Trans. A, Vol. CLXXXVI. pp. 221, 238, 1895. [Vol. iv. p. 130.]
J Loc. cit. p. 221. [Vol. iv. p. 165.]
I! Roy. Soc. Proc. February, 1888 [Vol. m. p. 37] ; February, 1892 [Vol. in. p. 534] ; March,
1893 [Vol. iv. p. 39].
§ Phil. Trans, loc. cit. p. 219. [Vol. iv. p. 162.]
216 ON SOME PHYSICAL PROPERTIES OF ARGON AND HELIUM. [218
nitrogen, were removed to the country and transferred to a special apparatus
where it could be prepared for weighing.
For this purpose the purifying vessel had to be arranged somewhat
differently from that employed in the preliminary absorption of nitrogen.
When the gas is withdrawn for weighing, the space left vacant must be filled
up with liquid, and afterwards, when the gas is brought back for repurification,
the liquid must be removed. In order to effect this, the working vessel
(Fig. 7)* communicates by means of a siphon with a 10-litre "aspirating
bottle," the ends of the siphon being situated in both cases near the bottom
of the liquid. In this way the alkaline solution may be made to pass back-
wards and forwards, in correspondence with the desired displacements of gas.
There is, however, one objection to this arrangement which requires to be
met. If the reserve alkali in the aspirating bottle were allowed to come into
contact with air, it would inevitably dissolve nitrogen, and this nitrogen would
be partially liberated again in the working vessel, and so render impossible
a complete elimination of that gas from the mixture of argon and oxygen.
By means of two more aspirating bottles an atmosphere of oxygen was main-
tained in the first bottle, and the outermost bottle, connected with the second
by a rubber hose, gave the necessary control over the pressure.
Five glass tubes in all were carried through the large rubber cork by
which the neck of the working vessel was closed. Two of these convey the
electrodes: one is the siphon for the supply of alkali, while the fourth and
fifth are for the withdrawal and introduction of the gas, the former being
bent up internally, so as to allow almost the whole of the gaseous contents
to be removed. The fifth tube, by which the gas is returned, communicates
with the fall-tube of the Topler pump, provision being made for the overflow
of mercury. In this way the gas, after weighing, could be returned to the
working vessel at the same time that the globe was exhausted. It would be
tedious to describe in detail the minor arrangements. Advantage was fre-
quently taken of the fact that oxygen could always be added with impunity,
its presence in the working vessel being a necessity in any case.
When the nitrogen had been so far removed that it was thought desirable
to execute a weighing, the gas on its way to the globe had to be freed from
oxygen and moisture. The purifying tubes contained copper and copper
oxide maintained at a red heat, caustic soda, and phosphoric anhydride.
In all other respects the arrangements were as described in the memoir on
the densities of the principal gases f, the weighing globe being filled at 0°,
and at the pressure of the manometer gauge.
The process of purification with the means at my command proved to be
* Phil. Trans, loc. cit. p. 218. [Vol. iv. p. 163.]
t Roy. Soc. Proc. Vol. LIII. p. 134, 1893. [Vol. iv. p. 39.]
1896] ON SOME PHYSICAL PROPERTIES OF ARGON AND HELIUM. 217
extremely slow. The gas contained more nitrogen than had been expected,
and the contraction went on from day to day until I almost despaired of
reaching a conclusion. But at last the visible contraction ceased, and soon
afterwards the yellow line of nitrogen disappeared from the spectrum of the
jar discharge*. After a little more sparking, a satisfactory weighing was
obtained on May 22, 1895 ; but, in attempting to repeat, a breakage occurred,
by which a litre of air entered, and the whole process of purification had to
be re-commenced. The object in view was to effect, if possible, a series of
weighings with intermediate sparkings, so as to obtain evidence that the
purification had really reached a limit. The second attempt was scarcely
more successful, another accident occurring when two weighings only had
been completed. Ultimately a series of four weighings were successfully
executed, from which a satisfactory conclusion can be arrived at.
May 22 3-2710
June 4 3-2617
June 7 . 3-2727
June 13 3-2652
June 18 ...... 3-2750 ]
June 25 3*2748 I 3'2746
July 2 3-2741 )
The results here recorded are derived from the comparison of the weighings
of the globe " full " with the mean of the preceding and following weighings
" empty," and they are corrected for the errors of the weights and for the
shrinkage of the globe when exhausted, as explained in former papers. In
the last series, the experiment of June 13 gave a result already known to be
too low. The gas was accordingly sparked for fourteen hours more. Between
the weighings of June 18 and June 25 there was nine hours' sparking, and
between those of June 25 and July 2 about eight hours' sparking. The mean
of the last three, viz. 3'2746, is taken as the definitive result, and it is
immediately comparable with 2'6276, the weight under similar circumstances
of oxygen -f-. If we take O,2 = 16, we obtain for argon
19-940,
in very close agreement with Professor Ramsay's result.
The conclusion from the spectroscopic evidence that the gases isolated
from the atmosphere by magnesium and by oxygen are essentially the same
is thus confirmed.
* Jan. 29. — When the argon is nearly pure, the arc discharge (no jar connected) assumes
a peculiar purplish colour, quite distinct from the greenish hue apparent while the oxidation of
nitrogen is in progress and from the sky-blue observed when the residue consists mainly of
oxygen.
t Rmj. Soc. Proc. Vol. LIII. p. 144, 1893. [Vol. iv. p. 48.]
218 ON SOME PHYSICAL PROPERTIES OF ARGON AND HELIUM. [218
The Refractivity of Argon and Helium.
The refractivity of argon was next investigated, in the hope that it might
throw some light upon the character of the gas. For this purpose absolute
measurements were not required. It sufficed to compare the pressures
necessary in two columns of air and argon of equal lengths, in order to
balance the retardations undergone by light in traversing them.
The arrangement was a modification of one investigated by Fraunhofer,
depending upon the interference of light transmitted through two parallel
vertical slits placed in front of the object-glass of a telescope. If there be
only one slit, and if the original source, either a distant point or a vertical
line of light, be in focus, the field is of a certain width, due to " diffraction,"
and inversely as the width of the slit. If there be two equal parallel slits
whose distance apart is a considerable multiple of the width of either, the
field is traversed by bands of width inversely as the distance between the
slits. If from any cause one of the portions of light be retarded relatively
to the other, the bands are displaced in the usual manner, and can be brought
back to the original position only by abolishing the relative retardation.
When the object is merely to see the interference bands in full perfection,
the use of a telescope is not required. The function of the telescope is really
to magnify the slit system*, and this is necessary when, as here, it is desired
to operate separately upon the two portions of light. The apparatus is,
however, extremely simple, the principal objection to it being the high
magnifying power required, leading under ordinary arrangements to a great
attenuation of light. I have found that this objection may be almost entirely
overcome by the substitution of cylindrical lenses, magnifying in the hori-
zontal direction only, for the spherical lenses of ordinary eye-pieces. For
many purposes a single lens suffices, but it must be of high power. In the
measurements about to be described most of the magnifying was done
by a lens of home manufacture. It consisted simply of a round rod,
about £in. (4 mm.) in diameter, cut by Mr Gordon from a piece of plate
glass f. This could be used alone ; but as at first it was thought necessary
to have a web, serving as a fixed mark to which the bands could be referred,
the rod was treated as the object-glass of a compound cylindrical microscope,
the eye-piece being a commercial cylindrical lens of 1^ in. (31 mm.) focus.
Both lenses were mounted on adjustable stands, so that the cylindrical axes
could be made accurately vertical, or, rather, accurately parallel to the length
of the original slit. The light from an ordinary paraffin lamp now sufficed,
although the magnification was such as to allow the error of setting to be
* Brit. Assoc. Report, 1893, p. 703. [Vol. iv. p. 76.]
t Preliminary experiments had been made with ordinary glass cane and with tubes charged
with water.
1896] ON SOME PHYSICAL PROPERTIES OF ARGON AND HELIUM. 219
less than 1/20 of a band interval. It is to be remembered that with this
arrangement the various parts of the length of a band correspond, not to the
various parts of the original slit, but rather to the various parts of the object-
glass. This departure from the operation of a spherical eye-piece is an
advantage, inasmuch as optical defects show themselves by deformation of
the bands instead of by a more injurious encroachment upon the distinction
between the dark and bright parts.
Fig. i.
B
The collimating lens A (Fig. 1) is situated 23 ft. (7 metres) from the source
of light. B, C are the tubes, one containing dry air, the other the gas to be
experimented upon. They are 1 ft. (30'5 cm.) long, and of \ in. (1'3 cm.) bore,
and they are closed at the ends with small plates of parallel glass cut from
the same strip. E is the object-glass of the telescope, about 3 in. (7'6 cm.)
in diameter. It is fitted with a cap, D, perforated by two parallel slits. Each
slit is ^ in. (6 mm.) wide, and the distance between the middle lines of the
slits is 1^ in. (38 mm.).
The arrangements for charging the tubes and varying the pressures of
the gases are sketched in Fig. 2. A gas pipette, DE, communicates with the
tube C, so that by motion of the reservoir E and consequent flow of mercury
through the connecting hose, part of the gas may be transferred. The
pressure was measured by a U-shaped manometer F, containing mercury.
This was fitted below with a short length of stout rubber tubing G, to which
was applied a squeezer H. The object of this attachment was to cause
a rise of mercury in both limbs immediately before a reading, and thus to
avoid the capillary errors that would otherwise have entered. A similar-
pipette and manometer were connected with the air-tube B. In order to be
able, if desired, to follow with the eye a particular band during the changes
of pressure (effected by small steps and alternately in the two tubes), diminu-
tive windlasses were provided by which the motions of the reservoirs (E)
could be made smooth and slow. In this way all doubt was obviated as to
the identity of a band ; but after a little experience the precaution was found
to be unnecessary*.
The manner of experimenting will now be evident. By adjustment of
pressures the centre of the middle band was brought to a definite position,
* [For a description of a modified apparatus capable of working with an extremely small
quantity of gas, see Proc. Roy. Soc. Vol. LXIV. p. 97, 1898.]
220 ON SOME PHYSICAL PROPERTIES OF ARGON AND HELIUM. [218
determined by the web or otherwise, and the pressures were measured. Both
pressures were then altered and adjusted until the band was brought back
precisely to its original position. The ratio of the changes of pressure is the
inverse ratio of the refractivities (u. — 1) of the gases. The process may be
repeated backwards and forwards any number of times, so as to eliminate in
great degree errors of the settings -and of the pressure readings.
Fig. 2.
To pump.
Scale
During these observations a curious effect was noticed, made possible
by the independent action of the parts of the object-glass situated at various
levels, as already referred to. When the bands were stationary, they appeared
straight, or nearly so, but when in motion, owing to changes of pressure, they
became curved, even in passing the fiducial position, and always in such
a manner that the ends led. The explanation is readily seen to depend upon
the temporary changes of temperature which accompany compression or
rarefaction. The full effect of a compression, for example, would not be
attained until the gas had cooled back to its normal temperature, and this
recovery of temperature would occur more quickly at the top and bottom,
where the gas is in proximity to the metal, than in the central part of
the tube.
The success of the measures evidently requires that there should be no
apparent movement of the bands apart from real retardations in the tubes.
1896] ON SOME PHYSICAL PROPERTIES OF ARGON AND HELIUM. 221
As the apparatus -was at first arranged, this condition was insufficiently
satisfied. Although all the parts were carried upon the walls of the room,
frequent and somewhat sudden displacements of the bands relatively to the
web were seen to occur, probably in consequence of the use of wood in some
of the supports. The observations could easily be arranged in such a manner
that no systematic error could thence enter, but the agreement of individual
measures was impaired. Subsequently a remedy was found in the use of
a second system of bands, formed by light which passed just above the tubes,
to which, instead of to the web, the moveable bands were referred. The
coincidence of the two systems could be observed with accuracy, and was
found to be maintained in spite of movements of both relatively to the web.
In the comparisons of argon and air (with nearly the same refractivities)
the changes of pressure employed were about 8 in. (20 cm.), being deductions
from the atmospheric pressure. In one observation of July 26, the numbers,
representing suctions in inches of mercury, stood
Argon Air
8-54 9-96
0-01 177
8-53 819
Ratio = 0-961,
signifying that 8'53 in. of argon balanced 819 in. of dry air. Four sets,
during which the air and argon (from the globe as last filled for weighing)
were changed, taken on July 17, 18, 19, 26, gave respectively for the final
ratio 0-962, 0*961, 0'961, 0*960, or as the mean
Refractivity of argon _
Refractivity of air
The evidence from the refractivities, as well as from the weights, is very
unfavourable to the view that argon is an allotropic form of nitrogen such as
would be denoted by N3.
The above measurements, having been made with lamp-light, refer to the
most luminous region of the spectrum, say in the neighbourhood of D. But
since no change in the appearance of the bands at the two settings could
be detected, the inference is that the dispersions of the two gases are
approximately the same, so that the above ratio would not be much changed,
even if another part of the spectrum were chosen. It may be remarked that
the displacement actually compensated in the above experiments amounted
to about forty bands, each band corresponding to about ^ in. (5 mm.) pressure
of mercury.
Similar comparisons have been made between air and helium. The
latter gas, prepared by Professor Ramsay, was brought from London by
222 ON SOME PHYSICAL PROPERTIES OF ARGON AND HELIUM. [218
Mr W. Randall, who farther gave valuable assistance in the manipulations.
It appeared at once that the refractivity of helium was remarkably low, 13 in.
pressure of the gas being balanced by less than 2 in. pressure of air. The
ratios given by single comparisons on July 29 were 0147, 0'146, 0145, 0146,
mean 0146 ; and on July 30 0147, 0147, 0145, 0145, mean 0146. The
observations were not made under ideal conditions, on account of the smallness
of the changes of air pressure ; but we may conclude that with considerable
approximation
Refractivity of helium _ ¥
Refractivity of air
The lowest refractivity previously known is that of hydrogen, nearly 0'5
of that of air.
Viscosity of Argon and Helium.
The viscosity was investigated by the method of passage through capillary
tubes. The approximate formula has been investigated by O. Meyer^, on
the basis of Stokes' theory for incompressible fluids. If the driving pressure
(pi~Pa) i& no* too great, the volume F2 delivered in time t through a tube
of radius R and length \ is given by
the volume being measured at the lower pressure p2} and rj denoting the
viscosity of the gas. In the comparison of different gases F2, plt p2, R, X
may be the same, and then 97 is proportional to t.
In the apparatus employed two gas pipettes and manometers, somewhat
similar to those shown in Fig. 2, were connected by a capillary tube of very
small bore and about 1 metre long. The volume F2 was about 100 c.c., and
was caused to pass by a pressure of a few centimetres of mercury, maintained
as uniform as possible by means of the pipettes. There was a difficulty,
almost inherent in the use of mercury, in securing the right pressures during
the first few seconds of an experiment ; but this was not of much importance
as the whole time t amounted to several minutes. The apparatus was tested
upon hydrogen, and was found to give 'the received numbers with sufficient
accuracy. The results, referred to dry air, were for helium 0'96; and for
argon T21, somewhat higher than for oxygen which at present stands at
the head of the list of the principal gases J.
* [1902. The sample must have contained impurity — probably hydrogen. Prof. Ramsay's
latest result for the refractivity of helium referred to air is -1238 (Proc. Roy. Soc. LXVII. p. 331,
1900).]
t Pogg. Ann. Vol. cxxvii. p. 270, 1866.
t [1902. Schultze (Drude Ann. vi. p. 310, 1901) finds for helium 1-086 in place of 0'96.]
1896] ON SOME PHYSICAL PROPERTIES OF ARGON AND HELIUM. 223
Gas from the Bath Springs.
In the original memoir upon argon* results were given of weighings of
the residue from the Bath gas after removal of oxygen, carbonic anhydride,
and moisture, from which it appeared that the proportion of argon was only
one-half of that contained in the residue, after similar treatment, from the
atmosphere. After the discovery of helium by Professor Ramsay, the question
presented itself as to whether this conclusion might not be disturbed by the
presence in the Bath gas of helium, whose lightness would tend to compensate
the extra density of argon.
An examination of the gas which had stood in my laboratory more than
a year having shown that it still contained no oxygen, it was thought worth
while to remove the nitrogen so as to determine the proportion that would
refuse oxidation. For this purpose 200 c.c. were worked up with oxygen until
the volume, free from nitrogen, was reduced to 8 c.c. On treatment with
pyrogallol and alkali the residue measured 3'3 c.c., representing argon, and
helium, if present. On sparking the residue at atmospheric pressure and
examining the spectrum, it was seen to be mainly that of argon, but with an
unmistakable exhibition of D3. At atmospheric pressure this line appears
very diffuse in a spectroscope of rather high power, but the place was correct.
From another sample of residue from the Bath gas, vacuum tubes were
charged by my son, Mr R. J. Strutt, and some of them showed D3 sharply
denned and precisely coincident with the line of helium in a vacuum tube
prepared by Professor Ramsay.
Although the presence of helium in the Bath gas is not doubtful, the
quantity seems insufficient to explain the low density found in October,
1894. In order to reconcile that density with the proportion of residue
(3'3/200 = 0'016) found in the experiment just described, it would be necessary
to suppose that the helium amounted to 25 per cent, of the whole residue of
argon and helium. Experiment, however, proved that a mixture of argon
and helium containing 10 per cent, of the latter gas showed D3 more plainly
than did the Bath residue. It is just possible that some of the helium was
lost by diffusion during the long interval between the experiments whose
results are combined in the above estimate.
Buxton Gas.
Gas from the Buxton springs, kindly collected for me by Mr A. McDougall,
was found to contain no appreciable oxygen. The argon amounted to about
* Rayleigh and Ramsay, Phil. Trans. A, Vol. CLXXXVI. p. 227, 1895. [Vol. iv. p. 172.]
224 ON SOME PHYSICAL PROPERTIES OF ARGON AND HELIUM. [218
2 per cent, of the volume. When its spectrum was examined, the presence
of D3 was suspected, but the appearance was too feeble to allow of a definite
statement being made. The proportion of helium is in any case very much
lower than in the Bath gas.
Is Helium contained in the Atmosphere?
Apart from its independent interest, this question is important in con-
nection with the density of atmospheric argon. Since the spectrum of this
gas does not show the line D3, we may probably conclude that the proportion
of helium is less than 3 per cent. ; so that there would be less than 3 x 10~4
of helium in the atmosphere. The experiment about to be described was
an attempt to carry the matter further, and is founded upon the observation
by Professor Ramsay, that the solubility of helium in water is only O007, less
than one-fifth of that which we found for argon*.
It is evident that if a mixture of helium and argon be dissolved in water
until there is only a small fraction remaining over, the proportion of helium
will be much increased in the residue. Two experiments have been made,
of which that on October 6, 1895, was the more elaborate. About 60 c.c.
of argon were shaken for a long time with well-boiled water contained in
a large flask. When the absorption had ceased, the residue of 30 c.c. was
sparked with a little oxygen until no nitrogen could be seen in the spectrum.
It was then treated a second time with boiled water until its volume was
reduced to 1£ c.c. With this vacuum tubes were charged by my son at two
different pressures. In none of them could D3 be detected; nor was there
any marked difference to be seen between the spectra of the washed and the
unwashed argon. If helium be present in the atmosphere, it must be in very
small quantity, probably much less than a ten-thousandth part^*.
* Phil. Trans. A, Vol. CLXXXVI. p. 225, 1895. [VoL iv. p. 170.]
t [1902. The presence of traces of helium in the atmosphere is not doubtful.]
219.
ON THE AMOUNT OF ARGON AND HELIUM CONTAINED IN
THE GAS FROM THE BATH SPRINGS*
[Proceedings of the Royal Society, LX. pp. 56, 57, 1896.]
THE presence of helium in the residue after removal of nitrogen from this
gas was proved in a former paperf, but there was some doubt as to the
relative proportions of argon and helium. A fresh sample, kindly collected
by Dr Richardson, has therefore been examined. Of this 2,500 c.c., submitted
to electric sparks in presence of oxygen, gave a final residue of 37 c.c., after
removal of all gases known until recently. The spectrum of the residue,
observed at atmospheric pressure, showed argon, and the D3 line of helium
very plainly.
The easy visibility of D3 suggested the presence of helium in some such
proportion as 10 per cent., and this conjecture has been confirmed by a
determination of the refractivity of the mixture. It may be remembered
that while the refractivity of argon approaches closely that of air, the relative
number being 0'961, the refractivity of helium (as supplied to me by Pro-
fessor Ramsay) is very low, being only 0'146 on the same scale. If we assume
that any sample of gas is a mixture of these two, its refractivity will deter-
mine the proportions in which the components are present.
The observations were made by an apparatus similar in character to that
already described, but designed to work with smaller quantities of gas. The
space to be filled is only about 12 c.c., and if the gas be at atmospheric
pressure its refractivity may be fixed to about 1/1000 part. By working at
pressures below atmosphere very fair results could be arrived at with quan-
tities of gas ordinarily reckoned at only 3 or 4 c.c.
The refractivity found for the Bath residue after desiccation was 0'896
referred to air, so that the proportional amount of helium is 8 per cent.
Referred to the original volume, the proportion of helium is 1'2 parts per
thousand.
* I am reminded by Mr Whitaker that helium is appropriately associated with the Bath
waters, which, according to some antiquaries, were called by the Romans Aqua Solis.
t Boy. Soc. Proc. Vol. LIX. p. 206, 1896. [Vol. iv. p. 223.]
R. IV. 15
220.
THE REPRODUCTION OF DIFFRACTION GRATINGS.
[Nature, LIV. pp. 332, 333, 1896*.]
I HAVE first to apologise for the very informal character of the communi-
cation which I am about to make to the Club ; I have not been able to put
anything down upon paper, but I thought it might be interesting to some to
hear an account of experiments that have now been carried on at intervals
for a considerable series of years in the reproduction — mainly the photographic
reproduction — of diffraction gratings. Probably most of you know that these
consist of straight lines ruled very closely, very accurately, and parallel to
one another, upon a piece of glass or speculum metal. Usually they are
ruled with a diamond by the aid of a dividing machine ; and in late years,
particularly in the hands of Rutherfurd and Rowland, an extraordinary
degree of perfection has been attained. It was many years ago — nearly
25 years, I am afraid — that I first began experiments upon the photographic
reproduction of these divided gratings, each in itself the work of great time
and trouble, and costing a good deal of money. At that time the only
gratings available were made by Nobert, in Germany, of which I had two,
each containing about a square inch of ruled surface, one of about 3,000
lines to the inch, and the other of about 6,000. It happened, by an accident,
that the grating with 3,000 lines was the better of the two, in that it
was more accurately ruled, and gave much finer definition upon the solar
spectrum; the 6,000 line grating was brighter, but its definition was
decidedly inferior, so that both had certain advantages according to the
particular object in view.
If it comes to the question of how to make a grating by photography,
probably the first idea to occur to one would be that it might be a com-
paratively simple matter to make a grating upon a large scale, and then
* [From a report of] an address delivered at the eighth annual conference of the Camera Club.
1896] THE REPRODUCTION OF DIFFRACTION GRATINGS. 227
reduce it by photography, but if one goes into the figures the project is
not found so promising. Take, for instance, a grating with 10,000 lines to
the inch ; if you magnified that, say 100 times, your lines would then be 100
to the inch; if you magnified it 1,000 times, they would still be 10 to the
inch, and that would be a convenient size so far as interval between the lines
was concerned ; but think what would be the area required to hold a grating
magnified to that extent. By the time you have magnified the inch by 100
or 1,000, you would want a wall of a house or of a cathedral to hold the
grating. If the problem were proposed of ruling a grating with 6,000 lines,
with a high degree of accuracy, it would be easier to do it on a microscopic
scale than upon a large scale, leaving out of consideration the difficulty of
reproducing it. And those difficulties would be insuperable, because, al-
though with a good microscopic object-glass it would be easy to photograph
lines which are much closer together than 3,000 or 6,000 to the inch,
yet that could only be achieved over a very small area of surface — nothing
like a square inch ; and if it were required to cover a square inch with lines
6,000 to the inch, it would be beyond the power, not only, I believe, of any
microscope, but of any lens that was ever made. So that that line of
investigation does not fulfil the promise which at first it might appear to
give ; and, in fact, there is nothing simpler or better than to copy the original
ruled by a dividing engine, by the simple process of contact printing.
For this purpose some precautions are required. You must use very flat
glass, by preference it should be optically worked glass, although very good
results may be obtained on selected pieces of ordinary plate. Of course, no
one would think of making such a print by diffused daylight, but the sun
itself, or a point of light from any suitable source, according to the nature
of the photographic process which is adopted, permits quite well of the
reproduction of any grating of a moderate degree of fineness. I have used
almost all varieties of photographic processes in my time. In the days
when I first worked, the various dry collodion processes were better under-
stood than they are now; the old albumen process was extremely suitable
for such work as this, on account of the almost complete absence of structure
in the film, and the very convenient hardness of the surface, which made the
result comparatively little liable to injury. I used with success the dry
collodion processes, the tannin process among others, and also some of the
direct printing methods, such as the collodio-chloride. The latter method,
worked upon glass, gave excellent results, particularly if the finished print
was treated with mercury in the way commonly used for intensification,
except that, in the treatment of a grating with mercury, it is desirable to
stop at the mercury and not to go on to the blackening process used in the
intensification of negatives. From the visual point of view, the grating,
after intensification — if one may use the term — with mercury, looks much
less intense than before, but, nevertheless, the spectra seen when a point or
15—2
228 THE REPRODUCTION OF DIFFRACTION GRATINGS. [220
slit of light is looked at through the grating becomes very much more
brilliant.
I used another process at that time, more than twenty years ago, which
gave excellent results, but had not the degree of certainty that I aimed at,
namely, a bichromated gelatine process, similar to carbon printing, except
that no pigment was employed. A glass plate was simply coated with
bichromated gelatine of a suitable thickness — and a good deal depended
upon hitting that off correctly; if the coating was too thin the grating
showed a deficiency of brightness, whereas, if it was too thick, there might
be a difficulty in getting it sufficiently uniform and smooth on the surface.
However, I obtained excellent gratings by that process, most of them capable
of showing the nickel line between the two well-known sodium or D lines
in the solar spectrum, when suitably examined. The collodio-chloride process
was comparatively slow, and bichromated gelatine required two or three
minutes exposure to sunlight to produce a proper effect; but for the more
sensitive developed negative processes a very much less powerful light or
a reduced exposure was needed.
The performance of the copies was quite good, and, except where there
was some obvious defect, I never could see that they were worse than the
originals; in fact, in respect of brightness it not unfrequently happened
that the copies were far superior to the originals, so that in many cases
they would be more useful. I do not mean by that, however, that I would
rather have a copy than an original if anyone wanted to make me a
present. There seems to be some falling off in copies ; so that they cannot
well be copied again, and if you want to work upon spectra of an extremely
high order, dispersed to a great extent laterally from the straight line,
a copy would not be satisfactory. The reproduction of gratings on bi-
chromated gelatine is easily and quickly accomplished; there is only the
coating of the glass over-night, rapid drying to avoid crystallisation in the
film, exposure, washing, and drying. In order to get the best effect it is
usually desirable to treat the bichromated copies with hot water. It is
a little difficult to understand what precisely happens. All photographers
know that the action of light upon bichromated gelatine is to produce
a comparative insolubility of the gelatine. In the carbon process, and
many others in which gelatine is used, the gelatine which remains soluble,
not having been sufficiently exposed to light, is fairly washed away in
the subsequent treatment with warm water, but for that effect it is generally
necessary to get at the back of the gelatine film, because on its face there
is usually a layer which is so insoluble as not to allow of the washing away
of any of the gelatine situated behind. But in the present case there is
no question of transferring the film, which remains fixed to the glass, and
therefore it is difficult to see how any gelatine could be dissolved out.
However, under the action of water, the less exposed gelatine no doubt
1896] THE REPRODUCTION OF DIFFRACTION GRATINGS. 229
swells more than that which has received more exposure and has thus
lost its affinity for water; and while the gelatine is wet it is reasonable
that a rib-like structure should ensue, which is what would be required
in order to make a grating, but when the gelatine dries, one would suppose
that all would again become flat, and indeed that happens to a certain
extent. The gratings lost a great deal of intensity in drying, but, if properly
treated with warm water, the reduction does not go too far, and a considerable
degree of intensity is left when the photograph is dry.
Although it belongs to another branch of the subject, a word may not
be out of place as to the accuracy with which the gratings must be made.
It seems a wonderful thing at first sight, to rule 6,000 lines to an inch
at all, if you think of the smallest interval that you can readily see with
the eye, perhaps one-hundredth of an inch, and remember that in these
gratings there are sixty lines in the space of one-hundredth of an inch,
and all disposed at rigorously equal intervals. Those familiar with optics
will understand the importance of extreme accuracy if I give an illustration.
Take the case of the two sodium lines in the spectrum, the D lines ; they
differ in wave-length by about a thousandth part; the dispersion — the
extent to which the light is separated from the direct line — is in proportion
to the wave-length of the light, and inversely as the interval between
the consecutive lines on the grating ; so that, if we had a grating in which
the first half was ruled at the rate of 1,000 to the inch, and the second
half at the rate of 1,001 to the inch, the one half would evidently do the
same thing for one soda line as the other half of the grating was doing
for the other soda line, and the two lines would be mixed together
and confused. In order, therefore, to do anything like good work, it is
necessary, not only to have a very great number of lines, but to have
them spaced with most extraordinary precision ; and it is wonderful what
success has been reached by the beautiful dividing machines of Rutherfurd
and Rowland. I have seen Rowland's machine at Baltimore, and have
heard him speak of the great precautions required to get good results.
The whole operation of the machine is automatic; the ruling goes on
continuously day and night, and it is necessary to pay the most careful
regard to uniformity of temperature, for the slightest expansion or con-
traction due to change of temperature of the different parts of the machine
would bring utter confusion into the grating and its resulting spectrum.
The contact in printing has to be pretty close and the finer the
grating the closer must the contact be. I experimented upon that point :
one can get some kind of result, theoretically, by preparing a photographic
film with a slightly convex surface and using that for the print; then,
where the contact was closest, the original of course was very well im-
pressed, and round that, one got different degrees of increasingly imperfect
contact, and one could trace in the result what the effect of imperfect
230 THE REPRODUCTION OF DIFFRACTION GRATINGS. [220
contact is. I found that, both with gratings of 3,000 and 6,000 lines to
the inch, good enough contact was obtained with ordinary flat glass ; but
when you come to gratings of 17,000 or 20,000 lines to the inch the contact
requires to be extremely close, and in order to get a good copy of a grating
with 20,000 lines per inch it is necessary that there should nowhere be
one ten-thousandth of an inch between the original and the printing
surface — a degree of closeness not easily secured over the entire area. It
is rather singular that though I published full accounts of this work a long
time ago, and distributed a large number of copies, the process of repro-
ducing gratings by photography did not become universally known, and
was re-discovered in France, by Izarn, only two or three years since.
One reason why photographic reproduction is not practised to a very
great extent, is, that the modern gratings — such as Rowland's — are ruled
almost universally upon speculum metal. A grating upon speculum metal
is very excellent for use, but does not well lend itself to the process of
photographic copying, although I have succeeded to a certain extent in
copying a grating ruled upon speculum metal. For this purpose the light
had to pass first through the photographic film, then be reflected from the
speculum metal, and so pass back again through the film. Gratings, such
as could easily be made by copying from a glass original, are not readily
produced from one on speculum metal, and I think that is the reason why
the process has not come into more regular use. Glass is much more
trying than speculum metal to the diamond, and that accounts for the
latter being generally preferred for gratings ; it is very hard, but has not
ruinous effects upon the diamond; indeed the principal difficulty consists
in getting a good diamond point, and maintaining it in a shape suitable for
making the very fine cut which is required.
I may now allude to another method of photographic reproduction which
I tried only last summer. It happened that I then went with Professor
Meldola over Waterlow's large photo-mechanical printing establishment,
and I was much interested, among other very interesting things, to see
the use of the old bitumen process — the first photographic process known.
It is used for the reproduction of cuts in black and white. A carefully
cleansed zinc plate is coated with a varnish of bitumen dissolved in benzole,
and exposed to sunlight for about two hours under a negative giving great
contrast. Where the light penetrates the negative the bitumen becomes
comparatively insoluble, and where it has been protected from the action
of light it retains its original degree of solubility. When the exposed plate
is treated with a solvent, turpentine or some milder solvent than benzole,
the protected parts are dissolved away, leaving the bare metal; whereas
the parts that have received the sunlight, being rendered insoluble, remain
upon the metal and protect it in the subsequent etching process. I did
not propose to etch metal, and, therefore, I simply used the bitumen varnish
1896] THE REPRODUCTION OF DIFFRACTION GRATINGS. 231
spread upon glass plates, and exposed the plates so prepared to sunshine
for about two hours in contact with the grating. They were then developed,
if one may use the phrase, with turpentine; and this is the part of the
process which is the most difficult to manage. If you stop development
early you get [without difficulty] a grating which gives fair spectra, but
it may be deficient in intensity and brightness; if you push development
the brightness increases up to a point at which the film disintegrates
altogether. In this way one is tempted to pursue the process to the very
last point, and, although one may succeed so far as to have a film which
is quite intact so long as the turpentine is upon it, I have not succeeded
in finding any method of getting rid of the turpentine without causing
the disintegration of the film. In the commercial application of the
process the bitumen is treated somewhat brutally — the turpentine is rinsed
off with a jet of water; I have tried that, and many of my results have
been very good. I have also tried to sling off the turpentine with the aid
of a kind of centrifugal machine, but by either plan the [too tender] film
is liable not to survive the treatment required for getting rid of the
turpentine. If the solvent is allowed to remain we are in another difficulty,
because then the developing action is continued and the result is lost.
But if the process is properly managed, and development stopped at the
right point, and if the film be of the right degree of thickness, you get
an excellent copy. I have one here, 6,000 lines to the inch, which I think
is about the very best copy I have ever made. The method gives results
somewhat superior to the best that can be got with gelatine ; but I would
not recommend it in preference to the latter, because it is much more difficult
to work unless some one can hit upon an improved manipulation.
I will not enlarge upon the importance of gratings; those acquainted
with optics know how very important is the part played by diffraction
gratings in optical research, and how the most delicate work upon spectra,
requiring the highest degree of optical power, is made by means of gratings,
ruled on speculum metal by Rowland. I suppose the reason why no pro-
fessional photographer has taken up the production of photographic gratings,
is the difficulty of getting the glass originals; they are very expensive,
and indeed I do not know where they are now to be obtained. It seems
a pity that photographic copies should not be more generally available.
I have given a great many away myself; but educational establishments
are increasing all over the country, and for the purpose of instructing
students it is desirable that reasonably good gratings should be placed in
their hands, to make them familiar with the measurements by which the
wave-length of light is determined.
[1902. For earlier papers upon this subject see Vol. I. pp. 160, 199, 504.]
221.
THE ELECTRICAL RESISTANCE OF ALLOYS.
[Nature, LIV. pp. 154, 155, 1896.]
THE recent researches of Profs. Dewar and Fleming upon the electrical
resistance of metals at low temperatures have brought into strong relief
the difference between the behaviour of pure metals and of alloys. In the
former case the resistance shows every sign of tending to disappear altogether
as the absolute zero of temperature is approached, but in the case of alloys
this condition of things is widely departed from, even when the admixture
consists only of a slight impurity.
Some years ago it occurred to me that the apparent resistance of an
alloy might be partly made up of thermo-electric effects, and as a rough
illustration I calculated the case of a conductor composed of two metals
arranged in alternate laminae perpendicular to the direction of the current.
Although a good many difficulties remain untouched, I think that the calcu-
lation may perhaps suggest something to those engaged upon the subject.
At any rate it affords a priori ground for the supposition that an important
distinction may exist between the resistances of pure and alloyed metals.
The general character of the effect is easily explained. According to the
discovery of Peltier, when an electric current flows from one metal to another
there is development or absorption of heat at the junction. The temperature
disturbance thus arising increases until the conduction of heat through the
laminae balances the Peltier effects at the junctions, and it gives rise to a
thermo-electromotive force opposing the passage of the current. Inasmuch as
the difference of temperature at the alternate j unctions is itself proportional
to the current, so is also the reverse electromotive force thereby called into
play. Now a reverse electromotive force proportional to current is indistin-
guishable experimentally from a resistance; so that the combination of
1896] THE ELECTRICAL RESISTANCE OF ALLOYS. 233
laminated conductors exhibits a false resistance, having (so far as is known)
nothing in common with the real resistance of the metals.
If e be the thermo-electric force of the couple for one degree difference
of temperature of the junctions; t, t' the actual temperatures; then the
electromotive force for one couple is e (t — t'}. If we suppose that there are
n similar couples per unit of length perpendicular to the lamination, the
whole reverse electromotive force per unit of length is ne (t — t'). Again, if
C be the current corresponding to unit of cross-section, the development of
heat per second at each alternate junction is per unit of area 273 x e x C,
the actual temperature being in the neighbourhood of zero Cent. This is
measured in ergs, and is to be equated to the heat conducted per second
towards the cold junctions on the two sides. If k, k' be the conductivities
for heat of the two metals, I and I' the corresponding thicknesses, the heat
conducted per second is
or if lf(l + l')=p}
the conducted heat is
n(t-t')\k/p + k'/q}.
In this expression p + q = 1 , the symbols p and q denoting the proportional
amounts by volume in which the two metals are associated. Thus when a
stationary state is reached
273 x e x C = n (t - t') [k/p + k'/q}.
This determines (t — t') when C is given ; and the whole back electromotive
force per unit of thickness is rC, where
273 x e2
k/p + k'/q'
This is the expression for the false resistance per unit of thickness, which,
it should specially be noted, is independent of n, the number of couples.
The number of couples which co-operate is indeed increased by finer lamina-
tion, but the efficiency of each is decreased in the same proportion by the
readier conduction of heat between the junctions. It is scarcely necessary
to point out that the false resistance is called into play only by currents
which flow across the laminae.
In my original calculation the metals chosen for illustration were iron
and copper. In this case (Everett's C.G.8. System of Units, p. 192) c = 1600.
The conductivities are to be measured in ergs. For iron, k = '164 x 4'2 x 107;
234 THE ELECTRICAL RESISTANCE OF ALLOYS. [221
for copper, k' = I'll x 4'2 x 107. Thus, if the metals are in equal volumes
_2x 273x1600*
4-2 x 107 x 1-27
This is the thermo-electric addition to the true specific resistance, and is
about 1£ per cent, of that of copper. Such an addition may seem small ;
but it should be remembered that for the more distinctively thermo-electric
metals e is much larger, and that it enters by its square. In any case it
seems desirable that this complication should be borne in mind. The
consequences which follow from recognised laws for laminated structures,
however fine, must surely have some bearing upon the properties of alloys,
although in this case the fineness is molecular.
222.
ON THE THEORY OF OPTICAL IMAGES, WITH SPECIAL
REFERENCE TO THE MICROSCOPE.
[Philosophical Magazine, XLII. pp. 167—195, 1896.]
THE special subject of this paper has been treated from two distinct
points of view. In the work of Helmholtz* the method followed is analogous
to that which had long been used in the theory of the telescope. It consists
in tracing the image representative of a mathematical point in the object,
the point being regarded as self-luminous. The limit to definition depends
upon the fact that owing to diffraction the image thrown even by a perfect
lens is not confined to a point, but distends itself over a patch or disk of
light of finite diameter. Two points in the object can appear fully separated
only when the representative disks are nearly clear of one another. The
application to the microscope was traced by means of a somewhat extended
form of Lagrange's general optical theorem, and the conclusion was reached
that the smallest resolvable distance e is given by
e = iX./sina, ................................. (1)
X being the wave-length in the medium where the object is situated, and
a the divergence -angle of the extreme ray (the semi-angular aperture) in
the same medium. If X0 be the wave-length in vacuum,
/i being the refractive index of the medium; and thus
€ = ^X0/yu,sin a ............................... (3)
The denominator /z sin a is the quantity now well known (after Abbe) as
the " numerical aperture."
The extreme value possible for a is a right angle, so that for the micro-
scopic limit we have
Pogg. Ann. Jubelband, 1874.
236 ON THE THEORY OF OPTICAL IMAGES, [222
The limit can be depressed only by a diminution in X0, such as photography
makes possible, or by an increase in //,, the refractive index of the medium
in which the object is situated.
This method, in which the object is considered point by point, seems
the most straightforward, and to a great extent it solves the problem
without more ado. When the representative disks are thoroughly clear
of one another, the two points in which they originate are resolved, and
on the other hand, when the disks overlap the points are not distinctly
separated. Open questions can relate only to intermediate cases of partial
overlapping and various degrees of resolution. In these cases (as has been
insisted upon by Dr Stoney) we have to consider the relative phases of the
overlapping lights before we can arrive at a complete conclusion.
If the various points of the object are self-luminous, there is no per-
manent phase-relation between the lights of the overlapping disks, and
the resultant illumination is arrived at by simple addition of separate
intensities. This is the situation of affairs in the ordinary use of a telescope,
whether the object be a double star, the disk of the sun, the disk of the
moon, or a terrestrial body. The distribution of light in the image of
a double point, or of a double line, was especially considered in a former
paper*, and we shall return to the subject later.
When, as sometimes happens in the use of the telescope, and more
frequently in the use of the microscope, the overlapping lights have per-
manent phase-relations, these intermediate cases require a further treatment ;
and this is a matter of some importance as involving the behaviour of the
instrument in respect to the finest detail which it is capable of rendering.
We shall see that the image of a double point under various conditions
can be delineated without difficulty.
In the earliest paper by Prof. Abbef, which somewhat preceded that
of Helmholtz, similar conclusions were reached; but the demonstrations
were deferred, and, indeed, they do not appear ever to have been set forth
in a systematic manner. Although some of the positions then taken up,
as for example that the larger features and the finer structure of a micro-
scopic object are delineated by different processes, have since had to be
abandoned^, the publication of this paper marks a great advance, and has
contributed powerfully to the modern development of the microscope§. In
* " Investigations in Optics, "with special reference to the Spectroscope," Phil. Mag. Vol. vin.
p. 266 (1879). [Vol. i. p. 415.]
t Archivf. Mikr. Anat. Vol. ix. p. 413 (1873).
J Dallenger's edition of Carpenter's Microscope, p. 64, 1891.
§ It would seem that the present subject, like many others, has suffered from over-specializa-
tion, much that is familiar to the microscopist being almost unknown to physicists, and vice versa.
For myself I must confess that it is only recently, in consequence of a discussion between
1896]
WITH SPECIAL REFERENCE TO THE MICROSCOPE.
237
Prof. Abbe's method of treating the matter the typical object is not a
luminous point, but a grating illuminated by plane waves. Thence arise
the well-known diffraction spectra, which are focused near the back of
the object-glass in its principal focal plane. If the light be homogeneous
the spectra are reduced to points, and the final image may be regarded
as due to the simultaneous action of these points acting as secondary centres
of light. It is argued that the complete representation of the object
requires the co-operation of all the spectra. When only a few are present,
the representation is imperfect ; and when there is only one — for this purpose
the central image counts as a spectrum — the representation wholly fails.
That this point of view offers great advantages, at least when the object
under consideration is really a grating, is at once evident. More especially
is this the case in respect of the question of the limit of resolution. It
is certain that if one spectrum only be operative, the image must consist
of a uniform field of light, and that no sign can appear of the real periodic
structure of the object. From this consideration the resolving-power is
readily deduced, and it may be convenient to recapitulate the argument
for the case of perpendicular incidence. In Fig. 1 AB represents the axis,
Fig. 1.
A being in the plane of the object (grating) and B in the plane of the
image. The various diffraction spectra are focused by the lens LL' in
the principal focal plane, S0 representing the central image due to rays
which issue normally from the grating. After passing S0 the rays diverge
in a cone corresponding to the aperture of the lens and illuminate a circle
CD in the plane of the image, whose centre is B. The first lateral
spectrum /S\ is formed by rays diffracted from the grating at a certain
angle; and in the critical case the region of the image illuminated by the
rays diverging from & just includes B. The extreme ray 8^ evidently
Mr L. Wright and Dr G. 3. Stoney in the English Mechanic (Sept., Oct., Nov., 1894; Nov. 8,
Dec. 13, 1895; Jan. 17, 1896), that I have become acquainted with the distinguishing features of
Prof. Abbe's work, and have learned that it was conducted upon different lines to that of
Helmholtz. I am also indebted to Dr Stoney for a demonstration of some of Abbe's experiments.
238 ON THE THEORY OF OPTICAL IMAGES, [222
proceeds from A, which is the image of B. The condition for the co-
operation at B of the first lateral spectrum is thus that the angle of diffraction
do not exceed the semi-angular aperture a. By elementary theory we
know that the sine of the angle of diffraction is X/e, so that the action of
the lateral spectrum requires that e exceed X/sin a. If we allow the incidence
upon the grating to be oblique, the limit becomes ^ X/sin a, as in (1).
We have seen that if one spectrum only illuminate B, the field shows
no structure. If two spectra illuminate it with equal intensities, the field
is occupied by ordinary interference bands, exactly as in the well-known
experiments of Fresnel. And it is important to remark that the character
of these bands is always the same, both as respects the graduation of light
and shade, and in the fact that they have no focus. When more than two
spectra co-operate, the resulting interference phenomena are more com-
plicated, and there is opportunity for a completer representation of the special
features of the original grating*.
While it is certain that the image ultimately formed may be considered
to be due to the spectra focused at $0, S^.., the degree of conformity of
the image to the original object is another question. From some of the
expositions that have been given it might be inferred that if all the spectra
emitted from the grating were utilized, the image would be a complete
representation of the original. By considering the case of a very fine
grating, which might afford no lateral spectra at all, it is easy to see that
this conclusion is incorrect, but the matter stands in need of further eluci-
dation. Again, it is not quite clear at what point the utilization of a
spectrum really begins. All the spectra which the grating is competent
to furnish are focused in the plane S^; and some of them might be
supposed to operate partially even although the part of the image under
examination is outside the geometrical cone defined by the aperture of
the object-glass. For these and other reasons it will be seen that the
* These effects were strikingly illustrated in some observations upon gratings with 6,000 lines
to the inch, set up vertically in a dark room and illuminated by sunlight from a distant vertical
slit. The object-glass of the microscope was a quarter-inch. When the original grating, divided
upon glass (by Nobert), was examined in this way, the lines were well seen if the instrument was
in focus, but, as usual, a comparatively slight disturbance of focus caused all structure to disappear.
When, however, a photographic copy of the same glass original, made with bitumen [p. 231], was
substituted for it, very different effects ensued. The structure could be seen even although the
object-glass were drawn back through 1£ inch from its focused position ; and the visible lines
were twice as close, as if at the rate of 12,000 to the inch. The difference between the two cases
is easily explained upon Abbe's theory. A soda flame viewed through the original showed a strong
central image (spectrum of zero order) and comparatively faint spectra of the first and higher
orders. A similar examination of the copy revealed very brilliant spectra of the first order on both
sides, and a relatively feeble central image. The case is thus approximately the same as when in
Abbe's experiment all spectra except the first (on the two sides) are blocked out.
1896] WITH SPECIAL REFERENCE TO THE MICROSCOPE. 239
spectrum theory*, valuable as it is, needs a good deal of supplementing,
even when the representation of a grating under parallel light is in
question.
When the object under examination is not a grating or a structure in
which the pattern is repeated an indefinite number of times, but for
example a double point, or when the incident light is not parallel, the
spectrum theory, as hitherto developed, is inapplicable. As an extreme
example of the latter case we may imagine the grating to be self-luminous.
It is obvious that the problem thus presented must be within the scope
of any complete theory, and equally so that here there are no spectra
formed, as these require the radiations from the different elements of the
grating to possess permanent phase-relations. It appears, therefore, to be
a desideratum that the matter should be reconsidered from the older point
of view, according to which the typical object is a point and not a grating.
Such a treatment illustrates the important principle that the theory of
resolving-power is essentially the same for all instruments. The peculiarities
of the microscope arise from the fact that the divergence-angles are not
limited to be small, and from the different character of the illumination
usually employed; but, theoretically considered, these are differences of
detail. The investigation can, without much difficulty, be extended to
gratings, and the results so obtained confirm for the most part the conclusions
of the spectrum theory.
It will be convenient to Commence our discussion by a simple investiga-
tion of the resolving-power of an optical instrument for a self-luminous
double point, such as will be applicable equally to the telescope and to
the microscope. In Fig. 2 AB represents the axis, A being a point of the
Fig. 2.
object and B a point of the image. By the operation of the object-glass LL'
all the rays issuing from A arrive in the same phase at B. Thus if A be
self-luminous, the illumination is a maximum at B, where all the secondary
waves agree in phase. B is in fact the centre of the diffraction disk which
constitutes the image of A. At neighbouring points the illumination is
* The special theory initiated by Prof. Abbe is usually called the "diffraction theory," a
nomenclature against which it is necessary to protest. Whatever may be the view taken, any
theory of resolving power of optical instruments must be a diffraction theory in a certain sense,
so that the name is not distinctive. Diffraction is more naturally regarded as the obstacle to fine
definition, and not, as with some exponents of Prof. Abbe's theory, the machinery by which good
definition is brought about.
240 ON THE THEORY OF OPTICAL IMAGES, [222
less, in consequence of the discrepancies of phase which there enter. In
like manner, if we take a neighbouring point P in the plane of the object,
the waves which issue from it will arrive at B with phases no longer
absolutely accordant, and the discrepancy of phase will increase as the
interval AP increases. When the interval is very small, the discrepancy
of phase, though mathematically existent, produces no practical effect, and
the illumination at B due to P is as important as that due to A, the
intensities of the two luminous centres being supposed equal. Under these
conditions it is clear that A and P are not separated in the image. The
question is, to what amount must the distance AP be increased in order
that the difference of situation may make itself felt in the image. This
is necessarily a question of degree; but it does not require detailed calcu-
lations in order to show that the discrepancy first becomes conspicuous
when the phases corresponding to the various secondary waves which travel
from P to B range over about a complete period. The illumination at B
due to P then becomes comparatively small, indeed for some forms of
aperture evanescent. The extreme discrepancy is that between the waves
which travel through the outermost parts of the object-glass at L and L';
so that, if we adopt the above standard of resolution, the question is, where
must P be situated in order that the relative retardation of the rays PL
and PL' may on their arrival at B amount to a wave-length (X). In
virtue of the general law that the reduced optical path is stationary in
value, this retardation may be calculated without allowance for the different
paths pursued on the further side of L, L', so that its value is simply
PL — PL. Now since AP is very small, AL' — PL' is equal to ^LP.sina,
where a is the semi-angular aperture L AB. In like manner PL - AL has
the same value, so that
According to the standard adopted, the condition of resolution is therefore
that AP, or e, should exceed ^X/sina, as in (1). If e be' less than this,
the images overlap too much ; while if e greatly exceed the above value
the images become unnecessarily separated.
In the above argument the whole space between the object and the
lens is supposed to be occupied by matter of one refractive index, and
X represents the wave-length in this medium of the kind of light employed.
If the restriction as to uniformity be violated, what we have ultimately to
do with is the wave-length in the medium immediately surrounding
the object.
The statement of the law of resolving-power has been made in a form
appropriate to the microscope, but it admits also of immediate application
to the telescope. If 2R be the diameter of the object-glass, and D the
1896] WITH SPECIAL REFERENCE TO THE MICROSCOPE. 241
distance of the object, the angle subtended by AP is e/D, and the angular
resolving-power is given by
X
the well-known formula.
This method of derivation makes it obvious that there is no essential
difference of principle between the two cases, although the results are
conveniently stated in different forms. In the case of the telescope we have
to do with a linear measure of aperture and an angular limit of resolution,
whereas in the case of the microscope the limit of resolution is linear and
it is expressed in terms of angular aperture.
In the above discussion it has been supposed for the sake of simplicity
that the points to be discriminated are self-luminous, or at least behave
as if they were such. It is of interest to enquire how far this condition
can be satisfied when the object is seen by borrowed light. We may imagine
that the object takes the form of an opaque screen, perforated at two points,
and illuminated by distant sources situated behind.
If the source of light be reduced to a point, so that a single train of
plane waves falls upon the screen, there is a permanent phase-relation
between the waves incident at the two points, and therefore also between
the waves scattered from them. In this case the two points are as far as
possible from behaving as if they were self-luminous. If the incidence
be perpendicular, the secondary waves issue in the same phase ; but in
the case of obliquity there is a permanent phase-difference. This difference,
measured in wave-lengths, increases up to e, the distance between the
points, the limit being attained as the incidence becomes grazing.
When the light originates in distant independent sources, not limited
to a point, there is no longer an absolutely definite phase-relationship
between the secondary radiations from the two apertures ; but this condition
of things may be practically maintained, if the angular magnitude of the
source be not too large. For example, if the source be limited to an angle 6
round the normal to the screen, the maximum phase-difference measured
in wave-lengths is esin#, so that if sin# be a small fraction of X/e, the
finiteness of 6 has but little effect. When, however, sin 6 is so great that
e sin 9 becomes a considerable multiple of X, the secondary radiations
become approximately independent, and the apertures behave like self-
luminous points. It is evident that even with a complete hemispherical
illumination this condition can scarcely be attained when e is less
than X.
The use of a condenser allows the widely-extended source to be dispensed
with. By this means an image of a distant source composed of indepen-
E. iv. 16
242
ON THE THEORY OF OPTICAL IMAGES,
[222
dently radiating parts, such as a lamp-flame, may be thrown upon the
object, and it might at first sight be supposed that the problem under
consideration was thus completely solved in all cases, inasmuch as the two
apertures correspond to different parts of the flame. But we have to
remember here and everywhere that optical images are not perfect, and
that to a point of the flame corresponds in the image, not a point, but
a disk of finite magnitude. When this consideration is taken into account,
the same limitation as before is encountered.
For what is the smallest disk into which the condenser is capable of
concentrating the light received from a distant point ? Fig. 2 and the
former argument apply almost without modification, and they show that
the radius AP of the disk has the value ^X/sina, where a is the semi-
angular aperture of the condenser. Accordingly the diameter of the disk
cannot be reduced below X ; and if e be less than X the radiations from the
two apertures are only partially independent of one another.
It seems fair to conclude that the function of the condenser in micro-
scopic practice is to cause the object to behave, at any rate in some degree,
as if it were self-luminous, and thus to obviate the sharply-marked inter-
ference-bands which arise when permanent and definite phase-relations are
permitted to exist between the radiations which issue from various points
of the object.
As we shall have occasion later to employ Lagrange's theorem, it may
be well to point out how an instantaneous proof of it may be given upon
the principles [especially that the optical distance measured along a' ray
is a minimum] already applied. As before, AB (Fig. 3) represents the
Fig. 3.
axis of the instrument, A and B being conjugate points. P is a point
near A in the plane through A perpendicular to the axis, and Q is its
image* in the perpendicular plane through B. Since A and B are conjugate,
the optical distance between them is the same for all [ray-] paths, e.g. for
* [1902. In the original diagram Q was shown upon the wrong side of B. I owe the
correction to a correspondence with Prof, Everett.]
1896] WITH SPECIAL REFERENCE TO THE MICROSCOPE. 243
AR8B and ALMB. [For the same reason the optical distance from P
to Q is the same along the various rays, one of which lies infinitely near
to PRSQ and another to PLMQ.] And, since AP, BQ are perpendicular
to the axis, the optical distance from P to Q is the same (to the first order
of small quantities [such as AP]) as from A to B. Consequently the optical
distance PRSQ is the same as ARSB. Thus, if /*, p! be the refractive
indices in the neighbourhood of A and B respectively, a and ft the divergence-
angles RAL, SBM for a given ray, we have
fi.AP.ama=p'.BQ.sm/3, ........................ (6)
where AP, BQ denote the corresponding linear magnitudes of the two
images. This is the theorem of Lagrange, extended by Helmholtz so as to
apply to finite divergence-angles*.
We now pass on to the actual calculation of the images to be expected
upon Fresnel's principles in the various cases that may arise. The origin
of coordinates (£ = 0, rj = 0) in the focal plane is the geometrical image of
the radiant point. If the vibration incident upon the lens be represented
by cos (27rVt/\), where V is the velocity- of light, the vibration at any
point £, i] in the focal plane isf
in which / denotes the focal length, and the integration with respect to as
and y is to be extended over the aperture of the lens. If for brevity
we write
, ....................... (8)
(7) may be put into the form
where
8 = //sin (pas + qy) dxdy, C = // cos (px + qy) dxdy. . . .(10, 11)
It will suffice for our present purpose to limit ourselves to the case where
the aperture is symmetrical with respect to as and y. We have then
8 = Q, -and
C = ffcospx cosqy dxdy, ........................ (12)
the phase of the vibration being the same at all points of the diffraction
pattern.
* I learn from Czapski's excellent Theorie der Optischen Instrumente that a similar derivation
of Lagrange's theorem from the principle of minimum path had already been given many years
ago by Hockin (Micros. Soc. Journ. Vol. iv. p. 337, 1884).
t See for example Enc. Brit. " Wave Theory," p. 430 (1878). [Vol. in. p. 80.]
16—2
244
ON THE THEORY OF OPTICAL IMAGES,
[222
When the aperture is rectangular, of width a parallel to x, and of
width b parallel to y, the limits of integration are from — \a to +^a for x,
and from — 16 to + 16 for y. Thus
sin(7ny6/X/)
ab-
.(13)
and by (9) the amplitude of vibration (irrespective of sign) is Cj\f. This
expression gives the diffraction pattern due to a single point of the object
whose geometrical image is at ff = 0, 77 = 0. Sometimes, as in the application
to a grating, we wish to consider the image due to a uniformly luminous
line, parallel to 17, and this can always be derived by integration from the
expression applicable to a point. But there is a distinction to be observed
according as the radiations from the various parts of the line are independent
or are subject to a fixed phase-relation. In the former case we have to
deal only with the intensity, represented by /2 or C2/xys; and we get
by means of the known integral
•dx
dx =
.(15)
This gives, as a function of (•, the intensity due to a self-luminous line
whose geometrical image coincides with g — 0.
Under the second head of a fixed phase-relation we need only consider
the case where the radiations from the various parts of the line start in
the same phase. We get, almost as before,
for the expression of the resultant amplitude corresponding to £.
In order to make use of these results we require a table of the values
of smu/u, and of sirfu/u?. The following will suffice for our purposes: —
TABLE I.
4u
IT
sin u
u
sin2 u
u2
4u
IT
sinu
u
sin2w
u2
4u
IT
sin u
u
sin2u
-tf~
0
+1-0000
1-0000
6
-•2122
•0450
12
•oooo
•oooo
1
•9003
•8105
7
- -1286
•0165
13
-•0692
•0048
2
•6366
•4053
8
•oooo
•oooo
14
- -0909
•0083
3
•3001
•0901
9
+ •1000
•0100
15
-•0600
•0036
4
•0000
•oooo
10
•1273
•0162
16
•oooo
•oooo
5
- -1801
•0324
11
•0818
•0067
1896]
WITH SPECIAL REFERENCE TO THE MICROSCOPE.
245
When we have to deal with a single point or a single line only, this
table gives directly the distribution of light in the image, u being equated
to Trga/Xf. The illumination first vanishes when u = ir, or ^/f=\/a.
On a former occasion* it has been shown that a self-luminous point
or line at u = — IT is barely separated from one at u = 0. It will be of
interest to consider this case under three different conditions as to phase-
relationship : (i) when the phases are the same, as will happen when the
illumination is by plane waves incident perpendicularly; (ii) when the
phases are opposite ; and (iii) when the phase -difference is a quarter period,
which gives the same result for the intensity as if the apertures were self-
luminous. The annexed table gives the numerical values required. In
TABLE II.
4w
sinu sin(M + 7r)
sin u sin (u + IT)
/ Isin2« sin2(« + ir)j
V n?~+ («+-)2 I
7T
M M + 7T
u 11 + ir
-4...
+ 1-0000
-1-0000
+ 1-000
-3...
+ 1-2004
- -6002
+ -949
-2...
+ 1-2732
•oooo
+ -900
-1...
+ 1-2004
+ '6002
+ -949
0...
+ 1-0000
+ 1-0000
+ 1-000
1...
+ -7202
+ 1-0804
+ '918
2...
+ -4244
+ -8488
+ -671
3...
+ -1715
+ -4287
+ -326
4...
•0000
•oooo
•000
5...
- -0800
- -2801
- -206
6...
- -0849
- -3395
- -247
7...
- -0468
- -2105
- -152
8...
•0000
•oooo
•000
9...
+ -0308
+ -1693
+ -122
10...
+ -0364
+ -2183
+ -156
11...
+ -0218
+ -1419
+ -101
12...
•0000
•oooo
•000
cases (i) and (iii) the resultant amplitude is symmetrical with respect to
the point U = — ^TT midway between the two geometrical images; in case (ii)
the sign is reversed, but this of course has no effect upon the intensity.
Graphs of the three functions are given in Fig. 4, the geometrical images
being at the points marked — TT and 0. It will be seen that while in case (iii),
relating to self-luminous points or lines, there is an approach to separation,
* Phil. Mag. Vol. vin. p. 266, 1879. [Vol. i. p. 420.]
246
ON THE THEORY OF OPTICAL IMAGES,
[222
nothing but an accurate comparison with the curve due to a single source
would reveal the duplicity in case (i). On the other hand, in case (ii),
where there is a phase-difference of half a period between the radiations, the
separation may be regarded as complete.
Fig. 4.
Wii/y
In a certain sense the last conclusion remains undisturbed even when
the double point is still closer, and also when the aperture is of any other
symmetrical form, e.g. circular. For at the point of symmetry in the image,
midway between the two geometrical images of the radiant points, the
component amplitudes are necessarily equal in numerical value and opposite
in sign, so that the resultant amplitude or illumination vanishes. For
example, suppose that the aperture is rectangular and that the points or
lines are twice as close as before, the geometrical images being situated at
T, u = 0. The resultant amplitude is represented by f(u), where
.. , __
.(17)
The values of f(u) are given in Table III. They show that the resultant
vanishes at the place of symmetry u = — \ IT, and rises to a maximum at
a point near u = %ir, considerably beyond the geometrical image at u = 0.
Moreover, the value of the maximum itself is much less than before, a
feature which would become more and more pronounced as the points were
1896]
WITH SPECIAL REFERENCE TO THE MICROSCOPE.
247
taken closer. At this stage the image becomes only a very incomplete
representation of the object ; but if the formation of a black line in the
centre of the pattern be supposed to constitute resolution, then resolution
occurs at all degrees of closeness*. We shall see later, from calculations
conducted by the same method, that a grating of an equal degree of closeness
would show no structure at all but would present a uniformly illuminated
field.
TABLE III.
4u
4«
4W
4u
TT
/(«)
TT
/(«)
TT
/(«)
7T
/(»)
-1
+ •00
2
+ •64
5
-•05
8
-•13
0
+ •36
3
+ •48
6
-•21
9
+ •02
1
+ •60
4
+ •21
7
-•23
But before proceeding to such calculations we may deduce by Lagrange's
theorem the interval e in the original object corresponding to that between
u = 0 and u = TT in the image, and thence effect a comparison with a grating
by means of Abbe's theory. The linear dimension (£) of the image cor-
responding to u = TT is given by £ = \fla\ and from Lagrange's theorem
e/ £ = sin /3 / sin a, (If a)
in which a is the " semi-angular aperture," and /3 = a/2/. Thus, corresponding
to U = TT,
The case of a double point or line represented in Fig. 4 lies therefore
at the extreme limit of resolution for a grating in which the period is the
* These results are easily illustrated experimentally. I have used two parallel slits, formed in
films of tin-foil or of chemically deposited silver, of which one is conveniently made longer than
the other. These slits are held vertically and are viewed through a small telescope, provided with
a high-power eye-piece, whose horizontal aperture is restricted to a small width. The distance
may first be so chosen that when backed by a neighbouring flame the double part of the slit just
manifests its character by a faint shadow along the centre. If the flame is replaced by sunlight
shining through a distant vertical slit, the effect depends upon the precise adjustment. When
everything is in line the image is at its brightest, but there is now no sign of resolution of
the double part of the slit. A very slight sideways displacement, in my case effected most
conveniently by moving the telescope, brings in the half-period retardation, showing itself by
a black bar down the centre. An increased displacement, leading to a relative retardation of
three halves of a period, gives much the same result, complicated, however, by chromatic effects.
In conformity with theory the black bar down the image of the double slit may still be
observed when the distance is increased much beyond that at which duplicity disappears under
flame illumination.
For these experiments I chose the telescope, not only on account -of the greater facility of
manipulation which it allows, but also in order to make it clear that the theory is general,
and that such effects are not limited, as is sometimes supposed, to the case of the microscope.
248 ON THE THEORY OF OPTICAL IMAGES, [222
interval between the double points. And if the incidence of the light upon
the grating were limited to be perpendicular, the period would have to be
doubled before the grating could show any structure.
When the aperture is circular, of radius R, the diffraction pattern is
symmetrical about the geometrical image (p = 0, q = 0), and it suffices to
consider points situated upon the axis of £ for which 77 (and q) vanish. Thus
from (12)
rr r + R
C= llcospxdxdy = 2 I cos px V(^2 - a?) doc (18)
JJ J -R
This integral is the Bessel function of order unity, definable by
Ji(z) = - pcos^cos^sin2^ (19)
7T Jo
Thus, if x = R cos $,
°-»**$P (20)
or, if we write u = TT| .
This notation agrees with that employed for the rectangular aperture if we
consider that 2R corresponds with a.
The illumination at various parts of the image of a double point may be
investigated as before, especially if we limit ourselves to points which lie
upon the line joining the two geometrical images. The only difference in
the calculations is that represented by the substitution of 2/j for sine. We
shall not, however, occupy space by tables and drawings such as have been
given for a rectangular aperture. It may suffice to consider the three prin-
cipal points in the image due to. a double source whose geometrical images
are situated at u = 0 and u = — TT, these being the points just mentioned,
and that midway between them at u = — ^TT. The values of the functions
required are
2/x (0)/0 = I'OOOO = V{ I'OOOO).
2J1(7r)/7r = -1812 = V{'03283}.
2«/i7r= -7217 =
In the case (corresponding to i. Fig. 4) where there is similarity of phase,
we have at the geometrical images amplitudes 1-1812 as against 1*4434 at
the point midway between. When there is opposition of phase, the first
becomes + '8188, and the last zerof. When the phases differ by a quarter
* Enc. Brit. " Wave Theory," p. 432. [Vol. in. p. 87.]
t The zero illumination extends to all points upon the line of symmetry.
1896] WITH SPECIAL REFERENCE TO THE MICROSCOPE. 249
period, or when the sources are self-luminous (iii. Fig. 4), the amplitudes at
the geometrical images are V{1'0328} or T0163, and at the middle point
V{1'0418} or 1'0207. The partial separation, indicated by the central de-
pression in curve iii. Fig. 4, is thus lost when the rectangular aperture is
exchanged for a circular one of equal width. It should be borne in mind
that these results do not apply to a double line, which in the case of a
circular aperture behaves differently from a double point.
There is one respect in which the theory is deficient, and the deficiency
is the more important the larger the angular aperture. The formula (7)
from which we start assumes that a radiant point radiates equally in all
directions, or at least that the radiation from it after leaving the object-
glass is equally dense over the whole area of the section. In the case of
telescopes, and microscopes of moderate angular aperture, this assumption
can lead to no appreciable error ; but it may be otherwise when the angular
aperture is very large. The radiation from an ideal centre of transverse
vibrations is certainly not uniform in various directions, and indeed vanishes
in that of primary vibration. If we suppose such an ideal source to be
situated upon the axis of a wide-angled object-glass, we might expect the
diffraction pattern to be less closely limited in that axial plane which includes
the direction of primary vibration than in that which is perpendicular to it.
The result for a double point illuminated by borrowed light would be a
better degree of separation when the primary vibrations are perpendicular
to the line of junction than when they are parallel to it.
Although it is true that complications and uncertainties under this head
are not without influence upon the theory of the microscopic limit, it is not
to be supposed that any considerable variation from that laid down by Abbe
and Helmholtz is admissible. Indeed, in the case of a grating the theory of
Abbe is still adequate, so far as the limit of resolution is concerned ; for, as
Dr Stoney has remarked, the irregularity of radiation in different directions
tells only upon the relative brightness and not upon the angular position of
the spectra. And it will remain true that there can be no resolution without
the cooperation of two spectra at least.
In Table II. and Fig. 4 we have considered the image of a double point
or line as formed by a lens of rectangular aperture. It is now proposed to
extend the calculation to the case where the series of points or lines is
infinite, constituting a row of points or a grating. The intervals are sup-
posed to be strictly equal, and also the luminous intensities. When the
aperture is rectangular, the calculation is the same whether we are dealing"
with a row of points or with a grating, but we have to distinguish according
as the various centres radiate independently, viz., as if they were self-luminous,
250 ON THE THEORY OF OPTICAL IMAGES, [222
or are connected by phase-relations. We will commence with the former
case.
If the geometrical images of the various luminous points are situated
at u = 0, u = ± v, u = ± 2v, «Scc., the expressions for the intensity at any point
u of the field may be written as an infinite series,
sin2M sin2 (u + t>) sin2 (u — v)
~vT" (u + v)2 " (u-v?
Being an even function of u and periodic in period v, (22) may be
expanded by Fourier's theorem in a series of cosines. Thus
-r f \ T T T ,nn.
J'(u) •/, + /! 008 -- + ... + /rcos - •+ ...... ; ......... (23)
and the character of the field of light will be determined when the values of
the constants /„, /,, &c., are known. For these we have as usual
7- 1 [v T / N T T- 2 [v T , . lirru 7
/„ = - I(u)du, Ir = -\ /(w)cos -- dw; ......... (24)
v J0 v Jo ^
and it only remains to effect the integrations. To this end we may observe
that each term in the series (22) must in reality make an equal contribution
to Ir. It will come to the same thing whether, as indicated in (24), we
integrate the sum of the series from 0 to w, or integrate a single term of it,
e.g. the first, from — oo to + oo . We may therefore take
, TT T snM , /OK 0/?N
/» = - — du= — : Ir=- —cos - du. ...(25,26)
Vj_oo U2 V' Wj-oo W2 W
To evaluate (26) we have
+oc sin2wcossw 7 r+°° 1 cZ , . „ . ,
- - - du = I - j- (sm2 u cos SM) rfw,
-oo w2 ) _<» udu^
and
.7 „ 2-4-s 2 _ s
-T- (sin2w cos su) = — -= sin su -\ -- - — sin (2 + s) u -\ -- - — sin (2 — s) u ;
so that by (15) (s being positive)
the minus sign being taken when 2 — s is negative.
Hence
according as w exceeds or falls short of rir.
1896] WITH SPECIAL EEFERENCE TO THE MICROSCOPE. 251
We may now trace the effect of altering the value of v. When v is large,
a considerable number of terms in the Fourier expansion (23) are of import-
ance, and the discontinuous character of the luminous grating or row of points
is fairly well represented in the image. As v diminishes, the higher terms
drop out in succession, until when v falls below 2ir only /0 and 7t remain.
From this point onwards 7j continues to diminish until it also finally dis-
appears when v drops below TT. The field is then uniformly illuminated,
showing no trace of the original structure. The case v = TT is that of Fig. 4,
and curve iii. shows that at a stage when an infinite series shows no struc-
ture, a pair of luminous points or lines of the same closeness are still in
some degree separated. It will be remembered that v = TT corresponds to
e = l\/sin a, e being the linear period of the original object and a the semi-
angular aperture.
We will now pass on to consider the case of a grating or row of points
perforated in an opaque screen and illuminated by plane waves of light. If
the incidence be oblique, the phase of the radiation emitted varies by equal
steps as we pass from one element to the next. But for the sake of
simplicity we will commence with the case of perpendicular incidence, where
the radiations from the various elements all start in the same phase. We
have now to superpose amplitudes, and not as before intensities. If A be
the resultant amplitude, we may write
, _ sin u sin (u + v) sin (u — v)
u u + v u — v
2?rw
= A0+A1coa - + . . . + Ar cos - + ................ (28)
When v is very small, the infinite series identifies itself more and more
nearly with the integral
1 r+°° sin u -. . TT
— du, viz. — .
V J _oo U V
In general we have, as in the last problem,
1 f+cc sin u -. 2 f+co sin u 2-rrru .
A9 — -l - du; Ar=- —cos -- du: ...... (29)
Wj-oo U V J -x, U V
so that A0 = TT/V. As regards Ar, writing s for Z-rrr/v, we have
lprin(H.,)« + rin(l-.). ;
V J _«, U V ^ '
the lower sign applying when (1 — s) is negative. Accordingly,
4(«)-£{i+2oM — +2ooB — *...}., ............ (30)
V [ V V )
the series being continued so long as 2?rr < v.
252 ON THE THEORY OF OPTICAL IMAGES, [222
If the series (30) were continued ad infinitum, it would represent a
discontinuous distribution, limited to the points (or lines) u = 0, u = + v,
u = ± 2v, &c., so that the image formed would accurately correspond to the
original object. This condition of things is most nearly realised when v is
very great, for then (30) includes a large number of terms. As v diminishes
the higher terms drop out in succession, retaining however (in contrast with
(27)) their full value up to the moment of disappearance. When v is less
than 2?r, the series is reduced to its constant term, so that the field becomes
uniform. Under this kind of illumination, the resolving-power is only half
as great as when the object is self-luminous.
These conclusions are in entire accordance with Abbe's theory. The first
term of (30) represents the central image, the second term the two spectra
of the first order, the third term the two spectra of the second order, and
so on. Resolution fails at the moment when the spectra of the first order
cease to cooperate, and we have already seen that this happens for the case
of perpendicular incidence when v = 2?r. The two spectra of any given order
fail at the same moment.
If the series stops after the lateral spectra of the first order,
, ..................... (31)
showing a maximum intensity when u = 0, or \v, and zero intensity when
u = %v, or ft;. These bands are not the simplest kind of interference bands.
The latter require the operation of two spectra only ; whereas in the present
case there are three — the central image and the two spectra of the first
order.
We may now proceed to consider the case when the incident plane waves
are inclined to the grating. The only difference is that we require now to
introduce a change of phase between the image due to each element and its
neighbour. The series representing the resultant amplitude at any point u
may still be written
u + v
For perpendicular incidence m = 0. If 7 be the obliquity, e the grating-
interval, \ the wave-length,
(33)
The series (32), as it stands, is not periodic with respect to u in period v,
but evidently it can differ from such a periodic series only by the factor eimu.
1896] WITH SPECIAL REFERENCE TO THE MICROSCOPE. 253
The series
e-imu sjn u e-im (u+v) 8jn (M + 3,)
u u + v
e~im <»-*> sin (u - v) e~im(u+2v) sin (u + 2v)
— V
is truly periodic, and may therefore be expanded by Fourier's theorem in
periodic terms :
(34) = A0 + iBQ + (A, + iB,) cos (2-n-u/v) + (C, + iD,) sin (2-rru/v) + . ..
+ (Ar + iBr} cos (2r7ru/v) + (Cr + iDr) sin (2nru/v) + (35)
As before, if s = 2r7r/v,
f +°° e~imu sin u cos su ,
$v(Ar + iBr)=\ — du',
so that Br = Q, while
f + °° cos mu sin u cos su 7 /0 «,
%v.Ar=l — du (36)
In like manner Cr = 0, while
sinm^sin^sin^ (37)
In the case of the zero suffix
•°° cos mu sin u
n / cos mi* sn u, /oox
0 = 0, vA0= - — du ................ (38)
When the products of sines and cosines which occur in (36) &c. are
transformed in a well-known manner, the integration may be effected by
(15). Thus
cos mu sin u cos su = | {sin (1 + m + s) u + sin (1 — m — s) u
+ sin (1 + m — s) u + sin (1 — m + s) u} ;
so that
...(39)
where each symbol such as [1 + m + s] is to be replaced by + 1, the sign
being that of (1 + m + s). In like manner
[I + m + s]-[l-m-s]}. ...(40)
The rth terms of (35) are accordingly
|L|ei«u([i + m + s] + [i _m_s])+ er*»([l +m-s] + [1 -m + s])};
or for the original Series (32),
. ...(41)
254 ON THE THEOEY OF OPTICAL IMAGES, [222
For the term of zero order,
A0e™ = ei™([I+m] + [I-m]) ................ (42)
From (41) we see that the term in ei(m+g)u vanishes unless (ra + s) lies
between + 1, and that then it is equal to TT/V ,ei(m+g}u; also that the term in
ei(m-s)u vanishes unless (m-s) lies between ± 1, and that it is then equal to
TT/V . el(m~s]u. In like manner the term in eimu vanishes unless m lies between
±1, and when it does not vanish it is equal to Tr/v.eimu. This particular
case is included in the general statement by putting s = 0.
The image of the grating, or row of points, expressed by (32), is thus
capable of representation by the sum of terms
TT/V . [eimu + ei(m+s^u + ei(m-gi)u + e^m+s^}u + ...} ............ (43)
where sl = 2'rr/v, s2 = 4nr/v, &c., every term being included for which the
coefficient of u lies between ± 1. Each of these terms corresponds to a
spectrum of Abbe's theory, and represents plane progressive waves inclined
at a certain angle to the plane of the image. Each spectrum when it occurs
at all contributes equally, and it goes out of operation suddenly. If but one
spectrum operates, the field is of uniform brightness. If two spectra operate,
we have the ordinary interference bands due to two sets of plane waves
crossing one another at a small angle of obliquity*.
Any consecutive pair of spectra give the same interference bands, so far
as illumination is concerned. For
ZT }gtM[m+2r7r/»] -J- ei«[m+2(r+l)jr/o]J _ ^7r cog ^f gi»[i»+2 (r+i) JT/W]
V l V V
of which the exponential factor influences only the phase.
In (43) the critical value of v for which the rth spectrum disappears is
given by, when we introduce the value of m from (33),
or, since (as we have seen)
e (sin 7 + sin a) = + r\ ......................... (45)
This is the condition, according to elementary theory, in order that the
rays forming the spectrum of the rth order should be inclined at the angle
a, and so (Fig. 2) be adjusted to travel from A to B, through the edge of
the lens L.
* Enc. Brit. "Wave Theory," p. 425. [Vol. m. p. 59.]
1896] WITH SPECIAL REFERENCE TO THE MICROSCOPE. 255
The discussion of the theory of a rectangular aperture may here close.
This case has the advantage that the calculation is the same whether the
object be a row of points or a grating. A parallel treatment of other forms
of aperture, e.g. the circular form, is not only limited to the first alternative,
but applies there only to those points of the field which lie upon the
line joining the geometrical images of the luminous points. Although the
advantage lies with a more general method of investigation to be given
presently, it may be well to consider the theory of a circular aperture as
specially deduced from the formula (21) which gives the image of a single
luminous centre.
If we limit ourselves to the case of parallel waves and perpendicular
incidence, the infinite series to be discussed is
+ ...... (46)
u u + v u — v u + 2v
where w = 7rf.2R/\/. .............................. (47)
Since -A is necessarily periodic in period v, we may assume
A (u) = A0 + A j cos (Ztru/v) + . . . + Ar cos (Zrirujv) + ...; ...... (48)
and, as in the case of the rectangular aperture,
1 r + xJ1(M) , 2 [ + XJ1(U) 2T7TU ,
A0 = -l -=±**du, Ar = - -LA-^ cos -- du ....... (49)
V J -«> U V J -x U V
These integrals may be evaluated. If a and b be real, and a be positive *,
(50)
Multiplying by bdb and integrating from 0 to b, we find
V(* + »)-.
o
In this we write 6 = 1, a = is, where s is real. Thus
If s2 > 1, we must write i'V(s2 — 1) for V(l — s2). Hence, if s < 1,
= v(1_ rJ-Wsn-j, ... (62>68)
J0 x
while, if s > 1,
rjj (x) cos sx 1 _ f00 J1(x)sinsx .
-^-^ - dx = 0, -£-* - dx=-M-I) + s. ...(54,55)
& J Q X
* Gray and Mathews, BesseVs Functions, 1895, p. 72.
256 ON THE THEORY OF OPTICAL IMAGES, [222
We are here concerned only with (52), (54), and we conclude that A0 = 2/v,
and that
**, or 0, (56)
according as s is less or. greater than 1, viz. according as Ir-rr is less or greater
than v.
If we compare this result with the corresponding one (30) for a rect-
angular aperture of equal width (2R = a), we see that the various terms
representing the several spectra enter or disappear at the same time;
but there is one important difference to be noted. In the case of the
rectangular aperture the spectra enter suddenly and with their full effect,
whereas in the present case there is no such discontinuity, the effect of a
spectrum which has just entered being infinitely small. As will appear
more clearly by another method of investigation, the discontinuity has its
origin in the sudden rise of the ordinate of the rectangular aperture from
zero to its full value.
In the method referred to the form of the aperture is supposed to remain
symmetrical with respect to both axes, but otherwise is kept open, the
integration with respect to x being postponed. Starting from (12) and
considering only those points of the image for which 77 and q in equation
(8) vanish, we have as applicable to the image of a single luminous source
C = ffcospxdxdy = 2fy cospxdx (57)
in which *2y denotes the whole height of the aperture at the point x. This
gives the amplitude as a function of p. If there be a row of luminous points,
from which start radiations in the same phase, we have an infinite series of
terms, similar to (57) and derived from it by the addition to p of positive
and negative integral multiples of a constant (p^) representing the period.
The sum of the series A (p) is necessarily periodic, so that we may write
and, as in previous investigations, we may take
Ar= I ^Ccosspdp, (59)
s (not quite the same as before) standing for 2irjr/pli and a constant factor
being omitted. To ensure convergency we will treat this as the limit of
(60)
the sign of the exponent being taken negative, and h being ultimately made
to vanish. Taking first the integration with respect to p, we have
1896] WITH SPECIAL REFERENCE TO THE MICROSCOPE. 257
= hydx
'
and thus
in which h is to be made to vanish. In the limit the integrals receive
sensible contributions only from the neighbourhoods of x = + s ; and since
~
we get Ar = 7r(yx=^ + yx=+g) = 27ryx=s ................... (62)
From (62) we see that the occurrence of the term in Ar, i.e. the appear-
ance of the spectrum of the rth order, is associated with the value of a
particular ordinate of the object-glass. If the ordinate be zero, i.e. if the
abscissa exceed numerically the half- width of the object-glass, the term in
question vanishes. The first appearance of it corresponds to
in which a is the entire width of the object-glass and £ the linear period in
the image. By (17 a),
! e sn a e sn a
so that the condition is, as before,
e sin a = r\.
When Ar has appeared, its value is proportional to the ordinate at x = s.
Thus in the case of a circular aperture (a = 2R) we have
} ...................... (63)
The above investigation relates to a row of luminous points emitting
light of the same intensity and phase, and it is limited to those points of
the image for which 17 (and q) vanish. If the object be a grating radiating
under similar conditions, we have to retain cosqy in (12) and to make an
integration with respect to q. Taking this first, and introducing a factor
e±kq, we have
(64)
This is now to be integrated with respect to y between the limits — y
and + y. If this range be finite, we have
.(65)
17
258 ON THE THEORY OF OPTICAL IMAGES, [222
independent of the length of the particular ordinate. Thus
(66)
= l
J
the integration with respect to x extending over the range for which y is
finite, that is, over the width of the object-glass. If this be 2R, we have
(67)
From (67) we see that the image of a luminous line, all parts of which
radiate in the same phase, is independent of the form of the aperture of the
object-glass, being, for example, the same for a circular aperture as for a
rectangular aperture of equal width. This case differs from that of a self-
luminous line, the images of which thrown by circular and rectangular
apertures are of different types*.
The comparison of (67) with (20), applicable to a circular aperture, leads
to a theorem in Bessel's functions. For, when q is finite,
so that, setting 22 = 1, we get
The application to a grating, of which all parts radiate in the same phase,
proceeds as before. If, as in (58), we suppose
., (70)
we have Ar= I CiCosspdp', (71)
from which we find that Ar is 4?r2 or 0, according as the ordinate is finite or
not finite at x = s. The various spectra enter and disappear under the same
conditions as prevailed when the object was a row of points ; but now they
enter discontinuously and retain constant values, instead of varying with the
particular ordinate of the object-glass which corresponds to x = s.
We will now consider the corresponding problems when the illumination
is such that each point of the row of points or of the grating radiates in-
dependently. The integration then relates to the intensity of the field as
due to a single source.
* Enc. Brit. " Wave Theory," p. 434. [Vol. in. p. 92.]
t This may be verified by means of Neumann's formula (Gray and Mathews, BesseVs Functions
(70), p. 27).
1896] WITH SPECIAL REFERENCE TO THE MICROSCOPE. 259
By (9), (10), (11), the intensity 72 at the point (p, q) of the field, due to
a single source whose geometrical image is situated at (0, 0) is given by
Xy2 72 = ijrjeos (px + qy) dx<lyY + {//sin (px + qy) dxdy}*
= //cos (px + qy') dx'dy' x //cos (px + qy) dxdy
+ //sin (px' + qy') dxdy' x //sin (px + qy) dxdy
= ffflcos{p(x' -x) + q(y'-y)}dxdydx'dy', ............... (72)
the integrations with respect to x', y', as well as those with respect to x, y
being over the area of the aperture.
In the present application to sources which are periodically repeated,
the term in cos sp of the Fourier expansion representing the intensity
at various points of the image has a coefficient found by multiplying (72)
by cos sp and integrating with respect to p from p = — cc to p = + oo . If
the object be a row of points, we may take q = 0 ; if it be a grating, we
have to integrate with respect also to q from q = — oo to q = + <x> .
Considering the latter case, and taking first the integrations with respect
to p, q, we introduce the factors e*hp'fkq, the plus or minus being so chosen as
to make the elements of the integral vanish at infinity. After the operations
have been performed, h and k are to be supposed to vanish*. The integra-
tions are performed as for (60), (64), and we get the sum of the two terms
denoted by
We have still to integrate with respect to dxdy dx'dy'. As in (65), since the
range for y' always includes y,
and we are left with
[([ 2-rrhdxdydx' 7
JJJ K+(x'-x±sY
If s were zero, the integration with respect to x would be precisely
similar; but with s finite it will be only for certain values of x that
(x — x ± s) vanishes within the range of integration. Unless this evan-
escence takes place, the limit when h vanishes becomes zero. The effect
of the integration with respect to x' is thus to limit the range of the
subsequent integration with respect to x. The result may be written
(75)
* The process is that employed by Stokes in his evaluation of the integral intensity, Edin.
Trans, xx. p. 317 (1853). See also Enc, frit, " Wave Theory," p. 431. [Vol. ni. p. 86.]
17—2
260 ON THE THEORY OF OPTICAL IMAGES. [222
upon the understanding that, while the integration for y ranges over the
whole vertical aperture, that for x is limited to such values of x as bring
x + s (as well as x itself) within the range of the horizontal aperture. The
coefficient of the Fourier component of the intensity involving cos sp, or
cos(2r7rp/p1), is thus proportional to a certain part of the area of the aper-
ture. Other parts of the area are inefficient, and might be stopped off
without influencing the result.
The limit to resolution, corresponding to r = 1, depends only on the width
of the aperture, and is therefore for all forms of aperture the same as for
the case of the rectangular aperture already fully investigated.
If the object be a row of points instead of a row of lines, q = 0, and there
is no integration with respect to it. The process is nearly the same as above,
and the result for the coefficient of the rth term in the Fourier expansion is
proportional to fy*dx, instead of fydx, the integration with respect to x being
over the same parts of the aperture as when the object was a grating. The
application to a circular aperture would lead to an evaluation of
' J^ (u) cos su ,
^ du.
223.
THEORETICAL CONSIDERATIONS RESPECTING THE SEPARA-
TION OF GASES BY DIFFUSION AND SIMILAR PROCESSES.
[Philosophical Magazine, XLII. pp. 493 — 498, 1896.]
THE larger part of the calculations which follow were made in connexion
with experiments upon the concentration of argon from the atmosphere by
the method of atmolysis*. When the supply of gas is limited, or when it
is desired to concentrate the lighter ingredient, the conditions of the question
are materially altered; but it will be convenient to take first the problem
which then presented itself of the simple diffusion of a gaseous mixture
into a vacuum, with special regard to the composition of the residue. The
diffusion tends to alter this composition in the first instance only in the
neighbourhood of the porous walls ; but it will be assumed that the forces
promoting mixture are powerful enough to allow of our considering the
composition to be uniform throughout the whole volume of the residue,
and variable only with time, on account of the unequal escape of the
constituent gases.
Let x, y denote the quantities of the two constituents of the residue at
any time, so that — dx, — dy are the quantities diffused out in time dt. The
values of dxjdt, dy/dt will depend upon the character of the porous partition
and upon the actual pressure ; but for our present purpose it will suffice to
express dyjdx, and this clearly involves only the ratios of the constituents
and of their diffusion rates. Calling the diffusion rates //,, v, we have
In this equation x, y may be measured on any consistent system that
may be convenient. The simplest case would be that in which the residue
is maintained at a constant volume, when x, y might be taken to represent
* Kayleigh and Ramsay, Phil. Trans. CLXXXVI. p. 206 (1895). [Vol. iv. p. 130.]
262 ON THE SEPARATION OF GASES [223
the partial pressures of the two gases. But the equation applies equally well
when the volume changes, for example in such a way as to maintain the total
pressure constant.
The integral of (1) is
yil»=Co£l», (2)
where G is an arbitrary constant, or
If X, Y be simultaneous values of x, y, regarded as initial,
#-(*)""*• w
so that x = X(fj^J (5)
In like manner ^=F(z7FJ (6)
If we write ^= = r, (7)
r represents the enrichment of the residue as regards the second constituent,
and we have from (5), (6),
an equation which exhibits the relation between the enrichment and the
ratio of the initial and final total quantities of the mixture.
From (8), or more simply from (4), we see that as x diminishes with time
the enrichment tends to zero or infinity, indicating that the residue becomes
purer without limit, and this whatever may be the original proportions. Thus
if the first gas (x) be the more diffusive (//, > v), the exponent on the right
of (4) is negative; and this indicates that r becomes infinite, or that the
first gas is ultimately eliminated from the residue. When the degree of
enrichment required is specified, an easy calculation from (8) gives the degree
to which the diffusion must be carried.
In Graham's atmolyser the gaseous mixture is caused to travel along a
tobacco-pipe on the outside of which a vacuum is maintained. If the
passage be sufficiently rapid to preclude sensible diffusion along the length
of the pipe, the circumstances correspond to the above calculation ; but the
agreement with Graham's numbers is not good. Thus in one case given by
him* of the atmolysis of a mixture containing equal volumes of oxygen
and hydrogen, we have
Y/X = l, y\x = 92-78/7-22,
* Phil. Trans. Vol. CLIII. p. 403 (1863).
1896] BY DIFFUSION AND SIMILAR PROCESSES. 263
so that r = 13 nearly. Thus, if in accordance with the view usually held
fjb/v = 4, we should have from (8)
i x 13- * + * x 13-* = -229 ;
so that a reduction of the residue to '229 of the initial quantity should have
effected the observed enrichment. The initial and final volumes given by
Graham are, however, 7'5 litres and '45 litre, whose ratio is '06. The inferior
efficiency of the apparatus may have been due to imperfections in the
walls or joints of the pipes. Such an explanation appears to be more
probable than a failure of the law of independent diffusion of the component
gases upon which the theoretical investigation is founded.
In the concentration of argon from a mixture of argon and nitrogen we
have conditions much less favourable. In this case
If an enrichment of 2 : 1 is required and if the original mixture is
derived from the atmosphere by removal of oxygen, the equation is
= -99x2-6'13 + -01 x -2-8-13 = -0142 + "0029 = "0171,
y
A. +
expressing the reduction needed. The results obtained experimentally (loc.
cit.) were inferior in this case also.
When the object is the most effective separation of the components of a
mixture, it is best, as supposed in the above theory, to maintain a vacuum
on the further side of the porous wall. But we have sometimes to consider
cases where the vacuum is replaced by an atmosphere of fixed composition,
as in the well-known experiment of the diffusion of hydrogen into air through
a porous plug. We will suppose that there are only two gases concerned
and that the volume inside is given. The symbols x, y will then denote the
partial pressures within the given volume, the constant partial pressures
outside being a, /3. Our equations may be written
(9)
or on integration
, y = /3 + De-vt, .................. (10)
C, D being arbitrary constants.
After a sufficient time x, y reduce themselves respectively to a, ft, as was
to be expected.
The constants /i, v are not known beforehand, depending as they do upon
264 ON THE SEPARATION OF GASES [223
the specialities of the apparatus as well as upon the quality of the gases. If
we eliminate t, we get
y-ft = E(x-*yi*, ........................... (11)
in which only the ratio v/p is involved.
As a particular case suppose that initially the inside volume is occupied
by one pure gas and the outside by another, the initial pressures being unity.
Then in (10)
« = 0, /8 = 1, (7=1, D = -l;
we have x = e-*t, y=\-e~vt, ........................ (12)
and a; + y =1 + 6-^-6-* ........................ (13)
gives the total internal pressure. When this is a maximum or minimum,
e(iL-*)t — piv> anci the corresponding value is
a*)
Thus in the case of hydrogen escaping into oxygen, p/v = 4, and
# + 2/=l-3x4-J = -528,
the minimum being about half the initial pressure*.
Returning now to the separation of gases by diffusion into a vacuum,
let us suppose that the difference between the gases is small, so that
(v — /*)/(*, = K, a small quantity, and that at each operation one-half the total
volume of the mixture is allowed to pass. In this case (8) becomes
X - Y —
= T~K + T " =
so that
This gives the effect of the operation in question upon the composition of
the residual gas. If s denote the corresponding symbol for the transmitted
gas, we have
_ -_
(X-x)/X \-xlX~\-x\X~ ' 1-asjX
approximately, since r is nearly equal to unity. Accordingly
1 1
2-
= r nearly,
so that approximately s and r are reciprocal operations. For example, if
* The most striking effects of this kind are when nitrous oxide, or dry ammonia gas, diffuse
into the air through indiarubber. I have observed suctions amounting respectively to 53 und 64
centimetres of mercury.
1896]
BY DIFFUSION AND SIMILAR PROCESSES.
265
starting with any proportions we collect the transmitted half, and submit it
to another operation of the same sort, retaining the half not transmitted,
the final composition corresponding to the operations sr is the same (ap-
proximately) as the composition with which we started, and the same also
as would be obtained by operations taken in the reverse order, represented
by rs. A complete scheme* on these lines is indicated in the diagram.
Representing the initial condition by unity, we may represent the result 01
the first operation by
\r + $8, or | (r + s),
in which the numerical coefficient gives the quantity of gas whose character
is specified by the literal symbols. The second set of operations gives in the
first instance
or, after admixture of the second and third terms (which are of the same
quality),
In like manner the result of the third set of operations may be represented
by (j , and (as may be formally proved by "induction") of n sets of
operations by
C-fT <ie>
When we take account of the reciprocal character of r and s, this may be
written
L Ln + n
rn-2 + -
(17)
the number of parts into which the original quantity of gas is divided being
* It differs, however, from that followed by Prof. Ramsay iu his recent researches (Proc. Roy.
Soc. Vol. LX. p. 216, 1896).
266 ON THE SEPARATION OF GASES BY DIFFUSION. [223
n + 1. If n is even, the largest part, corresponding to the middle term, has
the original composition*.
It is to be observed, however, that so far as the extreme concentration of
the less diffusive constituent is concerned these complex operations are
entirely unnecessary. The same result, represented by (£)nrn will be reached
at a single operation by continuing the diffusion until the residue is reduced
to (£)n of the original quantity, when its composition will be that denoted by
rn. And even as regards the extreme member at the other end in which the
more diffusive constituent preponderates, it will be evident that the opera-
tions really required are comparatively simple, the extreme member in each
row being derived solely from the extreme member of the row preceding-f*.
If we abandon the supposition, adopted for simplicity, that the gas is
divided into equal parts at each operation, we may still express the results
in a similar manner. If p, a be the fractions retained and transmitted, then
p + a = 1, and in place of (15) we get
r = Pk (18)
The relation between r and s is
pr+a8=l; (19)
and the various portions into which the gas is divided after n sets of operations
are represented by the various terms of the expansion of
(pr + <rs)n, (20)
the Greek letters and the numerical coefficients giving the quantity of each
portion, and the Roman letters giving the quality. But it must not be for-
gotten that this theory all along supposes the difference of diffusivities to be
relatively small.
* There is here a formal analogy with the problem of determining the probability of a given
combination of heads and tails in a set of n tosses of a coin ; and the result of supposing n infinite
may be traced as in the theory of errors.
t Possibly a better plan for the concentration of the lighter constituent would be diffusion
along a column of easily absorbable gas, e.g. C02. The gas which arrives first at the remote end
is infinitely rich in this constituent. [1902. See Phil. Mag. i. p. 105, 1901.]
224.
THE THEORY OF SOLUTIONS.
[Nature, LV. pp. 253, 254, 1897.]
As some recent viva voce remarks of mine have received an interpretation
more wide than I intended, I shall be glad to be allowed to explain that
when (now several years ago) I became acquainted with the work of
van t' Hoff I was soon convinced of the great importance of the advances
due to him and his followers. The subject has been prejudiced by a good
deal of careless phraseology, and this is probably the reason why some dis-
tinguished physicists and chemists have refused their adhesion. It must be
admitted, further, that the arguments of van t' Hoff are often insufficiently
set out, and are accordingly difficult to follow. Perhaps this remark applies
especially to his treatment of the central theorem, viz. the identification of
the osmotic pressure of a dissolved gas with the pressure which would be
exercised by the gas alone if it occupied the same total volume in the absence
of the solvent. From this follows the formal extension of Avogadro's law to
the osmotic pressure of dissolved gases, and thence by a natural hypothesis
to the osmotic pressure of other dissolved substances, even although they
may not be capable of existing in the gaseous condition. If I suggest a
somewhat modified treatment, it is not that I see any unsoundness in van
t' Hoff's argument, but because of the importance of regarding a matter of
this kind from various points of view.
Let us suppose that we have to deal with an involatile liquid solvent, and
that its volume, at the constant temperature of our operations, is unaltered
by the dissolved gas — a question to which we shall return. We start with
a volume v of gas under pressure p0, and with a volume V of liquid just
sufficient to dissolve the gas under the same pressure, and we propose to find
what amount of work (positive or negative) must be done in order to bring
the gas into solution reversibly. If we bring the gas at pressure p0 into
contact with the liquid, solution takes place irreversibly, but this difficulty
may be overcome by a method which I employed for a similar purpose many
268 THE THEORY OF SOLUTIONS. [224
years ago*. We begin by expanding the gas until its rarity is such that no
sensible dissipation of energy occurs when contact with the liquid is es-
tablished. The gas is then compressed and solution progresses under rising
pressure until just as the gas disappears the pressure rises to p0. The opera-
tions are to be conducted at constant temperature, and so slowly that the
condition never deviates sensibly from that of equilibrium. The process is
accordingly reversible.
In order to calculate the amount of work involved in accordance with the
laws of Boyle and Henry, we may conveniently imagine the liquid and gas to
be confined under a piston in a cylinder of unit cross-section. During the
first stage contact is prevented by a partition inserted at the surface of the
liquid. If the distance of the piston from this surface be x, we have initially
a; = v. At any stage of the expansion (x) the pressure p is given by p =p0vjx,
and the work gained during the expansion is represented by
*"K
p,vlog^,
x being a very large multiple of v. During the condensation, after the
partition has been removed, the pressure upon the piston in a given position
x is less than before. For the gas which was previously confined to the
space x is now partly in solution. If s denote the solubility, the available
volume is practically increased in the ratio x : x + s V, so that the pressure in
position x is now given by
and the work required to be done during the compression is
f* dx x + sV
H0^TF = ^10S^F--
On the whole the work lost during the double operation is
x + sV
and of this the first part must be omitted, as a; is indefinitely great. As
regards the second part, we see that it is zero, since by supposition the
quantity of liquid is such as to be just capable of dissolving the gas, so
that sV = v. The conclusion then is that, upon the whole, there is no gain
or loss of work in passing reversibly from the initial to the final state of
things.
The remainder of the cycle, in which the gas is removed from solution
and restored to its original state, may now be effected by the osmotic process '
* "On the Work that may be gained during the mixing of Gases," Phil. Mag. Vol. XLIX.
p. 811, 1875. [Vol. i. p. 242.]
1897] THE THEOEY OF SOLUTIONS. 269
of van t' Hoff*. For this purpose one "semi-permeable membrane," per-
meable to gas but not to liquid, is introduced just under the piston which
rests at the surface of the liquid. A second, permeable to liquid but not to
gas, is substituted as a piston for the bottom of the cylinder, and may be
backed upon its lower side by pure solvent. By suitable proportional motions
of the two pistons, the upper one being raised through the space v, and the
lower through the space V, the gas may be expelled, the pressure of the gas
retaining the constant value p0, and the liquid (which has not yet been
expelled) retaining a constant strength, and therefore a constant osmotic
pressure P. When the expulsion is complete, the work done upon the lower
piston is PV, and that recovered from the gas is p0v, upon the whole
PV— p0v. Since this process, as well as the first, is reversible, and since the
whole cycle has been conducted at constant temperature, it follows from the
second law of thermo-dynamics that no work is lost or gained during the
cycle, or that PV=p0v. The osmotic pressure P is thus determined, and it
is evident that its value is that of the pressure which the gas, as a gas, would
exert in space V.
The objection may perhaps be taken that the assumption of unaltered
volume of the liquid as the gas dissolves in it unduly limits the application
of the argument. It is true that when finite pressures are in question, an
expansion (or contraction) of the liquid would complicate the results ; but we
are concerned only, or at any rate primarily, with the osmotic pressure of
dilute, solutions. In this case the complications spoken of relate only to the
second order of small quantities, and in our theory are accordingly to be
dismissed.
* Phil. Mag. Vol. xxvi. p. 88, 1888.
225.
OBSERVATIONS ON THE OXIDATION OF NITROGEN GAS.
[Chemical Society's Journal, 71, pp. 181—186, 1897.]
THE observations here described were made in connexion with the
isolation of argon by removal of the nitrogen from air, but they may, perhaps,
possess a wider interest as throwing light upon the behaviour of nitrogen
itself.
According to Davy*, the dissolved nitrogen of water is oxidised to nitrous
(or nitric) acid when the liquid is submitted to electrolysis. " To make the
experiment in as refined a form as possible, I procured two hollow cones of
pure gold containing about 25 grains of water each, they were filled with
distilled water connected together by a moistened piece of amianthus which
had been used in the former experiments, and exposed to the action of
a voltaic battery of 100 pairs. . . . In 10 minutes the water in the negative
tube had gained the power of giving a slight blue tint to litmus paper : and
the water in the positive tube rendered it red. The process was continued
for 14 hours ; the acid increased in quantity during the whole time, and the
water became at last very sour to the taste. . . . The acid, as far as its
properties were examined, agreed with pure nitrous acid having an excess
of nitrous gas" (p. 6).
Further (p. 10), " I had never made any experiments, in which acid
matter having the properties of nitrous acid was not produced, and the
longer the operation the greater was the quantity which appeared. . . .
It was natural to account for both these appearances, from the combination
of nascent oxygene and hydrogene respectively; with the nitrogene of the
common air dissolved in the water."
Davy was confirmed in his conclusion by experiments in which the
* Phil. Trans. 1807, p. 1.
1897] OBSERVATIONS ON THE OXIDATION OF NITROGEN GAS. 271
electrolytic vessels were placed in a vacuum or in an atmosphere of hydrogen.
There was then little or no reddening of the litmus, even after prolonged
action of the battery.
If nitrogen could be oxidised in this way, the process would be a con-
venient one for the isolation of argon, for it could be worked on a large scale
and be made self-acting. But it did not appear at all probable that nitrogen
could take a direct part in the electrolysis. In that case, its oxidation would
be a secondary action, due, perhaps, to the formation of peroxide of hydrogen.
This consideration led me to try the effect of peroxide of sodium on dissolved
nitrogen, but without success. The nitrogen dissolved in 1250 c.c. of tap
water and liberated by boiling, was found to be 19'1 c.c., and it was not
diminished by a previous addition of peroxide of sodium, with or without
acid. Having failed in this direction, I endeavoured to repeat Davy's ex-
periment nearly in its original form. The water was contained in two
cavities bored in a block of paraffin, and connected by a wick of asbestos
which had been previously ignited. By means of platinum terminals con-
nected with a secondary battery, a potential difference of 100 volts was
maintained between the cups. The whole was covered by a glass shade, to
exclude any saline matter that might be introduced from the atmosphere.
But, under these conditions, no difference in the behaviour of litmus when
moistened with water from the two cups could be detected, even after 14
days' exposure to the 100 volts. When, however, the cover was removed, the
litmus responded markedly after a day or two.
The failure of several attempts of this kind lead me to doubt the correct-
ness of Davy's view, that the dissolved nitrogen of water is oxidised during
electrolysis. At any rate, the action is so slow that the process holds out no
promise of usefulness on a large scale.
In the oxidation of nitrogen by gaseous oxygen under the action of
electric discharge, a question arises as to the influence of pressure. If the
mass absorbed were proportional to pressure, or the volume independent of
pressure, the electrical energy expended being the same, it might be desirable
to work with highly condensed gases, in spite of the serious difficulties that
must necessarily be encountered. That pressure would be favourable seems
probable a priori, and is suggested by certain observations of Dr Frankland.
My own early experiments pointed also in the same direction. A suitable
mixture of nitrogen and oxygen, standing in an inverted test-tube over alkali,
was sparked from a Ruhmkorff coil actuated by five Grove cells ; when the
total pressure was about three atmospheres, the mass absorbed was about
three times that absorbed in the same time at the ordinary pressure.
This result made it necessary to proceed to operations upon a larger
scale with the alternate current discharge. Experiments were first tried in
272 OBSERVATIONS ON THE OXIDATION OF NITROGEN GAS. [225
a small vessel (of 250 c.c.), which would be more easily capable of withstand-
ing internal pressure than a larger one. In order to protect the glass, which
at the top was almost in contact with the electric flame, and to promote
absorption of the combined nitrogen, the alkali was used in the form of
a fountain, which struck the glass immediately over the flame, and washed
the whole of the internal surface*. But, to my surprise, preliminary trials,
conducted at atmospheric pressure, showed that this apparatus was not
effective. The rates of absorption were about 1600 c.c. per hour, the runs
themselves being for half-an-hour. About double this rate had already been
obtained with the same electrical appliances and with stationary alkali.
Care having been taken that the quality of the mixture within the working
vessel was maintained throughout the run, the smaller efficiency could only
be connected with the confined space.
As to the reason why a confined space should be unfavourable, it is
difficult to give a decided opinion. Other things being the same, the surface
presented by the alkali will be diminished in a smaller vessel, and the ab-
sorption of the combined nitrogen may consequently be less rapid. But it is
difficult to accept this explanation, in view of the favourable conditions
secured by the use of a fountain. The gases, as they rise from the flame,
impinge directly upon the alkali, which is itself in rapid motion over the
whole internal surface. It would almost seem as if the combined nitrogen,
as it leaves the flame, is not yet ready for absorption, and only becomes so
after the lapse of a certain time. However this may be, the efficiency is in
practice improved by largely increasing the capacity of the working vessel.
A larger bottle, of 370 c.c. capacity, allowed a rate of 2000 c.c. per hour.
A flask of still greater capacity gave 3300 c.c. per hour, whilst with a larger
globe capable of holding 4^ litres, a rate of 6800 c.c. per hour was obtained.
These experiments were all made at atmospheric pressure with a fountain of
alkali and with the electric flame in as nearly as possible a constant condition.
In the case of the smallest vessel, it was thought that the separation of the
platinum terminals may have been insufficient for the best effect, but the
loss due to this cause must have been relatively small. Electrical instruments
connected with the primary circuit of the Ruhmkorff gave readings of 10
amperes and 41 volts.
When the comparatively small vessel of 370 c.c. was used at a pressure
of about one additional atmosphere, the volume absorbed was about the same
as in the experiments with the same vessel at atmospheric pressure, thus
indicating a double efficiency. This increased efficiency is, however, of no
practical importance, inasmuch as a higher efficiency still can be obtained at
atmospheric pressure by use of a larger vessel. In order to clear up the
question, it was necessary to compare the efficiencies in a large vessel at
* Rayleigh and Ramsay, Phil. Trans. 1895, p. 217. [Vol. iv. p. 162.]
1897] OBSERVATIONS ON THE OXIDATION OF NITROGEN GAS. 273
different pressures, an operation involving considerable difficulty and even
danger.
For this purpose, a glass globe, nearly spherical in form, and having
a capacity of about 7 litres, was employed. The extra pressure was nearly
an atmosphere a,nd was obtained by gravity, the feed and return pipes for
the alkaline fountain, as well as the pipe for the supply of water to the gas-
holder, being carried to a higher level than that at which the rest of the
apparatus stood. The rate of absorption (reduced to atmospheric pressure)
was 6880 c.c. per hour. Experiments conducted at atmospheric pressure
gave as a mean 6600 c.c.
In order to examine still further the influence of pressure, two ex-
periments were tried under a total pressure of half an atmosphere. The
reduced numbers were 5600, 5700 c.c. per hour. From these results, it
would appear that the influence of pressure is slightly favourable. But, in
comparing the results for one atmosphere and for half an atmosphere, it
should be remembered that, in the latter case, aqueous vapour is responsible
for a sensible part of the total pressure. At any rate, the results are much
more nearly independent of pressure than proportional to pressure ; so that
the cases of large and small vessels are sharply distinguished, pressure ap-
pearing to be advantageous only where the space is too confined to admit
of the best efficiency at a given pressure being reached.
Not sorry to be relieved from the obligation of designing a large scale
apparatus to be worked at a high pressure, such as 20 or 100 atmospheres,
I reverted to the ordinary pressure, and sought to obtain a high rate of
absorption by employing a powerful electric flame contained in a large vessel
whose walls were washed internally by an alkaline fountain. The electrical
arrangements have been the subject of much consideration, and require to be
different from what would naturally be expected. Since the voltage on the
final platinums during discharge is only from 1600 to 2000, as measured by
one of Lord Kelvin's instruments, it might be supposed that a commercial
transformer, transforming from 100 volts to 2400 volts, would suffice for the
purpose. When, however, the attempt is made, it is soon discovered that
such an arrangement is quite unmanageable. When, after some difficulty,
the arc is started, it is found that the electrical conditions are unstable.
Things may go well for a time, but after perhaps some hours the current
will rise and the platinums will become overheated and may melt. Even
when two transformers were employed, so connected as to give on open
secondary circuit nearly 4800 volts, the conditions were not steady enough
for convenient practice. The transformer used in the experiments about to
be described is by Messrs Swinburne, and is insulated with oil. On open
secondary, the voltage is nearly 8000*, but it falls to 2000 or less when the
* Probably 6000 would have sufficed.
H. IV. 18
274
OBSERVATIONS ON THE OXIDATION OF NITROGEN GAS.
[225
discharge is running. Even with this transformer, it was necessary to include
in its primary (thick wire) circuit a self-induction coil, provided with, a core
consisting of a bundle of iron wires, and adjustable in position. As finally
used, the adjustment was such that the electromotive force actually operative
on the primary was only about 30 volts out of the 100 volts available at the
mains of the public supply. This reduction of voltage does not, at any rate
from a theoretical point of view, involve any loss of economy, and some such
reduction seems to be essential to steadiness. Under these conditions, the
current taken amounted to 40 amperes.
It is scarcely necessary to say that the watts actually delivered to the
primary circuit of the transformer are less than the number (1200) derived
by multiplication of volts and amperes. From some experiments made
under similar conditions*, I have found that the factor of reduction — the
cosine of the angle of lag — is about two-thirds, so that the watts taken in
the above arrangement are about 800, representing a little more than a
horse-power.
The working vessel, A, was of glass, spherical in form, and of 50 litres
capacity. The neck was placed downwards, and
was closed by a large rubber stopper, through
which five tubes of glass penetrated. Two tubes
of substantial construction carried the electrodes,
B, C, arranged much as in a former apparatus f ;
two more, F and E, were required for the supply
tube of the fountain and for the drain of liquid,
whilst the fifth, D, was for the supply of gas.
The external drowning of the vessel, formerly
necessary, was now dispensed with; but a suit-
able cooling arrangement for the alkali (some-
thing like the worm of a condenser) had to be
provided to obviate excessive accumulation of
heat.
As the solution of alkali circulated entirely
in the closed apparatus, it could lose none of its
dissolved argon. It was maintained in circula-
tion by a small centrifugal pump constructed of iron and driven from an
electric motor.
The mixed gases (about 11 parts of oxygen to 9 parts of air) were
supplied from a large gas-holder ; but an auxiliary holder was also necessary
in order to observe the rate of absorption. When the rate became un-
*• I hope shortly to publish an account of the method employed. [Phil. Mag. XLIII. p. 343 ;
Art. 229 below.]
t Bayleigh and Ramsay, Phil. Trans. 1895, p. 218. [Vol. iv. p. 163.]
1897] OBSERVATIONS ON THE OXIDATION OF NITROGEN GAS. 275
satisfactory, the mixed gas in the working vessel was analysed and the
necessary rectification effected.
In the earlier stages of the operation, the rate of absorption was about
21 litres per hour, and this, by proper attention, could be maintained without
much loss until the accumulation of argon began to tell. If we take 20 litres
as corresponding to 800 watts, we have 25 c.c. per watt-hour, an efficiency
not very different from that found in operations on a much smaller scale.
The present apparatus works about three times as fast as the former one,
in which the vessel was smaller and the alkali stationary. It is also more
interesting to watch, as the electric flame is fully exposed to view. On the
other hand, it is more complicated, owing to the use of a circulating pump,
and probably requires closer attention. A failure of the fountain whilst the
flame was established would doubtless soon lead to a disaster.
I have been efficiently aided throughout by Mr Gordon, who has not only
fitted the apparatus, but has devised many of the contrivances necessary to
meet the ever-recurring difficulties which must be expected in work of this
character.
18—2
226.
ON THE PASSAGE OF ELECTRIC WAVES THROUGH TUBES,
OR THE VIBRATIONS OF DIELECTRIC CYLINDERS.
[Philosophical Magazine, XLIII. pp. 125—132, 1897.]
General Analytical Investigation.
THE problem here proposed bears affinity to that of the vibrations of
a cylindrical solid treated by Pochhammer* and others, but when the
bounding conductor is regarded as perfect it is so much simpler in its
conditions as to justify a separate treatment. Some particular cases of it
have already been considered by Prof. J. J. Thomson f. The cylinder is
supposed to be infinitely long and of arbitrary section ; and the vibrations
to be investigated are assumed to be periodic with regard both to the
time (t) and to the coordinate (2) measured parallel to the axis of the
cylinder, i.e., to be proportional to ei(mz+pt).
By Maxwell's Theory, the components of electromotive intensity in the
dielectric (P, Q, R) and those of magnetic induction (a, b, c) all satisfy
equations such as
d?R d?R &R_I_cPR
dtf dy*+ dz* " F2 dff' '
V being the velocity of light ; or since by supposition
R=0, ........................ (2)
where k2 = p*/V*-m* ................................. (3){
* Crette, Vol. xxxi. 1876.
t Recent Researches in Electricity and Magnetism, 1893, § 300.
J The fc2 of Prof. J. J. Thomson (loc. cit. % 262) is the negative of that here chosen for
convenience.
1897] ON THE PASSAGE OF ELECTRIC WAVES THROUGH TUBES, ETC. 277
The relations between P, Q, R and a, b, c are expressed as usual by
da_d_Q_dR
dt ~ dz dy"
and two similar equations ; while
da db dc dP dQ dR
-7 — r~3 — r T~ = v, j — 7 — =— = 0 (O. O)
dx dy dz dx dy dz
The conditions to be satisfied at the boundary are that the components of
electromotive intensity parallel to the surface shall vanish. Accordingly
dx/ds, dy/ds being the cosines of the angles which the tangent (ds) at any
point of the section makes with the axes of x and y.
Equations (2) and (7) are met with in various two-dimensional problems
of mathematical physics. They are the equations which determine the free
transverse vibrations of a stretched membrane whose fixed boundary coincides
with that of the section of the cylinder. The quantity k'2 is limited to certain
definite values, k-?, k*, ..., and to each of these corresponds a certain normal
function. In this way the possible forms of R are determined. A value of R
which is zero throughout is also possible.
With respect to P and Q we may write
D dd> d^lr ~ d<f> d~^r /n _. _ .
Jr = -j — I — ;— , (J = j j — ', (y, 1U1
dx dy dy dx
where </> and \Jr are certain functions, of which the former is given by
dP dQ dR
VSA = -T-+ T^=--r- = -imfi (11)
dx dy dz
There are thus two distinct classes of solutions ; the first dependent upon <£,
in which R has a finite value, while \|r = 0 ; the second dependent upon i/r, in
which R and <f> vanish.
For a vibration of the first class we have
P=d<j>ldz, Q=d<f>/dy, (12)
and (V' + fc2)<£ = 0 (13)
Accordingly by (11) </> = ~E, (14)
„ im dR ~ im dR n ^
QXKJL i = U = — — , \ •"•**/
k2 dx ' n dy
by which P and Q are expressed in terms of R supposed already known.
278 ON THE PASSAGE OF ELECTRIC WAVES THROUGH TUBES, [226
The boundary condition (7) is satisfied by the value ascribed to R, and
the same value suffices also to secure the fulfilment of (8), inasmuch as
• dx dyJmdR
ds+^ds fc2 ds
The functions P, Q, R being now known, we may express a, b, c. From (4)
da dR in? + k* dR
-rr = ipa = imQ — 7- = -- rr — ~r~ 5
dt r dy k? dy
so that a=
^-7— -j— , . ., ,
ipkz dy ipk* dx
In vibrations of the second class R = 0 throughout, so that (2) and (7) are
satisfied, while k2 is still at disposal. In this case
P=d+/dy, Q = -d^fdx, ..................... (17)
and (V2 + £2)x/r=0 ............................... (18)
By the third of equations (4)
dc . dP dQ
so that T/T = — ipc/k2, and
ipdc ipdc
f — — yr -j— , V^J — yr -j— , Xt — U
A;2 dy k* dx
im dc j im dc /OAX
AlBoby(4) « = ^^, b = ¥dy ................ ' ........... (20)
Thus all the functions are expressed by means of c, which itself satisfies
(V2+fc2)c = 0 .................................. (21)
We have still to consider the second boundary condition (8). This takes the
form
dc dx dc dy _
dy ds dx ds
requiring that dcfdn, the variation of c along the normal to the boundary at
any point, shall vanish. By (21) and the boundary condition
dc/dn = 0, ................................. (22)
the form of c is determined, as well as the admissible values of k2. The
problem as regards c is thus the same as for the two-dimensional vibrations of
gas within a cylinder which is bounded by rigid walls coincident with the
conductor, or for the vibrations of a liquid under gravity in a vessel of the
same form*.
All the values of k determined by (2) and (7), or by (21) and (22), are real,
* Phil. Mag. Vol. i. p. 272 (1876). [Vol. i. p. 265.]
1897] OR THE VIBRATIONS OF DIELECTRIC CYLINDERS. 279
but the reality of k still leaves it open whether m in (3) shall be real or
imaginary. If we are dealing with free stationary vibrations m is given
and real, from which it follows that p is also real. But if it be p that is
given, ra2 may be either positive or negative. In the former case the motion
is really periodic with respect to z ; but in the latter z enters in the forms
em'z, e~m'z, and the motion becomes infinite when z = + oo , or when z = — oo ,
or in both cases. If the smallest of the possible values of k2 exceeds pzj V2,
m is necessarily imaginary, that is to say no periodic waves of the frequency
in question can be propagated along the cylinder.
Rectangular Section.
The simplest case to which these formulae can be applied is when the
section of the cylinder is rectangular, bounded, we may suppose, by the lines
As for the vibrations of stretched membranes*, the appropriate value
of R applicable to solutions of the first class is
R=ei (mz+^ sin (/«ra?/a) sin (vn-y/fS) ; ............. (23)
from which the remaining functions are deduced so easily by (15), (16) that
it is hardly necessary to write down the expressions. In (23) & and v are
integers, and by (13)
whence m2 = p*/V* -TT* (^ + ^j (25)
The lowest frequency which allows of the propagation of periodic waves along
the cylinder is given by
7T 7T /cu~>\
If the actual frequency of a vibration having its origin at any part of the
cylinder be much less than the above, the resulting disturbance is practically
limited to a neighbouring finite length of the cylinder.
For vibrations of the second class we have
c=ef(wlz+^cos(^7rA-/a)cos(^7ry//3), (27)
the remaining functions being at once deducible by means of (19), (20).
The satisfaction of (22) requires that here again //., v be integers, and (21)
gives
*--<&+»• (28)
identical with (24).
* Theory of Sound, § 195.
280 ON THE PASSAGE OF ELECTRIC WAVES THROUGH TUBES, [226
If a > /3, the smallest value of k corresponds to p = 1, v = 0. When v = 0,
we have k = pir /a, and if the factor eiimz+pt) be omitted,
a = -^smfc#, 6 = 0, c = cos^, ,...(29)
k
a solution independent of the value of ft. There is no solution derivable from
Circular Section.
For the vibrations of the first class we have as the solution of (2) by means
of Bessel's functions,
R = Jn(kr)cosn0, (31)
n being an integer, and the factor ei(mi!+pt} being dropped for the sake of
brevity. In (31) an arbitrary multiplier and an arbitrary addition to 0 are
of course admissible. The value of k is limited to be one of those for which
Jn(kr') = 0 (32)
at the boundary where r = r.
The expressions for P, Q, a, b, c in (15), (16) involve only dR/dx, dR/dy.
For these we have
7 Tt J f> J f>
^ = ^F cos 0 - ^ sin 0 = kJn (kr) cos n0 cos 0 + -Jn (kr) sin n0 sin 0
da; dr ra0 r
( J }
= &k cos (n — 1) 0 \Jn + i—
I **•)
(kr), (33)
according to known properties of these functions ; and in like manner
dR dR . dR
-j- = -j— sm 0 + — ra cos 0
dy dr rd0
(34)
These forms show directly that dR/d.r, dR/dy satisfy the fundamental
equation (2). They apply when n is equal to unity or any greater integer.
When n = 0, we have
R = Jn(kr\ .................................... (35)
(36)
* For (18) would then become v2f = °; and this. with the boundary condition df/dn^O,
•would require that P and Q, as well as R, vanish throughout.
1897] OR THE VIBRATIONS OF DIELECTRIC CYLINDERS. 281
The expressions for the electromotive intensity are somewhat simpler
when the resolution is circumferential and radial :
circumf. component
= QCos0-Ps™eJ^^=-™^Jn(kr)Snn0, .......... (37)
radial component
= Pcos0 + Qsm0 = 'l^~ = ~Jn'(kr)cosn0 ................ (38)
rC CLT rC
If n = 0, the circumferential component vanishes.
Also for the magnetization
circ. comp. of magnetization
; /, • /) f/, .
= bcos0-asm0 = — r-= -- 5- = — r-r— J»(kr) cos nd, ........ (39)
ipk2 dr ipk
rad. comp. of mag. - a cos 6 + b sin 6
m*+k2 dR n(m> + k*) r ,. . . .,
= -- r-y- -- ^ = \ >0 — * Jn(kr) smn0 ..... ... .(40)
ipk2 rd0 ipk*r
The smallest value of k for vibrations of this class belongs to the series
n = 0, and is such that kr = 2'404, r being the radius of the cylinder.
For the vibrations of the second class R = 0, and by (21),
c = Jn(kr)cosnO, ....... . .................... (41)
k being subject to the boundary condition
Jn'(kr') = 0 .................................. (42)
As in (33), (34),
dc dc a dc . a
-y- = -y- cos 6 -- 77; sin 6
dx dr rdd
dc dc . „ dc
-j- = -j- sin 6 + —j-a cos 9
dy dr rd6
(43)
(44)
so that by (19), (20) all the functions are readily expressed.
When n = 0, we have
^ = - kJ, (kr) cos 6, (jf-=-kJ1(kr)sin0 ............... (45)
doc dy
For the circumferential and radial components of magnetization we get
282 ON THE PASSAGE OF ELECTRIC WAVES THROUGH TUBES, ETC. [226
(46)
circ. comp. of mag. = b cos 6 - a sin 6
im dc imn
rad. comp. of mag. = a cos 6 + b sin 6
imdc im T ,
= ~ = Jn
corresponding to (37), (38) for vibrations of the first class.
In like manner equations analogous to (39), (40) now give the components
of electromotive intensity. Thus
circ. comp. = Qcos 0-P sin d = ^ ^ =|? Jn' (kr) cosn0, ............ (48)
rad. comp. = Pcos B + Q sin 0 = -|? ^ = ^ Jn(kr} sin n0 ....... (49)
The smallest value of k admissible for vibrations of the second class is of
the series belonging to n = 1, and is such that kr' = 1-841, a smaller value than
is admissible for any vibration of the first class. Accordingly no real wave of
any kind can be propagated along the cylinder for which p/V is less than
1'841/r', where r denotes the radius. The transition case is the two-
dimensional vibration for which
c =*«*,/, (1-841 r//> cos 0, p= 1-841 V/r' ............. (50,51)
227.
ON THE PASSAGE OF WAVES THROUGH APERTURES IN
PLANE SCREENS, AND ALLIED PROBLEMS.
[Philosophical Magazine, XLIII. pp. 259—272, 1897.]
THE waves contemplated may be either aerial waves of condensation and
rarefaction, or electrical waves propagated in a dielectric. Plane waves of
simple type impinge upon a parallel screen. The screen is supposed to be
infinitely thin, and to be perforated by some kind of aperture. Ultimately
one or both dimensions of the aperture will be regarded as infinitely small
in comparison with the wave-length (X); and the method of investigation
consists in adapting to the present purpose known solutions regarding the
flow of incompressible fluids.
If <£ be a velocity-potential satisfying
d^jdt^V^^, .............................. (1)
where V2 = d*/dtf + d*/dy* + d2/dz2,
the condition at the boundary may be (i) that d<f>fdn= 0, or (ii) that <£ = 0.
The first applies directly to aerial vibrations impinging upon a fixed wall, and
in this connexion has already been considered*.
If we assume that the vibration is everywhere proportional to eint, (1)
becomes
(V2 + fc2) <£ = 0, ................................ (2)
where k = n/V=2Tr/\ ............................... (3)
It will conduce to brevity if we suppress the factor eint. On this un-
derstanding the equation of waves travelling parallel to x in the positive
direction, and accordingly incident upon the negative side of a screen
situated at x = 0, is
Theory of Sound, § 292.
284 ON THE PASSAGE OF WAVES THROUGH APERTURES [227
When the solution is complete, the factor eint is to be restored, and the
imaginary part of the solution is to be rejected. The realized expression
for the incident waves will therefore be
kx) ............................... (5)
Perforated Screen. — Boundary Condition d<f>/dn = 0.
If the screen be complete, the reflected waves under the above condition
have the expression <f> = eikx.
Let us divide the actual solution into two parts % and -fy, the first the
solution which would obtain were the screen complete, the second the
alteration required to take account of the aperture ; and let us distinguish
by the suffixes m and p the values applicable upon the negative (minus} and
upon the positive side of the screen. In the present case we have
Xp = 0 ......................... (6)
This ^-solution makes d^m/dn — 0, d%p/dn = 0 over the whole plane
x = 0, and over the same plane %m = 2, %p = 0.
For the supplementary solution, distinguished in like manner upon the
two sides, we have
(7)
where r denotes the distance of the point at which ty is to be estimated
from the element dS of the aperture, and the integration is extended
over the whole of the area of aperture. Whatever functions of position
^m, "Vp may be, these values on the two sides satisfy (2), and (as is evident
from symmetry) they make d-fymjdn, dtypjdn vanish over the wall, viz. the
unperforated part of the screen ; so that the required condition over the wall
for the complete solution (^ + \Jr) is already satisfied. It remains to consider
the further conditions that <f> and d<f>/dx shall be continuous across the
aperture.
These conditions require that on the aperture
2 + ^m = ^p, d^m/dx = d^p/dx ................... (8)*
The second is satisfied if % = - Vm ; so that
<*S, *, = -¥„ 6-~ dS, ............. (9)
making the values of ^m and -fy-p equal and opposite at all corresponding
points, viz. points which are images of one another in the plane x = 0. In
* The use of dx implies that the variation is in a fixed direction, while dn may be supposed
to be drawn outwards from the screen in both cases.
1897] IN PLANE SCREENS, AND ALLIED PROBLEMS. 285
order further to satisfy the first condition it suffices that over the area of
aperture
^m = -l, ^=1 ............................. (10)
and the remainder of the problem consists in so determining ^m that this
shall be the case.
In this part of the problem we limit ourselves to the supposition that
all the dimensions of the aperture are small in comparison with X. For
points at a distance from the aperture e~ikr/r may then be removed from
under the sign of integration, so that (9) becomes
The significance of JJ ^mdS is readily understood from an electrical inter-
pretation. For in its application to a point, itself situated upon the area
of aperture, e~ikr in (9) may be identified with unity, so that tym is the
potential of a distribution of density *Pm on S. But by (10) this potential
must have the constant value — 1 ; so that —ff^mdS, or ff^pdS, represents
the electrical capacity of a conducting disk having the size and shape of
the aperture, and situated at a distance from all other electrified bodies.
If we denote this by M, the solution applicable to points at a distance from
the aperture may be written
To these are to be added the values of ^ in (6). The realized solutions
are accordingly
MC^~— }, ............... (13)
(14)
The value of M may be expressed* for an ellipse of semi-major axis a
and eccentricity e. We have
M-
J(.y
F being the symbol of the complete elliptic function of the first kind.
When e = 0, F (e) = £TT ; so that for a circle M = 2a/7r.
It should be remarked that M^ in (9) is closely connected with the normal
velocity at dS. In general,
* Theory of Sound, §§ 292, 306, where is given a discussion of the effect of ellipticity when
area is given.
286 ON THE PASSAGE OF WAVES THROUGH APERTURES [227
At a point (x) infinitely close to the surface, only the neighbouring elements
contribute to the integral, and the factor e~ikr may be omitted. Thus
^ =
dx
¥ = -, .............................. (17) -
2-rr dn
d-fy/dn being the normal velocity at the point of the surface in question.
Boundary Condition <f> = 0.
We will now suppose that the condition to be satisfied on the walls is
<£ = 0, although this case has no simple application to aerial vibrations.
Using a similar notation to that previously employed, we have as the ex-
pression for the principal solution
Xm = e-**-e**, Xp=0, ........................ (18)
giving over the whole plane (x = 0), %m = 0, %p = 0, d^m \ dx = — 2ik,
/dx = 0.
The supplementary solutions now take the form
These give on the walls -\Jrm = -^rp = 0, and so do not disturb the condition of
evanescence already satisfied by %. It remains to satisfy over the aperture
^m = ^p> -2ik + d^m/dx = d^p/dx ................ (20)
The first of these is satisfied if Wm = — 'Wp, so that ^rm and typ are equal
at any pair of corresponding points upon the two sides. The values of
d^fm\dx, dfa/dx are then opposite, and the remaining condition is also
satisfied if
d^Jdx = ik, d^p/dx = -ik .................... (21)
Thus Wm is to be such as to make d-^rmjdx = ik ; and, as in the proof of (17),
it is easy to show that in (19)
¥ro = ^m/27T, % = -^/27T, ................... (22)
where T/rm, -fyp are the (equal) surface- values at dS.
When all the dimensions of 8 are small in comparison with the wave-
length, (19) in its application to points at a sufficient distance from S
assumes the form
and it only remains to find what is the value of ftypdS which corresponds
to d-dx = - ik.
1897] IN PLANE SCREENS, AND ALLIED PROBLEMS. 287
Now this correspondence is ultimately the same as if we were dealing
with an absolutely incompressible fluid. If we imagine a rigid and infinitely
thin plate (having the form of the aperture) to move normally through
unlimited fluid with velocity u, the condition is satisfied that over the re-
mainder of the plane the velocity-potential ty vanishes. In this case the
values of ^r at corresponding points upon the two sides are opposite ; but
if we limit our attention to the positive side, the conditions are the same
as in the present problem. The kinetic energy of the motion is proportional
to uz, and we will suppose that twice the energy upon one side is hitf. By
Green's theorem this is equal to —ff^.d-^ldn.dS, or -uffydS; so that
//•^rdS= — hu. In the present application u = — ik, so that the corresponding
value of ffypdS is ihk. Thus (23) becomes
The same algebraic expression gives -\|rm, if the minus sign be omitted; for
as x itself changes sign in passing from one side to the other, the values of
tym and typ at corresponding points are then equal.
The value of h can be determined in certain cases. For a circle* of
radius c
(26)
so that for a circular aperture the realized solution is
^ = -~ ~cos(nt-kr), .............................. (27)
<£m = 2 sinnt smkx + ~ — cos(nt-kr) ............. (28)
oA, T
It will be remarked that while in the first problem the wave (ty) divergent
from the aperture is proportional to the first power of the linear dimension,
in the present case the amplitude is very much less, being proportional to
the cube of that quantity.
The solution for an elliptic aperture is deducible from the general theory
of the motion of an ellipsoid (a, b, c) through incompressible fluid f, by
supposing a = 0, while b and c remain finite and unequal; but the general
expression does not appear to have been worked out. When the eccentricity
of the residual ellipse is small, I find that
A = $(te)l (!-&*). ........................... (29)
showing that the effect of moderate ellipticity is very small when the area
is given.
* Lamb's Hydrodynamics, § 105.
t Loc. ««., § 111.
288 ON THE PASSAGE OF WAVES THROUGH APERTURES [227
From the solutions already obtained it is possible to derive others by
differentiation. If, for example, we take the value of </> in the first problem
and differentiate it with respect to x, we obtain a function which satisfies (2),
which includes plane waves and their reflexion on the negative side, and
which satisfies over the wall the condition of evanescence. It would seem
at first sight as if this could be no other than the solution of the second
problem, but the manner in which the linear dimension of the aperture
enters suffices to show that it is not so. The fact is that although the
proposed function vanishes over the plane part of the wall, it becomes in-
finite at the edge, and thus includes the action of sources there distributed.
A similar remark applies to the solutions that might be obtained by differen-
tiation of the second solution with respect to y or z, the coordinates measured
parallel to the plane of the screen.
ing Plate. — d<f>/dn = 0.
We now pass to the consideration of allied problems in which the trans-
parent and opaque parts of the screen are interchanged. Under the above-
written boundary condition the case is that of plane aerial waves incident
upon a parallel infinitely thin plate, whose dimensions are ultimately sup-
posed to be small in comparison with \. The analytical process of solution
may be illustrated by the following argument. Suppose a motion commu-
nicated to the plate identical with that which the air at that place would
execute were the plate absent. It is evident that the propagation of the
primary wave will then be undisturbed. The supplementary solution, re-
presenting the disturbance due to the plate, must then correspond to the
reduction of the plate to rest, that is to a motion of the plate equal and
opposite to that just imagined. The supplementary solution is accordingly
analogous to that which occurs in the second of the problems already
treated.
Using a similar notation, we have for the principal solution upon the
two sides
Xm = Xp = e-i**> .............................. (30)
giving when x = 0
The supplementary solution is of the form (19), and gives upon the aperture,
viz. the part of the plane x = 0 unoccupied by the plate, -ty-m = ^jrp — 0, and
so does not disturb the continuity of </>. But in order that the continuity
of d(j>/dx may be maintained it is necessary that '<Pp = '(irm; and then the
1897] IN PLANE SCREENS, AND ALLIED PROBLEMS. 289
values of ^rm and typ are opposite at any pair of corresponding points upon
the two sides.
It remains to satisfy the necessary conditions at the plate itself.
These are
,
dx dx dx dx
or, since d-^rm/dx, d-$p/dx are equal,
d+m/dx = d+p/dx = ik. ........................ (31)
It follows that typ has the opposite value to that expressed in (25) ; and the
realized solution for a circular plate of radius c becomes
<j>p = cos (nt — kx) + -—2 -cos(nt — kr), .............. (32)
^cos(nt-kr), (33)
the analytical form being the same in the two cases.
It is important to notice that the reflexion from the plate is utterly
different from the transmission by a corresponding aperture in an opaque
screen, as given in (14), the former varying as the cube of the linear
dimension, and the latter as the first power simply.
Reflecting Plate.— <j> = 0.
For the sake of completeness it may be well to indicate the solution of
a fourth problem defined by the above heading. This has an affinity with
the first problem, analogous to that of the third with the second. The form
of % is the same as in (30), and those for -v/rTO, typ the same as in (7). These
make d^m/dx, d^pjdx vanish on the aperture, and so do not disturb the
continuity of d(f>/dx. But in order that the continuity of <f> may also be
maintained, we must have ^n = ^p, and not as in (9) Vm = — %. On the
plate itself we must have
Accordingly ^m is the same as in (12), while typ in (12) must have its
sign reversed. The realized solution is
»..,«,(,.- fa) -Jf- (34)
19
290 ON THE PASSAGE OF WAVES THROUGH APERTURES [227
Two-dimensional Vibrations.
In the class of problems before us the velocity-potential of a point-
source, viz. e~{kr/r, is replaced by that of a linear source; and this in
general is much more complicated. If we denote it by D(kr), the ex-
pressions are*
2" ' 22.42
+ ^-^^ + ^-^£3- , (35)
where 7 is Euler's constant ('5772...), and
Of these the first is " semiconvergent," and is applicable when kr is large ;
the second is fully convergent and gives the form of the function when
kr is small.
Since the complete analytical theory is rather complicated, it may be
convenient to give a comparatively simple derivation of the extreme forms,
which includes all that is required for our present purpose, starting from
the conception of a linear source as composed of distributed point-sources.
If p be the distance of any element dx of the linear source from 0, the
point at which the potential is to be estimated, and r be the smallest
value of p, so that p2 = r2 + x2, we may take as the potential, constant
factors being omitted,
(36)
We have now to trace the form of (36) when kr is very great, and also
when kr is very small. For the former case we replace p by r + y, thus
obtaining
When kr is very great, the approximate value of the integral in (37) may be
obtained by neglecting the variation of V(2r + y), since on account of the
rapid fluctuation of sign caused by the factor er^ we need attend only to
small values of y. Now, as is known,
rcos x dx _ r°° sin x dx _ //TT\
V* ~L V« "Y W§
* See for example Theory of Sound, § 341.
1897] IN PLANE SCREENS, AND ALLIED PROBLEMS. 291
so that in the limit
in agreement with (35).
We have next to deduce the limiting form of (36) when kr is very small.
For this purpose we may write it in the form
The first integral in (39) is well known. We have
= Ci (kr) - i {fr 4- Si (kr)}
-_
IT
i--
In the second integral of (39) the function to be integrated vanishes
when p is great compared to r, and when p is not great in comparison
with r, kp is small and e~ikf may be identified with unity. Thus in the
limit
and (39) becomes
^r = ry 4. bg kr + \ITT - log 2 = 7 + log (^ikr), (40)
in agreement with (35).
When kr is extremely small (40) may be considered for some purposes
to reduce to log kr ; but the term |tV is required in order to represent the
equality of work done in the neighbourhood of the linear source and at
a great distance from it.
We may now proceed to solve four problems relative to narrow slits
and reflecting blades analogous to the four already considered in which the
aperture or the reflecting plate was small in both its dimensions in com-
parison with the wave-length.
Narrow Slit. — Boundary Condition d(f)/dn = Q.
As in the former problem the principal solution is
Xp = Q, (41)
making dxm/dn, dxp/dn vanish over the whole plane # = 0 and over the
19—2
292 ON THE PASSAGE OF WAVES THROUGH APERTURES [227
same plane %m = 2, %p = 0. The supplementary solution, which represents
the effect of the slit, may be written
(42)
Vm, Vp being certain functions of y to be determined, and the integration
extending over the width of the slit from y = -b to y = + b.
These additions do not disturb the condition to be satisfied over the
wall. On the aperture continuity requires, as in (8), that
2 + ym = ^f, d^m/dx = d^p/dx.
The second of these is satisfied by taking *Pp = — '*Pmi so that at all corre-
sponding pairs of points ^m = — ^p. It remains to determine "9m so that on
the aperture ^frm = — 1 ; and then by what has been said -^rv = + 1.
At a sufficient distance from the slit, supposed to be very narrow, D (kr)
may be removed from under the integral sign and also be replaced by its
limiting form given in (35). Thus
(43>
The condition by which "Vm is determined is that for all points upon the
aperture
(44)
where, since kr is small throughout, the second limiting form given in
(35) may be introduced.
From the known solution for the flow of incompressible fluid through
a slit in an infinite plane we may infer that *¥m will be of the form
A (b2 — yz)~*, where A is some constant. Thus (44) becomes
In this equation the first integral is obviously independent of the position
of the point chosen, and if the form of Wm has been rightly taken the
second integral must also be independent of it. If its coordinate be 77,
lying between + 6,
and must be independent of 77. This can be verified without much difficulty
1897] IN PLANE SCREENS, AND ALLIED PROBLEMS. 293
by assuming if = b sin a, y = bsin0; but merely to determine A in (45) it
suffices to consider the particular case of tj = 0. Here
so that (43) becomes ? +
From this typ is derived by simply prefixing a negative sign.
The realized solution is obtained from (46) by omitting the imaginary
part after introduction of the suppressed factor eint. If the imaginary part
of Iog(£t%6) be neglected, the result is
corresponding to %m = 2 cos n£ cos kx ............................ (48)
The solution (47) applies directly to aerial vibrations incident upon a per-
forated wall, and to an electrical problem which will be specified later.
Perhaps the most remarkable feature of it is the very limited dependence
of the transmitted vibration on the width (26) of the aperture.
Narrow Slit. — Boundary Condition <f> = Q.
The principal solution is the same as in (18) ; and the conditions for the
supplementary solution, to be satisfied over the aperture, are those expressed
in (21). In place of (19)
the values of ¥m and % being opposite, and those of -»/rm and ^ equal at
corresponding points. At a distance we have
*-£/>.* .......................... (»)
i • i dD ikx / TT \* .^ /KI\
in which -J-=—(^r) e ........................... (51)
dx r \2ikrJ
There is a simple relation between the value of Vp at any point of the
aperture and that of ^p at the same point. For in the application of (49)
294 ON THE PASSAGE OF WAVES THROUGH APERTURES [227
to any point of the narrow aperture, dDldx = x/r2, showing that only those
elements of the integral are sensible which lie infinitely near the point
where ^rp is to be estimated. The evaluation is effected by considering
in the first instance a point for which x is finite, and afterwards passing
to the limit. Thus
so that (50) becomes fa = jg:if .......................... (52)
It remains only to express the connexion between f^pdy and the constant
value of d^-p/dx on the area of the aperture ; and this is effected by the
known solution for an incompressible fluid moving under similar conditions.
The argument is the same as in the corresponding problem where the
perforation is circular. In the motion (a) of a lamina of width (26) through
infinite fluid, the whole kinetic energy per unit of length may be denoted
by hu2, and it appears from Green's theorem that ftypdy = ihk. The value
of h* is ^7r&2; so that
(53)
The same algebraical expression gives tym, if the minus sign be omitted.
The realized solution from (53) is
)* «•<*-*- W ............... (54)
corresponding to >%m = 2 sin nt sin kx ............................ (55)
Reflecting Blade. — Boundary Condition d<f>/dn = 0.
We have now to consider two problems which differ from the last in
that the opaque and transparent parts of the screen are interchanged. As
in the case of the circular aperture, we shall find that the correspondence
lies between the reflecting blade under the condition d(f>/dn = Q and the
transmitting aperture under the condition </> = 0, and reciprocally.
The principal solution remains as in (30). The supplementary solution
must satisfy (31), where
and *PP must be equal in order that the continuity of d(j>/dx over
* Lamb's Hydrodynamics, § 71.
1897] IN PLANE SCREENS, AND ALLIED PROBLEMS. 295
the aperture may be maintained. Thus i/rm and ^rp have opposite values
at any pair of corresponding points.
If we compare these conditions with those by which (53) was determined,
we see that i/rm has the same value as in that case, but that the sign of ^p
must be reversed. Thus in the present problem
corresponding to %m = %p = cos (nt — kx) .......................... (58)
Reflecting Blade. — Boundary Condition <£ = 0.
In this case % still remains as in (30). The general forms for tym, -^p
are as in (42), which secure that d^m/dx, d-fy-pjdx shall vanish on the
aperture (i.e. the part of the plane # = 0 unoccupied by the blade). But
in order that the continuity of </> may also be maintained over that area
we must have ^fm = ^rp. Thus •^rm, -*frp have equal values at corresponding
points. On the blade itself •\lrm = ^p = — 1.
A comparison of these conditions with those by which (46) was deter-
mined shows that in the present case
When log i in the denominator of (59) may be omitted, the realized form is
that expressed by (47), and this corresponds to
X™ = XP - cos (nt ~ &*) .......................... (60)
Various Applications.
Of the eight problems, whose solutions have now been given, four have
an immediate application to aerial vibrations, viz. those in which the con-
dition on the walls is d$/dn = 0. The symbol <£ then denotes the velocity-
potential, and the condition expresses simply that the fluid does not penetrate
the boundary. The ' four problems relating to two dimensions have also
a direct application to electrical vibrations, if we suppose that the thin
material constituting the screen (or the blade) is a perfect conductor. For
if R denote the electromotive intensity parallel to z, the condition at the
face of the conductor is £ = 0; so that if R be written for ^ in (53), (59),
we have the solutions for a narrow aperture in an infinite screen, and for
a narrow reflecting blade respectively, corresponding to the incident wave
296 ON THE PASSAGE OF WAVES THROUGH APERTURES, ETC. [227
R = e-*x. A narrow aperture parallel to the electric vibrations transmits
very much less than is reflected by a conductor elongated in the same
direction.
The two other solutions relative to two dimensions find electrical appli-
cation if we identify <f> with c, the component of magnetic intensity parallel
to z. For when the other components a and b are zero, the condition to
be satisfied at the face of a conductor is dc/dn = 0. Thus (46), (57) apply
to incident vibrations represented by c = e~ikx. In this case the slit transmits
much more than the blade reflects.
It may be remarked that in general problems of electrical vibration in
two dimensions have simple acoustical analogues*. As an example we may
refer to the reflexion of plane electric waves incident perpendicularly upon
a corrugated surface, the acoustical analogue of which is treated in Theory
of Sound, 2nd ed. § 272 a, and to the reflexion of electric waves from a con-
ducting cylinder (§ 343).
* The comparison is not limited to the case of perfect conductors, but applies also when the
obstacles, being non-conductors, differ from the surrounding medium in specific inductive capacity,
or in magnetic permeability, or in both properties.
228.
THE LIMITS OF AUDITION.
[Royal Institution Proceedings, xv. pp. 417—418, 1897.]
IN order to be audible, sounds must be restricted to a certain range of
pitch. Thus a sound from a hydrogen flame vibrating in a large resonator
was inaudible, as being too low in pitch. On the other side, a bird-call,
giving about 20,000 vibrations per second, was inaudible, although a sensitive
flame readily gave evidence of the vibrations and permitted the wave-length
to be measured. Near the limit of hearing the ear is very rapidly fatigued ;
a sound in the first instance loud enough to be disagreeable, disappearing
after a few seconds. A momentary intermission, due, for example, to a rapid
passage of the hand past the ear, again allows the sound to be heard.
The magnitude of vibration necessary for audition at a favourable pitch
is an important subject for investigation. The earliest estimate is that of
Boltzmann. An easy road to a superior limit is to find the amount of energy
required to blow a whistle and the distance to which the sound can be heard
(e.g. one-half a mile). Experiments upon this plan gave for the amplitude
8 x 10~8 cm., a distance which would need to be multiplied 100 times in order
to make it visible in any possible microscope. Better results may be obtained
by using a vibrating fork as a source of sound. The energy resident in the
fork at any time may be deduced from the amplitude as observed under
a microscope. From this the rate at which energy is emitted follows when
we know the rate at which the vibrations of the fork die down (say to one-
half). In this way the distance of audibility may be reduced to 30 metres,
and the results are less liable to be disturbed by atmospheric irregularities.
If s be the proportional condensation in the waves which are just capable of
exciting audition, the results may be expressed : —
frequency = 256
:512
= 6'OxlO-9
= 4-6x 10-9
4-6x 10-9
298 THE LIMITS OF AUDITION. [228
showing that the ear is capable of recognising vibrations which involve far
less changes of pressure than the total pressure outstanding in our highest
In such experiments the whole energy emitted is very small, and contrasts
strangely with the 60 horse-power thrown into the fog-signals of the Trinity
House. If we calculate according to the law of inverse squares how far
a sound absorbing 60 horse-power should be audible, the answer is 2700 kilo-
metres ! The conclusion plainly follows that there is some important source
of loss beyond the mere diffusion over a larger surface. Many years ago
Sir George Stokes calculated the effect of radiation upon the propagation
of sound. His conclusion may be thus stated. The amplitude of sound
propagated in plane waves would fall to half its value in six times the interval
of time occupied by a mass of air heated above its surroundings in cooling
through half the excess of temperature. There appear to be no data by
which the latter interval can be fixed with any approach to precision ; but if
we take it at one minute, the conclusion is that sound would be propagated
for six minutes, or travel over about seventy miles, without very serious loss
from this cause.
The real reason for the falling off at great distances is doubtless to be
found principally in atmospheric refraction due to variation of temperature,
and of wind, with height. In a normal state of things the air is cooler over-
head, sound is propagated more slowly, and a wave is tilted up so as to
pass over the head of an observer at a distance. [Illustrated by a model.]
The theory of these effects has been given by Stokes and Reynolds, and their
application to the explanation of the vagaries of fog-signals by Henry.
Progress would be promoted by a better knowledge of what is passing in
the atmosphere over our heads.
The lecture concluded with an account of the observations of Preyer upon
the delicacy of pitch perception, and of the results of Kohlrausch upon the
estimation of pitch when the total number of vibrations is small. In illustra-
tion of the latter subject an experiment (after Lodge) was shown, in which
the sound was due to the oscillating discharge of a Leyden battery through
coils of insulated wire. Observation of the spark proved that the total
number of (aerial) vibrations was four or five. The effect upon the pitch
of moving one of the coils so as to vary the self-induction was very apparent.
229.
ON THE MEASUREMENT OF ALTERNATE CURRENTS BY
MEANS OF AN OBLIQUELY SITUATED GALVANOMETER
NEEDLE, WITH A METHOD OF DETERMINING THE ANGLE
OF LAG.
[Philosophical Magazine, XLIII. pp. 343 — 349, 1897.]
IT is many years* since, as the result of some experiments upon induction,
I proposed a soft iron needle for use with alternate currents in place of the
permanently magnetized steel needle ordinarily employed in the galvanometer
for the measurement of steady currents. An instrument of this kind designed
for telephonic currents has since been constructed by Giltay ; but, so far as
I am aware, no application has been made of it to measurements upon a large
scale, although the principle of alternately reversed magnetism is the founda-
tion of several successful commercial instruments.
The theory of the behaviour of an elongated needle is sufficiently simple,
so long as it can be assumed that the magnetism is made up of two parts,
one of which is constant and the other proportional to the magnetizing force.
If internal induced currents can be neglected, this assumption may be
regarded as legitimate so long as the forces are small f. In the ordinary case
of alternate currents, where upon the whole there is no transfer of electricity
in either direction, the constant part of the magnetism has no effect ; while
the variable part gives rise to a deflecting couple proportional on the one
hand to the mean value of the square of the magnetizing force or current,
and upon the other to the sine of twice the angle between the direction of
the force and the length of the needle. The deflecting couple is thus
evanescent when the needle stands either parallel or perpendicular to the
magnetizing force, and rises to a maximum at the angle of 45°. For practical
* Brit. Assoc. Report, 1868; Phil. Mag. Vol. in. p. 43 (1887). [Vol. i. p. 310.]
t Phil. Mag. Vol. xxur. p. 225 (1887). [Vol. n. p. 579.]
300
ON THE MEASUREMENT OF ALTERNATE CURRENTS BY MEANS
[229
purposes the law of proportionality to the mean square of current would
seem to be trustworthy so long as no great change occurs in the frequency
or type of current ; otherwise eddy currents in the iron might lead to error,
unless the metal were finely subdivided.
It is hardly to be supposed that for ordinary purposes a suspended
iron needle would compete in convenience with the excellent instruments
now generally available ; but having found it suitable for a special purpose
of my own, I think it may be worth while to draw to it the attention of those
interested. In experiments upon the oxidation of nitrogen by the electric
arc or flame it was desired to ascertain the relation between the electric
power absorbed and the amount of nitrogen oxidized. A transformer with
an unclosed magnetic circuit was employed to raise the potential from that
of the supply to the 3000 volts or more needed at the platinum terminals.
Commercial ampere-meters and volt-meters gave with all needed precision
the current and potential at the primary of the transformer ; but, as is well
known, these data do not suffice for an estimate of power. The latter depends
also upon the angle of lag, or retardation of current relatively to potential-
difference. If this angle be 0, the power actually employed is to be found
by multiplying the product of volts and amperes by cos 0, so that the actual
power may be less to any extent than the apparent power represented by
the simple product. Various watt-meters have been introduced for measuring
the actual power directly, but I could not hear of one suitable for the large
current of 40 amperes used at the Royal Institution. Working subsequently
in the country I returned to the problem, and succeeded in determining the
angle of lag very easily by means of the principle now to be explained.
The soft iron needle of 2 centim. in length, suspended by a fine torsion-
fibre of glass and carrying a mirror in the usual way, is inclined at 45° to
the direction of the magnetic force. This force is due to currents in two coils,
the common axis of the coils being horizontal and passing through the centre
of the needle. As in ordinary galvanometers, the mean plane of each coil
may include the centre of the needle ; but it was found better to dispose the
coils on opposite sides and at distances from the needle which could be varied.
A plan of the arrangement is sketched diagrammatically in the woodcut,
1897] OF AN OBLIQUELY SITUATED GALVANOMETER NEEDLE. 301
where MM, SS represent the two coils, the common axis HK passing through
the centre of the needle 37. If the currents in the coils are of the same
frequency and of simple type, the magnetizing forces along HK may be
denoted by A cos nt, B cos (nt — e), e being the phase-difference. If either
force act alone, the deflecting couple is represented by .A2 or by B2 ; but if
the two forces cooperate the corresponding effect is
(72 = A2 + B2 + 2AB cos e, (1)
reducing itself to (A + J5)2 or (A — B}z only in the cases where e is zero or two
right angles. The method consists in measuring upon any common scale all
the three quantities A*, B2, and (72, from which e can be deduced by trigono-
metrical tables, or more simply in many cases by constructing the triangle
whose sides are A, B, and C. The determination of the phase-difference
between the currents is thus independent of any measurement of their
absolute values.
The best method of estimating the deflecting couples may depend upon
the circumstances of the particular case. The most accurate in principle is
the restoration of the needle to the zero position by means of a torsion-head.
But when the conditions are so arranged that the angular deflexions are
moderate, it will usually suffice merely to read them, either objectively by
a spot of light thrown upon a scale, or by means of a telescope. In any case
where it may be desired to push the deflexions beyond the region where the
law of proportionality can be relied upon, all risk of error may be avoided by
comparison with another instrument of trustworthy calibration, one coil only
of the soft iron apparatus being employed.
In certain cases the advantages which accompany the restoration of the
zero position of the needle may be secured by causing the deflexions them-
selves to assume a constant value, e.g. by making known changes of resistance
in one or both of the circuits, or by motion of the coils altering their
efficiencies in a known ratio.
In the particular experiments for which the apparatus was set up the
coil MM (see woodcut) was reduced to a single turn of about 17 centim.
diameter and conveyed the main current (about 10 amperes) which traversed
the primary circuit of the transformer. This, it may be mentioned, was
a home-made instrument, somewhat of the Ruhmkorff type, and was placed
at a sufficient distance from the measuring apparatus. The shunt-coil SS
was of somewhat less diameter, and contained 32 convolutions. The shunt-
circuit included also two electric lamps, joined in series, and its terminals
were connected with two points of the main circuit outside the apparatus,
where the difference of potentials was about 40 volts. Provision was made
for diverting the main current at pleasure from MM, and by means of a re-
verser the direction of the current in SS could be altered, equivalent to
302 ON THE MEASUREMENT OF ALTERNATE CURRENTS BY MEANS [229
a change of e by 180°. The measurements to be made are the effects of MM
and of 88 acting separately, and of MM and 88 acting together in one or
both positions of the reverser.
The best arrangement of the details of observation will depend somewhat
upon the particular value of e to be dealt with. If this be 60°, or there-
abouts, the method can be applied with peculiar advantage. For by pre-
liminary adjustment of the coils, if movable, or by inclusion of (unknown)
resistance in the shunt-circuit, the deflexions due to MM and SS may be
made equal to one another ; so that in the case supposed the same deflexion
will ensue from the simultaneous action of the two currents in one of the
ways in which they may be combined.
This condition of things was somewhat approached in the actual measures
relating to the electric flame. Thus in one trial the coils were adjusted so
as to make the deflexions, due to each of the currents acting singly, equal
to one another. The value was 40 divisions of the scale. When both currents
were turned on, the deflexion was 26 J divisions. Thus
whence cose = '67, or e = 48°.
In a second experiment the deflexion due to both currents acting together
was made equal to that of the main acting alone. Here
whence cos e = '665.
The accuracy was limited by the unsteadiness of the electric flame and of the
primary currents (from a gas-driven De Meritens) rather than by want of
delicacy in the measuring apparatus.
When the phase-difference is about a quarter of a period, cose is small,
and its value is best found by observing the effect of reversing the shunt-
current while the main current continues running. The difference is 4>AB cos e,
from which, combined with a knowledge of A and B, the value of cos e is ad-
vantageously derived. If cos e is absolutely zero, the reversal does not alter
the reading.
If the currents are in the same, or in opposite phases, it is possible to
reduce the joint effect to zero by suitable adjustment of the coils or of the
shunt resistance.
The application of principal interest is when the shunt-current may be
assumed to have the same phase as the potential-difference at its terminals,
for then cos e is the factor by which the true watts may be derived from the
apparent watts. We will presently consider the question of the negligibility
1897] OF AN OBLIQUELY SITUATED GALVANOMETER NEEDLE. 303
of the self-induction of the shunt-current, but before proceeding to this it
may be well to show the application of the formulae when the currents deviate
from the sine type.
If a be the instantaneous current, and v the instantaneous potential-
difference at the terminals, the work done is fav dt. The readings of the soft
iron galvanometer for either current alone may be represented by
A2 = k2fa2dt, B'* = k*fv*dt, ............. ........ (2)
where h, k are constants depending upon the disposition of the apparatus.
When both currents act, we have the readings
Cj2 or C2z = j(ha±kvJ2dt .......... .................. (3)
Taking the first alternative, we find
C? = h?ja?dt + Zhkfavdt + k*Jv*dt,
The fraction on the right of (4) is the ratio of true and apparent watts ; and
we see that, whether the currents follow the sine law or not, the ratio is given
by cos e, where, as before, e is the angle of the triangle constructed with sides
proportional to the square roots of the three readings.
Another formula for cos e is
In the final formula (4) the factors of efficiency of the separate coils (h, k)
do not enter. This result depends, however, upon the fulfilment of the con-
dition of parallelism between the two coils. If the magnetic forces due to
the coils be inclined at different angles %, %' to the length of the needle, we
have in place of (3),
C2 = / (a cos x + v cos %') (a sin % + v sin %') dt
=/ [| a2 sin 2% + $v2 sin 2^' + av sin (^ + x')] dt ; ...... (6)
while ^2 = £sin2x/a2d£, Bt=±sm2tffiPdt ................ (7)
Accordingly
/ av dt _ C*-A*-B* V {sin 2# . sin 2^'}
{Jra^dtx~fv^df\^~ 2AB sin(x + x') ' ......... ( '
in which the second fraction on the right represents the influence of the
defect in parallelism. If ^ and %' are both nearly equal to 45°, then approxi-
mately
V {sin 2 X. sin 2^} _
' *(* %> ...................... (
304 ON THE MEASUREMENT OF ALTERNATE CURRENTS, ETC. [229
We have now to consider under what conditions the shunt-current may
be assumed to be proportional to the instantaneous value of the potential-
difference at its terminals. The obstacles are principally the self-induction of
the shunt-coil itself, and the mutual induction between it and the coil which
conveys the main current. As to the former, we know* that if the mean
radius of a coil be a, and if the section be circular of radius c, and if n be the
number of convolutions,
"*T-i} <10>
To take an example from the shunt-coil used in the experiments above
referred to, where
a = 6 cm., c = 1 cm., n = 32,
L is of the order 105cm. The time-constant of the shunt-circuit (T) is equal
to L/R, where R is the resistance in C.G.S. measure. If r be the resistance
measured in ohms, R = r x 109, so that
1
~ r x 104 '
What we are concerned with is the ratio of T to the period of the currents ;
if the latter be y^ second, the ratio is l/100r, so that if r be a good number
of ohms — it must have exceeded 100 in the particular experiments — there is
nothing to fear from self-induction. It would seem to follow generally that
if the voltage be not too small, say not falling below 10 volts, there should be
no difficulty in obtaining sufficient effect from a shunt-coil whose self-induction
may be neglected. It may be remarked that since the efficiency of the coil
varies as n, while L varies as n2, it will be advantageous to keep n (and r)
down so long as the self-induction of the whole shunt-circuit is mainly that
of the coil.
If the main and the shunt-coils were wound closely together, the disturb-
ance due to mutual induction would be of the same order of magnitude as
that due to self-induction. If the coils are separated, as is otherwise con-
venient, the influence of mutual induction will be less, and may be neglected
under the conditions above defined.
As to the effect of self-induction, if present, we know that the lag (<£) is
given by
tan </> = Lp I R, (11)
where p = ZTT x frequency. The angle of lag of the main current (6), which
it is the object of the measurements to determine, is then given by
6 = e + <f>, (12)
e being the phase-difference of the two currents as found directly from the
observations.
* Maxwell's Electricity, § 706.
230.
ON THE INCIDENCE OF AERIAL AND ELECTRIC WAVES
UPON SMALL OBSTACLES IN THE FORM OF ELLIPSOIDS
OR ELLIPTIC CYLINDERS, AND ON THE PASSAGE OF
ELECTRIC WAVES THROUGH A CIRCULAR APERTURE IN
A CONDUCTING SCREEN.
[Philosophical Magazine, XLIV. pp. 28 — 52, 1897.]
THE present paper may be regarded as a development of previous
researches by the author upon allied subjects. When the character of the
obstacle differs only infinitesimally from that of the surrounding medium,
a solution may be obtained independently of the size and the form which
it presents. But when this limitation is disregarded, when, for example,
in the case of aerial vibrations the obstacle is of arbitrary compressibility
and density, or in the case of electric vibrations when the dielectric constant
and the permeability are arbitrary, the solutions hitherto given are confined
to the case of small spheres, or circular cylinders. In the present investiga-
tion extension is made to ellipsoids, including flat circular disks and thin
blades.
The results arrived at are limiting values, strictly applicable only when
the dimensions of the obstacles are infinitesimal, and at distances outwards
which are infinitely great in comparison with the wave-length (X). The
method proceeds by considering in the first instance what occurs in an inter-
mediate region, where the distance (r) is at once great in comparison with
the dimensions of the obstacle and small in comparison with X. Throughout
this region and within it the calculation proceeds as if A, were infinite, and
depends only upon the properties of the common potential. When this
problem is solved, extension is made without much difficulty to the exterior
region where r is great in comparison with X, and where the common
potential no longer avails.
R. iv. 20
306 ON THE INCIDENCE OF AERIAL
At the close of the paper a problem of some importance is considered
relative to the escape of electric waves through small circular apertures
in metallic screens. The case of narrow elongated slits has already been
treated*.
Obstacle in a Uniform Field.
The analytical problem with which we commence is the same whether
the flow be thermal, electric, or magnetic, the obstacle differing from the
surrounding medium in conductivity, specific inductive capacity, or per-
meability respectively. If <f> denote its potential, the uniform field is
defined by
wz; ........................... (1)
u, v, w being the fluxes in the direction of fixed, arbitrarily chosen, rectangular
axes. If i/r be the potential in the uniform medium due to the obstacle, so
that the complete potential is </> -I- ^r, ^ may be expanded in the series of
spherical harmonics
the origin of r being within the obstacle. Since there is no source, S0
vanishes. Further, at a great distance S2)S3,... maybe neglected, so that
ir there reduces to
The disturbance (3) corresponds to (1). If we express separately the
parts corresponding to u, v, lu, writing A' = Alu + A2v + A3w, &c., we have
r3^ = u (AlX + B,y + C.z) + v (A2a; + B2y + C2z) + w (A3x + B3y + C3z) ;
......... (4)
but the nine coefficients are not independent. By the law of reciprocity the
coefficient of the #-part due to v must be the same as that of the y-part due
to u, and so on-f*. Thus B± = A2, &c., and we may write (4) in the form
dF dF dF
^ + = Udx+Vdy+Wdz> ........................ (5>
where F=^A1a? + \E^ + $C3z2 + B^xy + C2yz + G&x .......... (6)
In the case of a body, like an ellipsoid, symmetrical with respect to three
planes chosen as coordinate planes,
B, = C2 = 0, = 0,
* Phil. Mag. Vol. xmi. p. 272. [Vol. iv. p. 295.]
t Theory of Sound, § 109. u and v may be supposed to be due to point-sources situated at a
great distance B along the axes of x and y respectively.
1897] AND ELECTRIC WAVES UPON SMALL OBSTACLES. 307
and (4) reduces to
(7)
It will now be shown that by a suitable choice of coordinates this reduc-
tion may be effected in any case. Let u, v, w originate in a source at distance
R, whose coordinates are x', y', /, so that u = x'jR3, &c. Then (5) becomes
•i rr -I-TI jrr
^^ = X> dx + y' dy + Z> dz = AlXX' + E*yy' + °3ZZ'
+ B, (x'y + y'x} + C2 (y'z + z'y} + C, (z'x + x'z)
= F(x + x', y + y', z + /) - F (x, y,z}-F (x', y', /).
Now by a suitable transformation of coordinates F(x, y, z), and therefore
F (x, y', z') and F (x + x', y + y', z + /), may be reduced to the form
A^ + B2y2 + C,z*, &c.
If this be done,
r^R3^ = A^xx + B2yy + C3zz',
or reverting to u, v, w, reckoned parallel to the new axes,
r3^ = A^ux + B^vy+ C3ivz, ........................ (8)
as in (7) for the ellipsoid. It should be observed that this reduction of the
potential at a distance from the obstacle to the form (8) is independent of
the question whether the material composing the obstacle is uniform.
For the case of the ellipsoid (a, b, c) of uniform quality the solution may
be completely carried out. Thus*, if T be the volume, so that
T=%7rabc, ................................. (9)
we have AlU = -AT, B,v = - BT, G3w^-GT, ............ (10)
(1
where L = ^abcQ ^^--—^ ............ (12)
with similar expressions for M and N.
In (11) K denotes the susceptibility to magnetization. In terms of the
permeability /*, analogous to conductivity in the allied problems, we have, if
/jf relate to the ellipsoid and n to the surrounding medium,
(13)
so that
with similar equations for B and C.
* The magnetic problem is considered in Maxwell's Electricity and Magnetism, 1873, § 437,
and in Mascart's Lecons, 1896, §§ 52, 53, 276.
20—2
308 ON THE INCIDENCE OF AERIAL [230
Two extreme cases are worthy of especial notice. If /////* = oo , the
general equation for ty becomes
r^r ux vy wz
T L+M+N-
On the other hand, if /*' ///. = 0,
r3-^ ux vy wz
In the case of the sphere (a)
L = M=N = $7r;
so that (15) becomes
^=-^(wc + vy + wz), ..................... (18)
giving, when r = a, 0 + ^ = 0. This is the case of the perfect conductor.
In like manner for the non-conducting sphere (16) gives
* = j^(ux + vy + wz) ......................... (19)
If the conductivity of the sphere be finite (//),
which includes (18) and (19) as particular cases.
If the ellipsoid has two axes equal, and is of the planetary or flattened
form,
6 = c = a r = f7rcV(l-*2); ............ (21)
(22)
(23)
In the extreme case of a disk, when e = 1 nearly.
L = 47r - 27rV(l - e2), ........................ (24)
M=N = -n*</(l-el) ......................... (25)
Thus in the limit from (14), (21) TA = 0, unless // = 0 ; and when /*' = 0,
In like manner the limiting values of TB, TO are zero, unless // = oo ,
and then
(27)
1897] AND ELECTRIC WAVES UPON SMALL OBSTACLES. 309
In all cases
t,_n*?+a>+ca (28)
gives the disturbance due to the ellipsoid.
If the ellipsoid of revolution be of .the ovary or elongated form,
tt = 6 = 0^(1-0; (29)
In the case of a very elongated ovoid L and M approximate to the value 2?r,
while N approximates to the form
(32)
vanishing when e = 1.
/w Two Dimensions.
The case of an elliptical cylinder in two dimensions may be deduced from
(12) by making c infinite, when the integration is readily effected. We find
T 4nrb M 4?ra
L = - 7, M = --- ,
a + b a+b
A and B are then given by (14) as before, and finally
_ ab (a + 6) ( (// -p)ux (// - /*)
corresponding to
(35)
In the case of circular section L = M = 2?r, so that
When b = 0, that is when the obstacle reduces itself to an infinitely thin
blade, ty vanishes unless /*' = 0 or // = oo . In the first case
0.-.0) +-^; ........................ (37)
in the second
. , a?ux
(/-oo) ^r=- — ...................... (38)
* There are slight errors in the values of L, M, N recorded for this case in both the works
cited.
310 ON THE INCIDENCE OF AERIAL [230
Aerial Waves.
We may now proceed to investigate the disturbance of plane aerial waves
by obstacles whose largest diameter is small in comparison with the wave-
length (X). The volume occupied by the obstacle will be denoted by T ; as
to its shape we shall at first impose ho restriction beyond the exclusion of
very special cases, such as would involve resonance in spite of the small
dimensions. The compressibilities and densities of the medium and of the
obstacle are denoted by m, m' ; a-, a ; so that if V, V be the velocities of
propagation
F2 = m/o-, V*=m'/<r' ......................... (39)
The velocity-potential of the undisturbed plane waves is represented by
$ = eikvt.eik*, .............................. (40)
in which k = 27r/\. The time factor eikvt, which operates throughout, may be
omitted for the sake of brevity.
The velocity-potential (T/T) of the disturbance propagated outwards from
T may be expanded in spherical harmonic terms *
+ ...}, ............ (41)
n(n + l) , (n-l).
where /. <0;r) = 1 + -^^ + 2 4
+ ...... +'2.4.6...2n(tfcr)» ................ (42)
At a great distance from the obstacle fn(ikr} = 1; and the relative importance
of the various harmonic terms decreases in going outwards with the order of
the harmonic. For the present purpose we shall need to regard only the
terms of order 0 and 1. Of these the term of order 0 depends upon the
variation of compressibility, and that of order 1 upon the variation of density.
The relation between the variable part of the pressure Bp, the conden-
sation s, and 0 is
rw-t-SE;
at a
so that during the passage of the undisturbed primary waves the rate at
which fluid enters the volume T (supposed for the moment to be of the same
quality as the surrounding medium) is
If the obstacle present an unyielding surface, its effect is to prevent the
entrance of the fluid (43) ; that is, to superpose upon the plane waves such a
* Theory of Sound, §§ 323, 324.
1897] AND ELECTRIC WAVES UPON SMALL OBSTACLES. 311
disturbance as is caused by the introduction of (43) into the medium. Thus,
if the potential of this disturbance be
^ = 8^, .............................. (44)
S0 is to be determined by the condition that when r — 0
so that £0= - k*TI4nr, and
This result corresponds with m' = oo representing absolute incompressibility.
The effect of finite compressibility, differing from that of the surrounding
medium, is readily inferred by means of the pressure relation (8p = ms). The
effect of the variation of compressibility at the obstacle is to increase the rate
of introduction of fluid into T from what it would otherwise be in the ratio
m : m' ; and thus (45) now becomes
or if we restore the factor eikvt and throw away the imaginary part of the
solution,
TT L tn — m , ,-.r , / 1)-,\
•f = -- — — cos k ( Vt - r) ................... (47)
A/r m
This is superposed upon the primary waves
x) ............................ (48)
When m = 0, i.e., when the material composing the obstacle offers no
resistance to compression, (47) fails. In this case the condition to be satisfied
at the surface of T is the evanescence of Bp, or of the total potential (<£ + ^).
In the neighbourhood of the obstacle <£ = 1 ; and thus if M ' denote the
electrical " capacity " of a conducting body of form T situated in the open,
•fr = —M'/r, r being supposed to be large in comparison with the linear
dimension of T but small in comparison with X. The latter restriction is
removed by the insertion of the factor e~ikr ; and thus, in place of (46). we
now have
t— ^ ............................ (49)
The value of M' may be expressed when T is in the form of an ellipsoid.
For a sphere of radius R,
M'=R; ................................. (50)
for a circular plate of radius R,
M' = 2RlTr ............................... (51)
312 ON THE INCIDENCE OF AERIAL [230
When the density of the obstacle (a-') is the same as that of the sur-
rounding medium, (47) constitutes the complete solution. Otherwise the
difference of densities causes an interference with the flow of fluid, giving
rise to a disturbance of order 1 in spherical harmonics. This disturbance is
independent of that already considered, and the flow in the neighbourhood of
the obstacle may be calculated as if the fluid were incompressible. We thus
fall back upon the problem considered in the earlier part of this paper, and
the results will be applicable as soon as we have established the corre-
spondence between density and conductivity.
In the present problem, if ^ denote the whole velocity-potential, the
conditions to be satisfied at any part of the surface of the obstacle are the
continuity of d%/dn and of <r%, the latter of which represents the pressure.
Thus, if we regard <T% as the variable, the conditions are the continuity of
(o"x) and of a-~l d (ay) I dn. In the conductivity problem the conditions to be
satisfied by the potential (^') are the continuity of ^' and of ^d-^jdn.
In an expression relating only to the external region where <r is constant,
it makes no difference whether we are dealing with o-% or with ^; and
accordingly there is correspondence between the two problems provided that
we suppose the ratio of //,'s in the one problem to be the reciprocal of the
ratio of the cr's in the other.
We may now proceed to the calculation of the disturbance due to an
obstacle, based upon the assumption that there is a region over which r is
large compared with the linear dimension of T, but small in comparison
with X. Within this region i/r is given by (8) if the motion be referred to
certain principal axes determined by the nature and form of the obstacle, the
quantities u, v, w being the components of flow in the primary waves. By
(41), (42), this is to be identified with
p-ikr t 1 x
+ = &— (l + ~}, (52)
r \ ikr)
when r is small in comparison with \ ; so that
C3wz) .„„.
— ................... (oo)
At a great distance from T, (52) reduces to
^ = ik (A,ux + Bzvy + C,wz) e~ikr (
—a term of order 1, to be added to that of zero order given in (46).
In general, the axis of the harmonic in (54) is inclined to the direction of
propagation of the primary waves ; but there are certain cases of exception.
For example, v and w vanish if the primary propagation be parallel to # (one
1897] AND ELECTRIC WAVES UPON SMALL OBSTACLES. 313
of the principal axes). Again, as for a sphere or a cube, A1} Bz, G3 may
be equal.
We will now limit ourselves to the case of the ellipsoid, and for brevity
will further suppose that the primary waves move parallel to x, so that
v — w — 0. The terms corresponding to u and v, if existent, are simply
superposed. If, as hitherto, <J> = eikx, u = ik; so that by (14), a being sub-
stituted for // and a' for /A,
In the intermediate region by (28) ^ = — TAxfr*, and thus at a great
distance
(56)
or on substitution of the values of A and k,
to(<r'-<r)
X2r2 to<r' + (<r-<r')L' '
Equations (46), (57) express the complete solution in the case supposed.
For an obstacle which is rigid and fixed, we may deduce the result by
supposing in our equations m' = oo , a-' = oo . Thus
Certain particular cases are worthy of notice. For the sphere L = |TT, and
If the ellipsoid reduce to an infinitely thin circular disk of radius c, T = 0
and the term of zero order vanishes. The term of the first order also
vanishes if the plane of the disk be parallel to x. If the plane of the disk be
perpendicular to x, 4nr — L is infinitesimal. By (21), (24) we get in this case
toT _8c3.
4>7T-L~ 3 '
(60)
If the axis of the disk be inclined to that of x, -fy retains its symmetry
with respect to the former axis, and is reduced in magnitude in the ratio of
the cosine of the angle of inclination to unity.
In the case of the sphere the general solution is
***
Theory of Sound, § 334. t L c. cit. § 335.
314 ON THE INCIDENCE OF AERIAL [230
Waves in Two Dimensions.
In the case of two dimensions (x, y) the waves diverging from a cylindrical
obstacle have the expression, analogous to (41),
1(kr)+..., .................. (62)
where S0> Sl ... are the plane circular functions of the various orders, and
3 A^r4
+..., ...... (63)
d(kr) ~ \2kr
As in the case of three dimensions already considered, the term of zero
order in -ty depends upon the variation of compressibility. If we again begin
with the case of an unyielding boundary, the constant S0 is to be found from
the condition that when r = 0
T denoting now the area of cross-section. When r is small,
dP0 (kr) _ 1
dr ~r'
and thus S0 = k2T/2-n;
................ (65)
when r is very great. This corresponds to (45).
In like manner, if the compressibility of the obstacle be finite,
The factor i~* = e~*iir ; and thus if we restore the time-factor e*7*, and reject
the imaginary part of the solution, we have
— /Tr, . ^.
2src08T<F'-r-*x)' ............... (67)
See Theory of Sound, § 341 ; Phil. Mag. April, 1897, p. 266. [Vol. iv. p. 290.]
1897] AND ELECTRIC WAVES UPON SMALL OBSTACLES. 315
corresponding to the plane waves
............................ (68)
In considering the term of the first order we will limit ourselves to the
case of the cylinder of elliptic section, and suppose that one of the principal
axes of the ellipse is parallel to the direction (#) of primary wave-propagation.
Thus in (34), which gives the value of -»/r at a distance from the cylinder
which is great in comparison with a and b, but small in comparison with X,
we are to suppose u = ik, v = 0, at the same time substituting a, a' for ///,
fi respectively. Thus for the region in question
,ab.ikx o-'-o-a + fr).
and this is to be identified with 8lD1(kr) when kr is small, i.e. with Sl/kr.
Accordingly
o _x ik2ab (a - <r) (a + 6)
r 2 a-'a + <rb
so that, at a distance r great in comparison with X, -fy becomes
'-- b) x
& >
T being written for trab. The complete solution for a great distance is given
by addition of (66) and (70), and corresponds to <£ = €**•*.
In the case of circular section (b = a) we have altogether *
+ =-k*a*e-*r (^ K-^ + ^ *1 , ...(71)
\2ikrJ ( 2m <r + cr r)
which may be realized as in (67). If the material be unyielding, the corre-
sponding result is obtained by making m' = oo , </ = oo in (71). The realized
value is then f
(72)
In general, if the material be unyielding, we get from (66), (70)
(73)
The most interesting case of a difference between a and b is when one of
them vanishes, so that the cylinder reduces to an infinitely thin blade. If
* Theory of Sound, § 343.
t Loc. cit. equation (17).
316 ON THE INCIDENCE OF AERIAL [230
b = 0, i/r vanishes as to both its parts ; but if a — 0, although the term of zero
order vanishes, that of the first order remains finite, and we have
(74)
in agreement with the value formerly obtained*.
It remains to consider the extreme case which arises when m' = 0. The
term of zero order in circular harmonics, as given in (66), then becomes
infinite, and that of the first order (70) is relatively negligible. The con-
dition to be satisfied at the surface of the obstacle is now the evanescence of
the total potential (<£ +--V/T), in which <£ = 1.
It will conduce to clearness to take first the case of the circular cylinder
(a). By (62), (63) the surface condition is
S0{y + \og($ika)} + l=Q ......................... (75)
Thus at a distance r great in comparison with A, we have
(76)
When the section of the obstacle is other than circular, a less direct
process must be followed. Let us consider a circle of radius p concentric
with the obstacle, where p is large in comparison with the dimensions of the
obstacle but small in comparison with X. Within this circle the flow may be
identified with that of an incompressible fluid. On the circle we have
(77)
(78)
of which the latter expresses the flow of fluid across the circumference. This
flow in the region between the circle and the obstacle corresponds to the
potential-difference (77). Thus, if R denote the electrical resistance between
the two surfaces (reckoned of course for unit length parallel to z),
S«{7 + log(W-27r.R} = l, ..................... (79)
and ^ = S0D0 (&r)> as usual.
The value of S0 in (79) is of course independent of the actual value of p,
so long as it is large. If the obstacle be circular,
The problem of determining R for an elliptic section (a, 6) can, as is well
known, be solved by the method of conjugate functions. If we take
x — c cosh £ cos 77, y = c sinh f sin 77, ............... (80)
* Phil. Mag. April 1897, p. 271. [Vol. rv. p. 295.] The primary waves are there supposed to
travel in the direction of + x, but here in the direction of - x.
1897] - AND ELECTRIC WAVES UPON SMALL OBSTACLES. 317
the confocal ellipses
are the equipotential curves. One of these, for which f is large, can be
identified with the circle of radius p, the relation between p and f being
An inner one, for which £=£„, is to be identified with the ellipse (a, b),
so that
a = c cosh f0, b = c sinh £0,
whence c2 = u? - 62, tanh %0 = bfa.
Thus 27r£ = £-£0 = log; ................... (32)
and then (79) gives as applicable at a great distance
— ,vy
T ~_~ i M »'7, /„ i JAI I 2ikr) '***v
The result for an infinitely thin blade is obtained by merely putting 6 = 0
in (83).
For some purposes the imaginary part of the logarithmic term may be
omitted. The realized solution is then
/jr \* coBft(Ft-r-fr)
UtoV 7 + log {fk (a + &)}'
7 + log {|fc(a + 6)}
corresponding, as usual, to
<f> = cosk(Vt + x) (85)
Electrical Applications.
The problems in two dimensions for aerial waves incident upon an
obstructing cylinder of small transverse dimensions are analytically identical
with certain electric problems which will now be specified. The general
equation (v2 + A;2) = 0 is satisfied in all cases. In the ordinary electrical
notation V2 = l/K/j,, F'2 = 1 jK'p! ; while in the acoustical problem F2 = ra/<r,
V'2 = m'/<r'. The boundary conditions are also of the same general form.
Thus if the primary waves be denoted by 7 = eikx, y being the magnetic force
parallel to z, the conditions to be satisfied at the surface of the cylinder are
the continuity of 7 and of K~l dy/dn. Comparing with the acoustical
conditions we see that K replaces or, and consequently (by the value of F2)
fj, replaces 1/w. These substitutions with that of 7, or c (the magnetic
induction), for ^ and </> suffice to make (66), (70) applicable to the electrical
318 ON THE INCIDENCE OF AERIAL [230
problem. For example, in the case of the circular cylinder, we have for the
dispersed wave
' (86)
corresponding to the primary waves
c = eikx ................................... (87)
An important particular case is obtained by making K' = oo , yu/ = 0, in
such a way that V remains finite. This is equivalent to endowing the
obstacle with the character of a perfect conductor, and we get
which, when realized, coincides with (72).
The other two-dimensional electrical problem is that in which everything
is expressed by means of R, the electromotive intensity parallel to z. The
conditions at the surface are now the continuity of R and of p^dR/dri.
Thus K and p are simply interchanged, /j, replacing a and K replacing 1/ra
in (66), (70), </> and i/r also being replaced by R. In the case of the circular
cylinder
' (89)
corresponding to the primary waves
R = e** .................................. (90)
If in order to obtain the solution for a perfectly conducting obstacle we
make K' = oo , // = 0, (89) becomes infinite, and must be replaced by the
analogue of (83). Thus for the perfectly conducting circular obstacle
which may be realized as in (84).
The problem of a conducting cylinder is treated by Prof. J. J. Thomson in
his valuable Recent Researches in Electricity and Magnetism, § 364 ; but his
result differs from (84), not only in respect to the sign of ^X, but also in the
value of the denominator*. The values here given are those which follow
from the equations (9), (17) of § 343 Theory of Sound.
Electric Waves in Three Dimensions.
In the problems which arise under this head the simple acoustical
analogue no longer suffices, and we must appeal to the general electrical
* It should be borne in mind that y here is the same as Prof. Thomson's log y.
1897] AND ELECTRIC WAVES UPON SMALL ORSTACLES. 319
equations of Maxwell. The components of electric polarization (f, g, h) and
of magnetic force (a, /3, 7), being proportional to eikvt, all satisfy the funda-
mental equation
(V2 + £2) = 0. .............................. (92)
and they are connected together by such relations as
da , Tro fdq dh\
or — = 4?rF *(-/--!- ), ....................... (94)
dt * dz dy)
in which any differentiation with respect to t is equivalent to the introduction
of the factor ikV. Further
dj, dk *« + *8 + £ g
as dy dz dx dy dz
The electromotive intensity (P, Q, R) and the magnetization (a, b, c) are
connected with the quantities already defined by the relations
a, b, c = /*(«, /S, 7); ......... (96)
in which K denotes the specific inductive capacity and /i the permeability ;
so that F~2 = Kft.
The problem before us is the investigation of the disturbance due to a
small obstacle (K', /*') situated at the origin, upon which impinge primary
waves denoted by
/o = 0, <7o = 0, A, = ««* ..................... (97)
or, as follows from (94),
a0 = 0, 00 = 47rFeto, 7o = 0 ................... (98)
The method of solution, analogous to that already several times employed,
depends upon the principle that in the neighbourhood of the obstacle and up
to a distance from it great in comparison with the dimensions of the obstacle
but small in comparison with \, the condition at any moment may be
identified with a steady condition such as is determined by the solution of a
problem in conduction. When this is known, the disturbance at a distance
from the obstacle may afterwards be derived.
We will commence with the case of the sphere, and consider first the
magnetic functions as disturbed by the change of permeability from ^ to /*'.
Since in the neighbourhood of the sphere the problem is one of steady
distribution, ot, /3, 7 are derivable from a potential. By (98), in which we
may write eikx = 1, the primary potential is 4nrVy; so that in (1) we are to
take u = 0, v = 4-rrF, w = 0. Hence by (20) a, ft, 7 for the disturbance are
given by
320 ON THE INCIDENCE OF AERIAL [230
where .' f__4,ri< ......................... (99)
In like manner f, g, h are derivable from a potential %. The primary
potential is z simply, so that in (1), u = 0, v = 0, w = 1. Hence by (20)
K'-K a?z
X = -£T^2K^> ........................ (1<
from which /, g, h for the disturbance are derived by simple differentiations
with respect to x, y, z respectively.
Since /. g, h, a, /8, 7 all satisfy (92), the values at a distance can be
derived by means of (41). The terms resulting from (99), (100) are of the
second order in spherical harmonics. When r is small,
and when r is great
r-i e-*r
so that, as regards an harmonic of the second order, the value at a distance
will be deduced from that in the neighbourhood of the origin by the intro-
duction of the factor - ^kzr2e~ikr. Thus, for example,/ in the neighbourhood
of the origin is
so that at a great distance we get
f__K^K.**^ ...................... (102)
In this way the terms of the second order in spherical harmonics are at
once obtained, but they do not constitute the complete solution of the
problem. We have also to consider the possible occurrence of terms of other
orders in spherical harmonics. Terms of order higher than the second are
indeed excluded, because in the passage from r small to r great they suffer
more than do the terms of the second order. But for a like reason it may
happen that terms of order zero and 1 in spherical harmonics rise in relative
importance so as to be comparable at a distance with the term of the second
order, although relatively negligible in the neighbourhood of the obstacle.
The factor, analogous to —%feir*e~ikr for the second order, is for the first order
ikre~i}cr, and for zero order e~ikr. Thus, although (101) gives the value of f
with sufficient completeness for the neighbourhood of the obstacle, (102) may
need to be supplemented by terms of the first and zero orders in spherical
harmonics of the same importance as itself. The supplementary terms may
be obtained without much difficulty from those already arrived at by means
of the relations (93), (94), *(95) ; but the process is rather cumbrous, and
1897] AND ELECTRIC WAVES UPON SMALL OBSTACLES. 321
it seems better to avail ourselves of the forms deduced by Hertz * for electric
vibrations radiated from a centre.
If we write Tl = Ae~ikr/r, the solution corresponding to an impressed
electric force acting at the origin parallel to z is
(104)
These values evidently satisfy (92) since H does so, and they harmonize with
(93), (94), (95).
In the neighbourhood of the origin, where kr is small, e~ikr may be
identified with unity, so that II = A jr. In this case (103) may be written
/•__^!E <M M
' dxdz' 9 dydz' dz* '
and all that remains is to identify — dU/dz with ^ in (100). Accordingly
^ = -a° ......................... <105>
The values of f, g, h in (103) are now determined. Those of a, /3, 7 are
relatively negligible in the neighbourhood of the origin. At a great distance
we have
f=-A
J ~ dxdz \ r ~ r dxdz
so that (103), (104) may be written
K' — K k*a?e~ikr ( xz yz a? + y2\
f>9>h=vr-^r- 7— (-^» ~^> -pr-)' <106)
a,
y x \
r' "r' °J
These equations give the values of the functions for a disturbance
radiating from a small spherical obstacle, so far as it depends upon (K' — K).
We have to add a similar solution dependent upon the change from /j. to ///.
In this (103), (104) are replaced by
_, ___.
2 * T 2 ' 2 ' '
_
dxdy ' F2 dx* rf*2 ' F2
* Ausbreitung tier electrischen Kraft, Leipzig, 1892, p. 150. It may be observed that the
solution for the analogous but more difficult problem relating to an elastic solid was given much
earlier by Stokes (Camb. Trans., Vol. ix. p. 1, 1849). Compare Theory of Sound, 2nd ed. § 378.
R. IV. 21
322 ON THE INCIDENCE OF AERIAL [230
where H = Be~ikrlr, corresponding to an impressed magnetic force parallel
to y. In the neighbourhood of the origin (108) becomes
a d2H ft _ d-Tl 7 _ d*H
Vz dy~ ' V* dzdy '
so that -f in (99) is to be identified with - V2dU/dy. Thus
-.'...'. ;• •B=-1f?T^ <110>
At a great distance we have
...(111)
a, ft, 7_ p' -p t?a3e-ikr( xy tf + z* _zy\
"4arV ~ p' + ty r' \ r* ' r* r* J '
By addition of (111) to (106) and of (112) to (107) we obtain the com-
plete values of f, g, h, a, ft, 7 when both the dielectric constant and the
permeability undergo variation. The disturbance corresponding to the
primary waves h = eikx is thus determined.
When the changes in the electric constants are small, (106), (111) may
be written
(113)
\ f\. i~ p. 'i /
' _., i &Kiiz\
9 = ^.e
where T=§TTO?, Ar=27r/X. These are the results given formerly* as applic-
able in this case to an obstacle of volume T and of arbitrary form. When
the obstacle is spherical and &KJ K is not small, it was further shown that
&KJK should be replaced by 3(K' — K)/(K' + 2^T)f, and similar reasoning
would have applied to A/A //A.
The solution for the case of a spherical obstacle having the character of a
perfect conductor may be derived from the general expressions by supposing
that K' = x , and (in order that V may remain finite) // = 0. We get
from (106), (111),
* "Electromagnetic Theory of Light," Phil. Mag. Vol. xii. p. 90 (1881). [Vol. i. p. 526.}
t [1902. The " 3 " was inadvertently omitted in the original of the present paper.]
1897] AND ELECTRIC WAVES UPON SMALL OBSTACLES. 323
(nfi)
in agreement with the results of Prof. J. J. Thomson*. As was to be
expected, in every case the vectors (f, g, h), (a, ft, 7), (x, y, z) are mutually
perpendicular.
Obstacle in the Form of an Ellipsoid.
The case of an ellipsoidal obstacle of volume T, whose principal axes are
parallel to those of x, y, z, i. e. parallel to the directions of propagation and of
vibration in the primary waves, is scarcely more complicated. The passage
from the values of the disturbance in the neighbourhood of the obstacle to
that at a great distance takes place exactly as in .the case of the sphere.
The primary magnetic potential in the neighbourhood of the obstacle is
4?r Vy, and thus, as before, u = 0, v = 4nrV, w = Q in (1). Accordingly, by (14),
A = 0, C = 0 ; and (28) gives
' gy, .................. (119)
3
r
47r/u, + (fjL —
corresponding to (99) for the sphere. In like manner the electric potential is
— /i on\
x==~ *7rK + (K'-K}N ^ '
These potentials give by differentiation the values of a, /3, 7 and f, g, h
respectively in the neighbourhood of the ellipsoid. Thus at a great distance
we obtain for the part dependent on (K1 — K}, as generalizations of (106),
(107),
y _x
' ' '
_ __
4-TrK ~±TrK + (K'-K)N r \r' r'
To these are to be added corresponding terms dependent upon (//— /i), viz.: —
'-, 0, -?); ...... (128)
r' rj
a, 0, 7 = ^ -n _ k*Te~ikr (xy x> + z* _zy
4-TrF 4nrfji + (»' - p) M r \ r* ' r2 r*
* Recent Researches, § 377, 1893.
21—2
324 ON THE INCIDENCE OF AERIAL [230
The sum gives the disturbance at a distance due to the impact of the
primary waves,
(125)
upon the ellipsoid T of dielectric capacity K' and of permeability /*'.
As in the case of the sphere, the result for an ellipsoid of perfect conduc-
tivity is obtained by making K' = oo , /*' = 0. Thus
(T xz T
, tee-** (T xz T z\
~^(N^ + 4^Mr)'
.(127)
Next to the sphere the case of greatest interest is that of a flat circular
disk (radius = R). The volume of the obstacle then vanishes, but the effect
remains finite in certain cases notwithstanding. Thus, if the axis of the disk
be parallel to x, that is to the direction of primary propagation, we have
(21), (25),
T 4R3 T
In spite of its thinness, the plate being a perfect conductor disturbs the
electric field in its neighbourhood; but the magnetic disturbance vanishes,
the zero permeability having no effect upon the magnetic flow parallel to its
face. If the axis of the disk be parallel to y (see (24)),
and if the axis be parallel to z,
£-0 * -0
N 4>-rr — M
so that in this case the obstacle produces no effect at all.
Circular Aperture in Conducting Screen.
The problem proposed is the incidence of plane waves (A0 = e***) upon an
infinitely thin screen at x = 0 endowed with perfect electric conductivity and
perforated by a circular aperture. In the absence of a perforation there
would of course be no waves upon the negative side, and upon the positive
side the effect of the screen would merely be to superpose the reflected waves
denoted by /*0 = - e~ikx. We wish to calculate the influence of a small
circular aperture of radius R,
1897] AND ELECTRIC WAVES UPON SMALL OBSTACLES. 325
In accordance with the general principle the condition of things is
determined by what happens in the neighbourhood of the aperture, and this
is substantially the same as if the wave-length were infinite. The problem
is then expressible by means of a common potential. The magnetic force at
a distance from the aperture on the positive side is altogether 87rV, and on
the negative side zero ; while the condition to be satisfied upon the faces of
the screen is that the force be entirely tangential. The general character of
the flow is indicated in Fig. 1.
Fig. 1. Fig. 2. Fig. 3.
The problem here proposed is closely connected with those which we have
already considered where no infinite screen was present, but a flat finite
obstacle, which may be imagined to coincide with the proposed aperture.
The primary magnetic field being /9 = 4>7rV, and the disk of radius R being of
infinite permeability, the potential at . a distance great compared with R (but
small compared with X) is by (27), (28)
(132)
By the symmetry the part of the plane x = 0 external to the disk is not
crossed by the lines of flow, and thus it will make no difference in the
conditions if this area be filled up by a screen of zero permeability. On the
other hand, the part of the plane # = 0 represented by the disk is met
normally by the lines of flow. This state of things is indicated in Fig. 2.
The introduction of the lamina of zero permeability effects the isolation
of the positive and negative sides. We may therefore now reverse the flow
upon the negative side, giving the state of things indicated in Fig. 3. But
the plate of infinite permeability then loses its influence and may be removed,
so as to re-establish a communication between the positive and negative sides
through an aperture. The passage from the present state of things to that
of Fig. 1 is effected by superposition upon the whole field of ft = 4-TrF, so as to
destroy the field at a distance from the aperture upon the negative side and
upon the positive side to double it.
326 ON THE INCIDENCE OF AERIAL AND ELECTRIC WAVES. [230
As regards the solution of the proposed problem we have then on the
positive side
and on the negative side
Thus on the negative side at a distance great in comparison with the wave-
length we get, as in (99), (111), (112),
- ^ -?
On the positive side these values are to be reversed, and addition made of
A, = eite_<r«* £0 = 47rF(e'^ + e-to), ......... (137)
representing the plane waves incident and reflected.
The solution for h in (135) may be compared with that obtained (27), (28)
in a former paper*, where, however, the primary waves were supposed to
travel in the positive, instead of, as here, in the negative direction. It had at
first been supposed that the solution for <£ there given might be applied
directly to h, which satisfies the condition (imposed upon <£) of vanishing
upon the faces of the screen. If this were admitted, as also g = 0 throughout,
the value of h would follow by (95). The argument was, however, felt to be
insufficient on account of the discontinuities which occur at the edge of the
aperture, and the value now obtained, though of the same form, is doubly
as great.
* *' On the Passage of Waves through Apertures in Plane Screens, and Allied Problems,"
Phil. Mag. Vol. XLIII. p. 264 (1897). [Vol. iv. p. 287.]
231.
r-
ON THE PROPAGATION OF ELECTRIC WAVES ALONG
CYLINDRICAL CONDUCTORS OF ANY SECTION.
[Philosophical Magazine, XLIV. pp. 199—204, 1897.]
THE problem of the propagation of waves along conductors has been
considered by Mr Heaviside and Prof. J. J. Thomson, for the most part with
limitation to the case of a wire of circular section with a coaxal sheath
serving as a return. For practical applications it is essential to treat the
conductivity of the wire as finite; but for some scientific purposes the
conductivity may be supposed perfect without much loss of interest. Under
this condition the problem is so much simplified that important extensions
may be made in other directions. For example, the complete solution may
be obtained for the case of parallel wires, even although the distance between
them be not great in comparison with their diameters.
We may start from the general equations of Maxwell involving the
electromotive intensity (P, Q, R) and the magnetic induction (a, b, c),
introducing the supposition that all the functions are proportional to ei{pt+mz},
and further that m=p/V, just as in the case of uninterrupted plane waves
propagated parallel to z. Accordingly d*/dt2 = V*d2/dz2, and any equation
such as
d*P d*P d*P 1 d*P
dtf + djf + df^T* ~dr
fJ2P d-P
reduces to ^ + ^ = ° <2>
They may be summarized in the form
+ £)CP, Q,R,a,b,c) = 0 (3)
dx2 dy*J ^
328 ON THE PROPAGATION OF ELECTRIC WAVES [231
The case to be here treated is characterized by the conditions R = 0, c = 0;
but it would suffice to assume one of them, say the latter. Since in general
throughout the dielectric
dc/dt = dP/dy-dQ/da;, ........................ (4)
*
it follows that P and Q are derivatives of a function (<£), also proportional to
ei(pt+mz)^ which as a function of x and y may be regarded as a potential since
it satisfies the form (2). Thus dP/dx + dQ/dy = Q, from which it follows
that dRjdz and R vanish. It will be convenient to express all the functions
by means of <f>. We have at once
P = d(f>/dx, Q = d<J)/dy, E = 0 ................... (5)
Again, by the general equation analogous to (4), since R = 0, ipa = imQ ;
so that
a=V-ld<j>/dy, b = -V-*d$ldx, c = 0 ............. (6)
Thus the same function 0 serves as a potential for P, Q and as a stream-
function for a, b.
The problem is accordingly reduced to dependence upon a simple potential
problem in two dimensions. Throughout the dielectric <f> satisfies
(7)
At the boundary of a conductor, supposed to be perfect, the condition is
that the electromotive intensity be entirely normal. So far as regards the
component parallel to z this is satisfied already, since R = 0 throughout.
The remaining condition is that </> be constant over the contour of any
continuous conductor. This condition secures also that the magnetic in-
duction shall be exclusively tangential.
It is to be observed that R is not equal to dffr/dz. The former quantity
vanishes throughout, while d<j>/dz remains finite, since <j> <x e{ {pt+mz} . In-
asmuch as <j> satisfies Laplace's equation in two dimensions, but not in three,
it will be convenient to use language applicable to two dimensions, referring
the conductors to their sections by the plane xy.
If a boundary of a conductor be in the form of a closed curve, the included
dielectric is incapable of any vibration of the kind now under consideration.
For a function satisfying (7) and retaining a constant value over a closed
contour cannot deviate from that value in the interior. Thus the derivatives
of </> vanish, and there is no disturbance. The question of dielectric vibrations
within closed tubes, when ra is not limited to equality with p/V, was con-
sidered in a former paper*.
* Phil. Mag. Vol. XLIII. p. 125 (1897). [Vol. iv. p. 276.]
1897] ALONG CYLINDRICAL CONDUCTORS OF ANY SECTION. 329
For the case of a dielectric bounded by two planes perpendicular to x we
may take
giving p = eHpt+mz)>
(10)
in which, as usual, m—pj V. Since Q = 0, R = 0 throughout, the dielectric
may be regarded as limited by conductors at any planes (perpendicular to x)
that may be desired.
If the dielectric be bounded by conductors in the form of coaxal circular
cylinders, we have the familiar wire with sheath return, first, I believe,
considered on the basis of these equations by Mr Heaviside. We may take,
with omission of a constant addition to log r which has here no significance,
(11)
gvng ,, = *,. y-, o) , ............... (12)
(13)
And here again it makes no difference to these forms at what points (rl} r2)
the dielectric is replaced by conductors.
For the moment these simple examples may suffice to illustrate the
manner in which the propagation along z takes place, and to show that $
is determined by conditions completely independent of p and its associated m.
In further discussions it will save much circumlocution to suppose that p and
m are zero and thus to drop the exponential factor. The problem is then
strictly reduced to two dimensions and relates to charges and steady currents
upon cylindrical conductors, the currents being still entirely superficial.
When <f> is once determined for any case of this kind, the exponential factor
may be restored at pleasure with an arbitrary value assigned to p and the
corresponding value, viz. p/V, to m.
The usual expressions for electric and magnetic energies will then apply,
everything being reckoned per unit length parallel to z. It suffices for
practical purposes to limit ourselves to the case of a single outgoing and a
single return conductor. We may then write
Electric energy .<&"&!., ........................ (14)
2 x capacity
Magnetic energy = |- x self-induction x (current)2 ; ......... (15)
and the value of the self-induction in the latter case is the reciprocal of that
of the capacity in the former.
330 ON THE PROPAGATION OF ELECTRIC WAVES [231
Thus, for a dielectric bounded by coaxal conductors at r = r^ and r = >:„
we have $ = log r, and
self-induction = (capacity)-1 = 2 log - ................ (16)
Among the cases for which the solution can be completely effected may be
mentioned that of a dielectric bounded by confocal elliptical cylinders.
More important in practice is the case of parallel circular wires. In
Lecher's arrangement, which has been employed by numerous experimenters,
the wires are of equal diameter ; and it is usually supposed to be necessary to
maintain them at a distance apart which is very great in comparison with
that diameter. The general theory above given shows that there is no need
for any such restriction, the manner and velocity of propagation along the
length being the same whatever may be the character of the cross-section of
the system.
The form of <f>, and the self-induction of the system, may be determined
in this case, whatever may be the radii (oj, a.2) of the wires and the distance
(6) between their centres. If rl} r2 are the distances of any point P in the
plane from fixed points Oi, 02, the equipotential curves for which <£, equal to
Iog(?*2/r1), assumes constant values are a system of circles, two of which can
be identified with the boundaries of the conductors. The details of the
investigation, consisting mainly of the geometrical relations between the
ultimate points 0j, 02 and the circles of radii a,, a,, are here passed over.
The result for the self-induction per unit length L, or for the capacity, may
be written*
As was to be expected, L vanishes when 6 = a, + a*, that is, when the
conductors are just in contact.
When Oj, a? are small in comparison with b, the approximate value is
.................. (18)
•or, if a, = «, = <«, i = 4log- ......................... (19)
The first term of (19) is the value usually given. The same expression
represents the reciprocal of the capacity of the system per unit length.
In the application of Lecher's arrangement to the investigation of re-
fractive indices, we have to consider the effect of a variation of the dielectric
* Compare Mncdonald, Camb. Phil. Trans. Vol. xv. p. 303 (1894).
1897] ALONG CYLINDRICAL CONDUCTORS OF ANY SECTION. 331
occurring at planes for which z is constant. It will be seen that no new
difficulty arises in the case of systems for which the appropriate function <£ in
two dimensions can be assigned.
Regarding <£ as a given function, e.g. log r for the case of a coaxal wire
and sheath (compare (11)), we may take as the solution for any length of
uniform dielectric
, , , , -,, (20)
dx dy )
j± \
...(21)
in which pa = a, &c., and provision is made for waves travelling in both
directions.
At a plane where the dielectric changes, the conditions to be satisfied are
the continuity of P, Q and of a, /8 ; and this is secured if
A eimz + Be~imz, (22)
~(Aeimz-Be-imz), (23)
are continuous. It will be seen that the conditions are altogether indepen-
dent of the section of the conductors, being the same in fact as if there were
no conductors and we were dealing with infinite plane waves represented
by <£ = x.
As a particular case we may suppose that waves travelling in the negative
direction in the dielectric ( F, //,) meet at z = 0 a dielectric of altered character
(V, //). The expressions (20), (21) represent the incident (A) and reflected
(B) waves. For the second medium it suffices to accent V and /*, writing
also A' for A and 0 for B. Thus (22), (23) give
by which B and A' are determined. For the reflected wave
or if the difference between the dielectrics relate only to the dielectric
constants (K, K'),
'--'
in agreement with Young's well-known optical formula.
Whether the dielectric consist of uniform portions with discontinuous
changes of character at the boundaries, or whether it be a continuous function
of z, the solution of the problem is the same, whatever be the character of the
cylindrical conductors. It is only the form of <£ that is influenced by the
latter consideration.
232.
THE ELECTRO-CHEMICAL EQUIVALENT OF SILVER.
[Nature, LVI. p. 292, 1897.]
IN Nature, Vol. LVI. p. 259, Mr Griffiths points out that recent comparisons
of the values of the mechanical equivalent of heat, obtained by mechanical
and electrical methods, suggest that the adopted value of the equivalent of
silver may be in error to the extent of Y^TT- This adopted value rests,
I believe, almost entirely upon experiments made by Kohlrausch, and by
myself with Mrs Sidgwick in 1882 ; and the question has been frequently put
to me as to the limits within which it is trustworthy. Such questions are
more easily asked than answered, and experience shows that estimates of
possible error given by experimenters themselves are usually framed in far too
sanguine a spirit.
When our work was undertaken the generally accepted number was '01136
obtained by Kohlrausch in 1873. Mascart had recently given '01124, sub-
sequently corrected to '011156. The uncertainty, therefore, at that time
amounted to at least 1 per cent. The experiments of Mrs Sidgwick and
myself were very carefully conducted, and we certainly hoped to have attained
an accuracy of -^wo^- So ^ar as errors that can be eliminated by repetition
are concerned, this was doubtless the case, as is proved by an examination of
our tabular results. But, as every experimenter knows, or ought to know,
this class of errors is not really the most dangerous. Security is only to be
obtained by coincidence of numbers derived by different methods and by
different individuals. It was, therefore, a great satisfaction to find our
number (Phil. Trans. 1884) ('011179) confirmed by that of Kohlrausch
('011183), resulting from experiments made at about the same time.
It would, however, in my opinion, be rash to exclude the question of an
error of y^. Indeed, I have more than once publicly expressed surprise at
the little attention given to this subject in comparison with that lavished
upon the ohm. I do not know of any better method of measuring currents
absolutely than that followed in 1882, but an ingenious critic would doubtless
be able to suggest improvements in details. The only thing that has occurred
to me is that perhaps sufficient attention was not given to the change in
dimensions that must accompany the heating of the suspended coil when
conveying the current of £ ampere. Recent experiments upon the coil (which
exists intact) show that, as judged by resistance, the heating effect due to
this current is 2|° C. But it does not appear possible that the expansion of
mean radius thence arising could be comparable with y^j. [See Vol. II. p. 278.]
233.
ON AN OPTICAL DEVICE FOR THE INTENSIFICATION OF
PHOTOGRAPHIC PICTURES.
[Philosophical Magazine, XLIV. pp. 282 — 285, 1897.]
WHETHER from insufficient exposure or from other causes, it not unfre-
quently happens that a photographic negative is deficient in density, the ratio
of light-transmissions for the transparent and opaque parts being too low for
effective contrast. In many cases an adequate remedy is found in chemical
processes of intensification, but modern gelatine plates do not always lend
themselves well to this treatment.
The method now proposed may be described as one of using the negative
twice over. Many years ago a pleasing style of portrait was current depen-
dent upon a similar principle. A thin positive transparency is developed
upon a collodion plate by acid pyrogallol. Viewed in the ordinary way by
holding up to the light, the picture is altogether too faint; but when the
film side is placed in contact with paper and the combination viewed
by reflected light, the contrast is sufficient. Through the transparent
parts the paper is seen with but little loss of brilliancy, while the opaque
parts act, as it were, twice over, once before the light reaches the paper, and
again after reflexion on its way to the eye. For this purpose it is necessary
that the deposit, constituting the more opaque parts of the picture, be of
such a nature as not itself to reflect light back to the eye in appreciable
degree — a condition very far from being satisfied by ordinary gelatine
negatives. But by a modification of the process the objection may be met
without much difficulty.
To obtain an intensified copy (positive) of a feeble negative, a small source
of illumination, e.g. a candle, is employed, and it is placed just alongside of
the copying-lens. The white paper is replaced by a flat polished reflector,
and the film side of the negative is brought into close contact with it. On
334 ON AN OPTICAL DEVICE FOR THE [233
the other side of the negative and pretty close to it is a field, or condensing,
lens of such power that the light from the candle is made parallel by it.
After reflexion the light again traverses the lens and forms an image of the
candle centred upon the photographic copying-lens. The condenser must be
large enough to include the picture and must be free from dirt and scratches ;
otherwise it does not need to be of good optical quality. If the positive is to
preserve the original scale, the focal length of the condenser must be about
twice that of the copying-lens.
In carrying this method into execution there are two points which require
special attention. The first is the elimination of false light reflected from the
optical surfaces employed. As regards the condensing-lens, the difficulty is
easily met by giving it a moderate slope. But the light reflected from the
glass face of the negative to be copied is less easily dealt with. If allowed
to remain, it gives a uniform illumination over the whole field, which in many
cases would go far to neutralize the advantages otherwise obtainable by the
method. The difficulty arises from the parallelism of the two surfaces of the
negative, and is obviated by using for the support of the film a glass whose
faces are inclined. The false light can then be thrown to one side and
rendered inoperative. In practice it suffices to bring into contact with the
negative (taken as usual upon a parallel plate) a wedge-shaped glass of equal
or greater area, the reflexion from the adjoining faces being almost destroyed
by the interposition of a layer of turpentine. By these devices the false
light is practically eliminated, and none reaches the sensitive film but what
has twice traversed the original negative.
The other point requiring attention is to secure adequate superposition of
the negative and its image in the associated reflector. On account of the
slight lateral interval between the copying-lens and the source of light, the
incidence of the rays upon the reflector is not accurately perpendicular, and
thus any imperfection of contact between the negative film and the reflector
leads to a displacement prejudicial to definition. The linear displacement is
evidently 2£ sin 6, if t denote the interval between the surfaces and 0 the
angle of incidence, and it can be calculated in any particular case. It is the
necessity for a small t that imposes the use of a speculum as a reflector. In
practice 20 can easily be reduced to ^ ; so that if t were ^ inch, the dis-
placement would not exceed -^ inch, and for most purposes might be
disregarded*. The obliquity 6 could be got rid of altogether by introducing
the light with the aid of a parallel glass reflector placed at 45°; but this com-
plication is hardly to be recommended.
The scale of the apparatus depends, of course, upon the size of the
negatives to be copied. In my own experiments ^-plates (4£ in. x 3£ in.)
* If the glass of the negative were flat, its approximation to the reflector might be much closer
than is here supposed.
1897]
INTENSIFICATION OF PHOTOGRAPHIC PICTURES.
335
were employed. The condenser is of plate-glass 6 in. diameter and 36 in.
focus. The reflector is of silver deposited on glass*. The wedge-shaped
glass f attached to the negative with turpentine is 4 x 4 ins. and the angle
between the faces is 2°. The photographic lens is of 3 inch aperture and
about 18 inch principal focus. It stands at about 36 inches from the negative
to be copied. [Inch = 2'54 cm.]
The accompanying sketch shows the disposition of some of the parts.
It represents a section by a horizontal plane. A is the condensing-lens,
B the wedge, C the negative temporarily cemented to B by fluid turpentine,
D the speculum.
[1902. An almost identical procedure had been described about three
years earlier by Mach (Eder's Jakrbuch fur Photographic}. The method of
double transmission was employed in a former research (Phil. Mag. Oct. 1892;
Vol. iv. of this collection, p. 10).]
* For a systematic use of the method a reflector of speculum metal would probably be
preferable.
f It is one of those employed for a similar purpose in the projection of Newton's rings (Proc.
Roy. Iiist. March, 1893 ; Nature, Vol. XLVIII. p. 212 [Vol. iv. p. 54]).
234.
ON THE VISCOSITY OF HYDROGEN AS AFFECTED BY
MOISTURE.
[Proceedings of the Royal Society, LXII. pp. 112—116, 1897.]
IN Sir W. Crookes's important work upon the viscosity of gases* the case
of hydrogen was found to present peculiar difficulty. " With each improve-
ment in purification and drying I have obtained a lower value for hydrogen,
and have consequently diminished the number expressing the ratio of the
viscosity of hydrogen to that of air. In 1876 I found the ratio to be 0'508.
In 1877 I reduced this ratio to 0'462. Last year, with improved apparatus,
I obtained the ratio O458, and I have now got it as low as 0'4439" (p. 425).
The difficulty was attributed to moisture. Thus (p. 422) : " After working at
the subject for more than a year, it was discovered that the discrepancy arose
from a trace of water obstinately held by the hydrogen — an impurity which
behaved as I explain farther on in the case of air and water vapour."
When occupied in 1888 with the density of hydrogen, I thought that
viscosity might serve as a useful test of purity, and I set up an apparatus
somewhat on the lines of Sir W. Crookes. A light mirror, 18 mm. in
diameter, was hung by a fine fibre (of quartz I believe) about 60 cm. long.
A small attached magnet gave the means of starting the vibrations whose
subsidence was to be observed. The viscosity chamber was of glass, and
carried tubes sealed to it above and below. The window, through which the
light passed to and fro, was of thick plate glass cemented to a ground face.
This arrangement has great optical advantages, and though unsuitable for
experiments involving high exhaustions, appeared to be satisfactory for the
purpose in hand, viz., the comparison of various samples of hydrogen at
atmospheric pressure. The Topler pump, as well as the gas generating
apparatus and purifying tubes, were connected by sealing. But I was not
able to establish any sensible differences among the various samples of
hydrogen experimented upon at that time.
* Phil. Trans. 1881, p. 387.
1897] ON THE VISCOSITY OF HYDROGEN AS AFFECTED BY MOISTURE. 337
In view of the importance of the question, I have lately resumed these
experiments. If hydrogen, carefully prepared and desiccated in the ordinary
way, is liable to possess a viscosity of 10 per cent, in excess, a similar un-
certainty in less degree may affect the density. I must confess that I was
sceptical as to the large effect attributed to water vapour in gas which had
passed over phosphoric anhydride. Sir W. Crookes himself described an
experiment (p. 428) from which it appeared that a residue of water vapour
in his apparatus indicated the viscosity due to hydrogen, and, without
deciding between them, he offered two alternative explanations. Either the
viscosity of water vapour is really the same as that of hydrogen, or under
the action of the falling mercury in the Sprengel pump decomposition
occurred with absorption of oxygen, so that the residual gas was actually
hydrogen. It does not appear that the latter explanation can be accepted, at
any rate as regards the earlier stages of the exhaustion, when a rapid current
of aqueous vapour must set in the direction of the pump ; but if we adopt
the former, how comes it that small traces of water vapour have so much
effect upon the viscosity of hydrogen ?
It is a fact, as was found many years ago by Kundt and Warburg* (and
as I have confirmed), that the viscosity of aqueous vapour is but little greater
than that of hydrogen. The numbers (relatively to air) given by them are
0-5256 and 0'488. It is difficult to believe that small traces of a foreign gas
having a six per cent, greater viscosity could produce an effect reaching to
10 per cent.
In the recent experiments the hydrogen was prepared from amalgamated
zinc and sulphuric acid in a closed generator constituting in fact a Smee cell,
and it could be liberated at any desired rate by closing the circuit externally
through a wire resistance. The generating vessel was so arranged as to admit
of exhaustion, and the materials did not need to be renewed during the
whole course of the -experiments. The gas entered the viscosity chamber
from below, and could be made to pass out above through the upper tube
(which served also to contain the fibre) into the pump head of the Tdpler.
By suitable taps the viscosity chamber could be isolated, when observations
were to be commenced.
The vibrations were started by a kind of galvanometer coil in connexion
(through a key) with a Leclanche cell. As a sample set of observations the
following relating to hydrogen at atmospheric pressure and at 58° F., which
had been purified by passage over fragments of sulphur and solid soda
(without phosphoric anhydride), may be given: —
* Fogg. Ann. 1875, Vol. CLV. p. 547.
22
338 ON THE VISCOSITY OF HYDROGEN AS AFFECTED BY MOISTURE. [234
ORSERVATIONS ON JUNE 7, 1897.
_
65-4
423-7
88-9
358-3
2-554
—
401-3
110-0
312-4
2-495
0-059
381-5
128-9
271-5
2-434
0-061
364-4
144-1
235-5
2-372
0-062
349-7
158-6
205-6
2-313
0-059
336-8
169-8
178-2
2-251
0-062
325-7
180-6
155-9
2-193
0-058
315-7
189-8
135-1
2-131
0-062
307-2
197-8
117-4
2-070
0-061
300-0
204-6 102-2
2-009
0-061
293-7
210-6
89-1
1-950
0-059
287-8
—
77-2
1-888
0-062
Meaft log. dec. =0-0604.
The two first columns contain the actually observed elongations upon the
two sides. They require no correction, since the scale was bent to a circular arc
centred at the mirror. The third column gives the actual arcs of vibration,
the fourth their (common) logarithms, and the fifth the differences of these,
which should be constant. The mean logarithmic decrement can be obtained
from the first and last arcs only, but the intermediate values are useful as a
check. The time of (complete) vibration was determined occasionally. It
was constant, whether hydrogen or air occupied the chamber, at 26'2 seconds.
The observations extended themselves over two months, and it would be
tedious to give the results in any detail. One of the points to which I
attached importance was a comparison between hydrogen as it issued from
the generator without any desiccation whatever and hydrogen carefully dried
by passage through a long tube packed with phosphoric anhydride. The
difference proved itself to be comparatively trifling. For the wet hydrogen
there were obtained on May 10, 11, such log. decs, as 0'0594, 0'0590, O0591,
or as a mean 0*0592. The dried hydrogen, on the other hand, gave 0'0588,
0'0586, 0'0584, 0'0590 on various repetitions with renewed supplies of gas,
or as a mean 0'0587, about 1 per cent, smaller than for the wet hydrogen.
It appeared that the dry hydrogen might stand for several days in the
viscosity chamber without alteration of logarithmic decrement. It should be
mentioned that the apparatus was set up underground, and that the changes
of temperature were usually small enough to be disregarded.
In the next experiments the phosphoric tube was replaced by others
containing sulphur (with the view of removing mercury vapour) and solid
soda. Numbers were obtained on different days such as 0'0591, 0-0586,
0-0588, 0-0587, mean 0'0588, showing that the desiccation by soda was practi-
cally as efficient as that by phosphoric anhydride.
1897] ON THE VISCOSITY OF HYDROGEN AS AFFECTED BY MOISTURE. 339
At this stage the apparatus was rearranged. As shown by observations
upon air (at 10 cm. residual pressure), the logarithmic decrements were
increased, probably owing to a slight displacement of the mirror relatively to
the containing walls of the chamber. The sulphur and soda tubes were
retained, but with the addition of one of hard glass containing turnings of
magnesium. Before the magnesium was heated the mean number for
hydrogen (always at atmospheric pressure) was 0*0600. The heating of the
magnesium to redness, which it was supposed might remove residual water,
had the effect of increasing the viscosity of the gas, especially at first*.
After a few operations the logarithmic decrement from gas which had passed
over the hot magnesium seemed to settle itself at 0*0606. When the
magnesium was allowed to remain cold, fresh fillings gave again 0'0602,
0-0601, 0-0598, mean O'OGOO. Dried air at 10cm. residual pressure gave
0-01114, 0-01122, 0-01118, 0-01126, 0-01120, mean 0-01120.
In the next experiments a phosphoric tube was added about 60 cm. long
and closely packed with fresh material. The viscosity appeared to be slightly
increased, but hardly more than would be accounted for by an accidental
rise of temperature. The mean unconnected number may be taken as 0*0603.
The evidence from these experiments tends to show that residual moisture
is without appreciable influence upon the viscosity of hydrogen ; so much so
that, were there no other evidence, this conclusion would appear to me to be
sufficiently established. It remains barely possible that the best desiccation
to which I could attain was still inadequate, and that absolutely dry hydrogen
would exhibit a less viscosity. It must be admitted that an apparatus
containing cemented joints and greased stop-cocks is in some respects at a
disadvantage. Moreover, it should be noticed that the ratio 0'0600 : 01120,
viz. 0*536, for the viscosities of hydrogen and air is decidedly higher than
that (0'500) deduced by Sir G. Stokes from Crookes's observations. Accord-
ing to the theory of the former, a fair comparison may be made by taking, as
above, the logarithmic decrements for hydrogen at atmospheric pressure, and
for air at a pressure of 10 cm. of mercury. I may mention that moderate
rarefactions, down say to a residual pressure of 5 cm., had no influence on the
logarithmic decrement observed with hydrogen.
I am not able to explain the discrepancy in the ratios thus exhibited.
A viscous quality in the suspension, leading to a subsidence of vibrations
independent of the gaseous atmosphere, would tend to diminish the apparent
differences between various kinds of gas, but I can hardly regard this cause
as operative in my experiments. For actual comparisons of widely differing
viscosities I should prefer an apparatus designed on Maxwell's principle, in
which the gas subjected to shearing should form a comparatively thin layer
bounded on one side by a moving plane and on the other by a fixed plane.
* The glass was somewhat attacked, and it is supposed that silicon compounds may have
contaminated the hydrogen.
22—2
235.
ON THE PROPAGATION OF WAVES ALONG CONNECTED
SYSTEMS OF SIMILAR BODIES.
[Philosophical Magazine, XLIV. pp. 356—362, 1897.]
FOR simplicity of conception the bodies are imagined to be similarly
disposed at equal intervals (a) along a straight line. The position of each
body, as displaced from equilibrium, is supposed to be given by one coor-
dinate, which for the rth body is denoted by -ty-r. A wave propagated in one
direction is represented by taking tyr proportional to ei(nt+r®. If we take an
instantaneous view of the system, the disturbance is periodic when rj3
increases by 2?r, or when ra increases by 2?ra/y8. This is the wave-length,
commonly denoted by X, ; so that, if k = 2-Tr/X,, k = /3/a. The velocity of
propagation (V) is given by V=nfk; and the principal object of the investi-
gation is to find the relation between n or V and X.
The forces acting upon each body, which determine the vibration of the
system about its configuration of equilibrium, are assumed to be due solely
to the neighbours situated. within a limited distance. The simplest case of
all is that in which there is no mutual reaction between the bodies, the kinetic
and potential energies of the system being then given by
T=\A£+*> P = i<702fr», .................. (1)
similarity requiring that the coefficients A0, C0 be the same for all values
of r. In this system each body vibrates independently, according to the
equation
and n* = C0/A0 .................................. (3)
The frequency is of course independent of the wave-length in which the
phases may be arranged to repeat themselves, so that n is independent of k,
while V equal to n/k varies inversely as k, or directly as X. The propagation
of waves along a system of this kind has been considered by Reynolds.
1897] ON WAVES ALONG SYSTEMS OF SIMILAR BODIES. 341
In the general problem the expression for P will include also products of
tyr with the neighbouring coordinates ...^r-2> tyr-i, tyr+i, tyr+z-', and a
similar statement holds good for T. Exhibiting only the terms which involve
r, we may write
- A2^r_2 - A2^r+2 -..., ....... (4)
P= ... + i<W- d Wv-i - £ Wr»
-<72^rfr_2-C2TMrr+2-..., ......... (5) '
where A1} A2, ... G1} C2) ... are constants, finite for a certain number of terms
and then vanishing. The equation for tyr is accordingly
A0$'r-Al-f>r-l - A^r+j. - A2fyr_2 - A^r+s - ......
+ C^r - C^M - C^r+l - C2+r_2 - C^r+2 - ...... = 0 ....... (6)
In the other equations of the system r is changed, but without entailing any
other alteration in (6). Since all the quantities i/r are proportional to eint,
the double differentiation is accounted for by the introduction of the factor
— n2. Making this substitution and remembering that tyr is also proportional
to eirP, we get as the equivalent of any one of the equations (6)
n? (A0 - A^-V - A^V - A2e~^ - A2e^ - ...)
= CQ - C^e-* - C.e* - C2e~w - C,e^ - ...,
2 _ C0 - 2^ cos ka - 2G2 cos 2feg - . . . ,,_,
~ 2o^"237cos ka - 1As cos 2ka - . . . ' '
in which ^ is replaced by its equivalent ka. By (7) n is determined as a
function of k and of the fundamental constants of the system.
In most of the examples which naturally suggest themselves A1} A2, ...
vanish, so that T has the same simple form as in (1). If we suppose for
brevity that A0 is unity, (7) becomes
n?= C0 - 20! cos ka - 2C2 cos 2ka - .................... (8)
When the waves are very long, k approximates to zero. In the limit
n2=C0-2Cl-2G2- ............................. (9)
If we call the limiting value C, we may write (8) in the form
sn
(10)
In an important class of cases C vanishes, that is the frequency diminishes
without limit as \ increases. If at the same time but one of the constants
G!, C2, ... be finite, the equation simplifies. For example, if Cl alone be
finite,
(11)
342 ON THE PROPAGATION OF WAVES ALONG [235
In any case when n is known V follows immediately. Thus from (10) with
C evanescent, we get
A. simple case included under (11) is that of a stretched string, itself
without mass, but carrying unit loads at equal intervals (a)*. The expression
for the potential energy is
T! representing the tension. Thus by comparison with (5)
ao = 22T1/a, Ci = Zya, 0, = 0, &c.;
so that by (8)
2rx 2T,
ri> = — - -- cos tea,
a a
IJL being introduced to represent the mass of each load with greater generality.
The value of V is obtained by division of (14) by k. In order more easily to
compare with a known formula we may introduce the longitudinal density p,
such that /i = ap. Thus
V=-= /(^} sin(^a> (15)
k V \P / %ka
reducing to the well-known value of the constant velocity of propagation
along a uniform string when a is made infinitesimal. Lord Kelvin's wave-
model (Popular Lectures and Addresses, Vol. I. 2nd ed. p. 164) is also included
under the class of systems for which P has the form (13).
Another example in which again C2, G3 . . . vanish is proposed by Fitzgerald-f*.
It consists of a linear system of rotating magnets (Fig. 1) with their poles
Fig. l.
close to one another and disturbed to an amount small compared with the
distance apart of the poles. The force of restitution is here proportional to
the sum of the angular displacements (-^) of contiguous magnets, so that P
is proportional to
* See Theory of Sound, §§ 120, 148.
t Brit. Assoc. Report, 1893, p. 689.
1897] CONNECTED SYSTEMS OF SIMILAR BODIES. 343
Here Ol = - £ <70, and (8) gives ?i2 =C0(l + cos ka\
or n = n0 . cos (Pa), ........................... (16)
if n0 represent the value of n appropriate to k — 0, i.e. to infinitely long waves.
Here n = 0, when \ = 2a. In this case
Fitzgerald considers, further, a more general linear system constructed by
connecting a series of equidistant wheels by means of indiarubber bands.
" By connecting the wheels each with its next neighbour we get the simplest
system. If to this be superposed a system of connexion of each with its
next neighbour but two, and so on. complex systems with very various
relations between wave-length and velocity can be constructed depending
on the relative strengths of the bands employed." If the bands may be
crossed, the potential energy takes the form
P = 7l r ± r-l)" + i7l (*r ± *r+l)*
r ± ^+2)2
which is only less general than (5) by the limitation
^±£±^±... = 0 ......................... (18)
Prof. Fitzgerald appears to limit himself to the lower sign in the alternatives,
so that C in (10) vanishes. This leads to (12), from which his result differs,
but probably only by a slip of the pen.
If we take the upper sign throughout, (8) becomes
- Jn« = 0icos'^ + a,cos* ^ + C,cos' ~ + .......... (19)
£ 2 . ~Z
It may be observed that Prof. Fitzgerald's system will have the most
general potential energy possible (5), if in addition to the elastic connexions
between the wheels there be introduced a force of restitution acting upon
each wheel independently.
As an example in which (72 is finite as well as C,, let us imagine a system
of masses of which each is connected to its immediate neighbours on the two
sides by an elastic rod capable of bending but without inertia. Here
P = . . . + *c (2^r_, - ^r_2 - ^ + i c (2f r - tr-i - ^+i)2
+ ic(2^r+1-tr-^+2)*+ ............. (20>
A comparison with (5) gives
00 = 6c, ai = 4c, Ca = -c,
so that 0=00-201-2C'2 = 0.
344 ON THE PROPAGATION OF WAVES ALONG [235
Accordingly by (10),
n2 = 16c sin3 (%ka) - 4c sin2 ka = 16c sin4 (%ka),
or n = 4c* sin2 ($ka) ............................ (21)
Thus far we have considered the propagation of waves along an unlimited
series of bodies. If we suppose that the total number is m and that they
form a closed chain, -^ must be such that
+r+m = +r, ................................. (22)
from which it follows that
@ = ka = 2s7rlm, .............................. (23)
s being an integer. Thus (8) becomes
n2 = (70 - 2 C1 cos ( 2s?r/w) - 2(72 cos (4s?r/m) - ....... ... (24)
When the chain, composed of a limited series of bodies, is open at the
ends instead of closed, the general problem becomes more complicated. A
simple example is that treated by I^agrange, of a stretched massless string,
carrying a finite number of loads and fixed at its extremities*. The open
chain of m magnets, for which
a + tm)2, ...... (25)
is considered by Fitzgerald. The equations are
^ (1 - n*) + >/r2 = 0,
=0,
of which the first and last may be brought under the same form as the others
if we introduce T/TO and ^m+i, such that
^o + ^i = 0, ^ + ^1 = 0 (27)
If we assume
^rr = cosnt sin (r/8 - £/8), (28)
the first of equations (27) is satisfied. The second is also satisfied provided
that
= 0, or @ = S7r/m (29)
* Theory of Sound, § 120.
1897] CONNECTED SYSTEMS OF SIMILAR BODIES. 345
The equations (26) are satisfied if
that is, if n = 2 cos (s7r/2m) ............................ (30)
In (29), (30) s may assume the ra values 1 to m inclusive. In the last case
n = 0, and £ = TT ; and from (28),
The equal amplitudes and opposite phases of consecutive coordinates, i.e.
angular displacements of the magnets, give rise to no potential energy, and
therefore to a zero frequency of vibration. In the first case (s = 1) the
angular deflexions are all in the same direction, and the frequency is the
highest admissible. If at the same time m be very great, n reaches its
maximum value, corresponding to parallel positions of all the magnets. If
we call this value N, the generalized form of (30), applicable to all masses
and degrees of magnetization, may be written
(31)
If m is great and s relatively small, (31) becomes approximately
so that as s diminishes we have a series of frequencies approaching N as an
upper limit, and are reminded (as Fitzgerald remarks) of certain groups of
spectrum lines. A nearer approach to the remarkable laws of Balmer for
hydrogen* and of Kayser and Runge for the alkalies is arrived at by
supposing s constant while m varies. In this case, instead of supposing that
the whole series of lines correspond to various modes of one highly compound
system, we attribute each line to a different system vibrating in a given
special mode. Apart from the better agreement of frequencies, this point of
view seems the more advantageous as we are spared the necessity of selecting
and justifying a special high value of m. If we were to take s = 2 in (31)
and attribute to m integral values 3, 4, 5, . . . , we should have a series of
frequencies of the same general character as the hydrogen series, but still
differing considerably in actual values.
There is one circumstance which suggests doubts whether the analogue
of radiating bodies is to be sought at all in ordinary mechanical or acoustical
systems vibrating about equilibrium. For the latter, even when gyratory
terms are admitted, give rise to equations involving the square of the
frequency; and it is only in certain exceptional cases, e.g. (31), that the
frequency itself can be simply expressed. On the other hand, the formulae
* Viz. n=tf (l-4m-2), with m=B, 4, 5, &c.
346 ON WAVES ALONG SYSTEMS OF SIMILAR BODIES. [235
and laws derived from observation of the spectrum appear to introduce more
naturally the first power of the frequency. For example, this is the case
with Balmer's formula. Again, when the spectrum of a body shows several
doublets, the intervals between the components correspond closely to a
constant difference of frequency, and could not be simply expressed in terms
of squares of frequency. Further, the remarkable law, discovered indepen-
dently by Rydberg and by Schuster, connecting the convergence frequencies of
different series belonging to the same substance, points in the same direction.
What particular conclusion follows from this consideration, even if force
be allowed to it, may be difficult to say. The occurrence of the first power
of the frequency seems suggestive rather of kinematic relations* than of those
of dynamics.
[1902. See further on the subject of the present paper, Phil. Mag. Dec.
1898, " On Iso-periodic Systems," Art. 242, below.]
* E.g. as in the phases of the moon.
236.
ON THE DENSITIES OF CARBONIC OXIDE, CARBONIC
ANHYDRIDE, AND NITROUS OXIDE.
[Proceedings of the Royal Society, LXII. pp. 204—209, 1897.]
THE observations here recorded were carried out by the method and with
the apparatus described in a former paper*, to which reference must be made
for details. It must suffice to say that the globe containing the gas to be
weighed was filled at 0° C., and to a pressure determined by a manometric
gauge. This pressure, nearly atmospheric, is slightly variable with tempera-
ture on account of the expansion of the mercury and iron involved. The
actually observed weights are corrected so as to correspond with a temperature
of 15° C. of the gauge, as well as for the errors in the platinum and brass
weights employed. In the present, as well as in the former, experiments I
have been ably assisted by Mr George Gordon.
Carbonic Oxide.
This gas was prepared by three methods. In the first method a flask,
sealed to the rest of the apparatus, was charged with 80 grams recrystallised
ferrocyanide of potassium and 360 c.c. strong sulphuric acid. The generation
of gas could be started by the application of heat, and with cares it could be
checked and finally stopped by the removal of the flame with subsequent
application, if necessary, of wet cotton-wool to the exterior of the flask. In
this way one charge could be utilised with great advantage for several fillings.
On leaving the flask the gas was passed through a bubbler containing potash
solution (convenient as allowing the rate of production to be more easily
estimated) and thence through tubes charged with fragments of potash and
phosphoric anhydride, all connected by sealing. When possible, the weight
* "On the Densities of the Principal Gases," Roy. Soc. Proc. Vol. LIII. p. 134, 1893.
[Vol. iv. p. 39.]
348 ON THE DENSITIES OF CARBONIC OXIDE, [236
of the globe full was compared with the mean of the preceding and following
weights empty. Four experiments were made with results agreeing to within
a few tenths of a milligram.
In the second set of experiments the flask was charged with 100 grams
of oxalic acid and 500 c.c. strong sulphuric acid. To absorb the large
quantity of CO2 simultaneously evolved, a plentiful supply of alkali was
required. A wash-bottle and a long nearly horizontal tube contained strong
alkaline solution, and these were followed by the tubes containing solid potash
and phosphoric anhydride as before.
For the experiments of the third set oxalic acid was replaced by formic,
which is more convenient as not entailing the absorption of large volumes of
C02. In this case the charge consisted of 50 grams formate of soda, 300 c.c.
strong sulphuric acid, and 150 c.c. distilled water. The water is necessary in
order to prevent action in the cold, and the amount requires to be somewhat
carefully adjusted. As purifiers, the long horizontal bubbler was retained
and the tubes charged with solid potash and phosphoric anhydride. In this
set there were four concordant experiments. The immediate results stand
thus :—
Carbonic Oxide.
From ferrocyanide 2*29843
„ oxalic acid 2*29852
formate of soda . 2*29854
Mean 2*29850
This corresponds to the number 2*62704 for oxygen*, and is subject to a
correction (additive) of 0*00056 for the diminution of the external volume of
the globe when exhausted.
The ratio of the densities of carbonic oxide and oxygen is thus
2*29906 : 2*62760 ;
so that if the density of oxygen be taken as 32, that of carbonic oxide will be
27*9989. If, as some preliminary experiments by Dr Scottf indicate, equal
volumes may be taken as accurately representative of CO and of 02, the
atomic weight of carbon will be 11*9989 on the scale of oxygen = 16.
The very close agreement between the weights of carbonic oxide prepared
in three different ways is some guarantee against the presence of an impurity
of widely differing density. On the other hand, some careful experiments
led Mr T. W. Richards J to the conclusion that carbonic oxide is liable to
* "On the Densities of the Principal Gases," Roy. Soc. Proc. Vol. LIII. p. 144, 1893.
[Vol. rv. p. 39.]
t Camb. Phil. Proc. Vol. ix. p. 144, 1896.
J Amer. Acad. Proc. Vol. xvin. p. 279, 1891.
1897] CARBONIC ANHYDRIDE, AND NITROUS OXIDE. 349
contain considerable quantities of hydrogen or of hydrocarbons. From
5£ litres of carbonic oxide passed over hot cupric oxide he collected no less
than 25 milligrams of water, and the evidence appeared to prove that the
hydrogen was really derived from the carbonic oxide. Such a proportion of
hydrogen would entail a deficiency in the weight of the globe of about 11
milligrams, and seems improbable in view of the good agreement of the
numbers recorded. The presence of so much hydrogen in carbonic oxide is
also difficult to reconcile with the well-known experiments of Professor Dixon,
who found that prolonged treatment with phosphoric anhydride was required
in order to render the mixture of carbonic oxide and oxygen inexplosive. In
the presence of relatively large quantities of free hydrogen (or hydrocarbons)
why should traces of water vapour be so important ?
In an experiment by Dr Scott*, 4 litres of carbon monoxide gave only
1*3 milligrams to the drying tube after oxidation.
I have myself made several trials of the same sort with gas prepared from
formate of soda exactly as for weighing. The results were not so concordant
as I had hoped -f, but the amount of water collected was even less than that
given by Dr Scott. Indeed, I do not regard as proved the presence of
hydrogen at all in the gas that I have employed J.
Carbonic A nhydride.
This gas was prepared from hydrochloric acid and marble, and after
passing a bubbler charged with a solution of carbonate of soda, was dried by
phosphoric anhydride. Previous to use, the acid was caused to boil for some
time by the passage of hydrochloric acid vapour from a flask containing
another charge of the acid. In a second set of experiments the marble was
replaced by a solution of carbonate of soda. There is no appreciable
difference between the results obtained in the two ways; and the mean,
corrected for the errors of weights and for the shrinkage of the globe when
exhausted, is 3'6349, corresponding to 2'6276 for oxygen. The temperature
at which the globe was charged was 0° C., and the actual pressure that of the
manometric gauge at about 20°, reduction being made to 15° by the use of
Boyle's law. From the former paper it appears that the actual height of the
mercury column at 15° is 762*511 mm.
* Chem. Soc. Trans. 1897, p. 564.
t One obstacle was the difficulty of re-oxidising the copper reduced by carbonic oxide. I have
never encountered this difficulty after reduction by hydrogen.
$ In Mr Richards' work the gas in an imperfectly dried condition was treated with hot
platinum black. Is it possible that the hydrogen was introduced at this stage?
350 ON THE DENSITIES OF CARBONIC OXIDE, [236
Nitrous Oxide.
In preliminary experiments the gas was prepared in the laboratory, at as
low a temperature as possible, from nitrate of ammonia, or was drawn from
the iron bottles in which it is commercially supplied. The purification was
by passage over potash and phosphoric anhydride. Unless special precautions
are taken the gas so obtained is ten or more milligrams too light, presumably
from admixture with nitrogen. In the case of the commercial supply, a better
result is obtained by placing the bottles in an inverted position so as to draw
from the liquid rather than from the gaseous portion.
Higher and more consistent results were arrived at from gas which had
been specially treated. In consequence of the high relative solubility of
nitrous oxide in water, the gas held in solution after prolonged agitation, of
the liquid with impure gas from any supply, will contain a much diminished
proportion of nitrogen. To carry out this method on the scale required, a
large (11 -litre) flask was mounted on an apparatus in connexion with the
lathe so that it could be vigorously shaken. After the dissolved air had been
sufficiently expelled by preliminary passage of N2O, the water was cooled to
near 0° C. and violently shaken for a considerable time while the gas was
passing in large excess. The nitrous oxide thus purified was expelled from
solution by heat, and was used to fill the globe in the usual manner.
For comparison with the results so obtained, gas purified in another
manner was also examined. A small iron bottle, fully charged with the com-
mercial material, was cooled in salt and ice and allowed somewhat suddenly
to blow off half its contents. The residue drawn from the bottle in one or
other position was employed for the weighings.
Nitrous Oxide (1896).
Aug. 15 Expelled from water 3'6359
,,17 „ „ 3-6354
„ 19 From residue after blow off, valve downwards 3'6364
„ 21 „ „ valve upwards . 3'6358
„ 22 „ „ valve downwards 3'6360
Mean 3'6359
The mean value may be taken to represent the corrected weight of the gas
which fills the globe at 0° C. and at the pressure of the gauge (at 15°), corre-
sponding to 2'6276 for oxygen.
One of the objects which I had in view in determining the density of
nitrous oxide was to obtain, if it were possible, evidence as to the atomic
weight of nitrogen. It may be remembered that observations upon the
1897] CARBONIC ANHYDRIDE, AND NITROUS OXIDE. 351
density of pure nitrogen, as distinguished from the atmospheric mixture
containing argon which, until recently, had been confounded with pure
nitrogen, led* to the conclusion that the densities of oxygen and nitrogen
were as 16 : 14'003, thus suggesting that the atomic weight of nitrogen might
really be 14 in place of 14'05, as generally received. The chemical evidence
upon which the latter number rests is very indirect, and it appeared that a
direct comparison of the weight of nitrous oxide and of its contained nitrogen
might be of value. A suitable vessel would be filled, under known conditions,
with the nitrous oxide, which would then be submitted to the action off a
spiral of copper or iron wire rendered incandescent by an electric current.
When all the oxygen was removed, the residual nitrogen would be measured,
from which the ratio of equivalents could readily be deduced. The fact that
the residual nitrogen would possess nearly the same volume as the nitrous
oxide from which it was derived would present certain experimental advan-
tages. If indeed the atomic weights were really as 14 : 16, the ratio (*•) of
volumes, after and before operations, would be given by
2-2996 xx 14
7 x 3-6359
Whence ' ° 11 x 2-2996 -1"0061'
3-6359 and 2*2996 being the relative weights of nitrous oxide and of
nitrogen which (at 0° C. and at the pressure of the gauge) occupy the same
volume. The integral numbers for the atomic weights would thus correspond
to an expansion, after chemical reduction, of about one-half per cent.
But in practical operation the method lost most of its apparent simplicity.
It was found that copper became unmanageable at a temperature sufficiently
high for the purpose, and recourse was had to iron. Coils of iron suitably
prepared and supported could be adequately heated by the current from a
dynamo without twisting hopelessly out of shape ; but the use of iron leads
to fresh difficulties. The emission of carbonic oxide from the iron heated in
vacuum continues for a very long time, and the attempt to get rid of this gas
by preliminary treatment had to be abandoned. By final addition of a small
quantity of oxygen (obtained by heating some permanganate of potash sealed
up in one of the leading tubes) the CO could be oxidised to CO2, and thus,
along with any H20, be absorbed by a lump of potash placed beforehand in
the working vessel. To get rid of superfluous oxygen, a coil of incandescent
copper had then to be invoked, and thus the apparatus became rather
complicated.
It is believed that the difficulties thus far mentioned were overcome, but
nevertheless a satisfactory concordance in the final numbers was not attained.
* Bayleigh and Ramsay, Phil. Trans. Vol. CLXXXVI. p. 190, 1895. [Vol. iv. p. 133.]
352 ON THE DENSITIES OF CARBONIC OXIDE, ETC. [236
In the present position of the question no results are of value which do not
discriminate with certainty between 14'05 and 14*00. The obstacle appeared
to lie in a tendency of the nitrogen to pass to higher degrees of oxidation.
On more than one occasion mercury (which formed the movable boundary of
an overflow chamber) was observed to be attacked. Under these circum-
stances I do not think it worth while to enter into further detail regarding
the experiments in question.
The following summary gives the densities of the various gases relatively
to air, all obtained by the same apparatus*. The last figure is of little
significance.
Air free from H20 and C02 . . . . . 1 '00000
Oxygen 110535
Nitrogen and argon (atmospheric) . . . 0'97209
Nitrogen 0'96737
Argon 1-37752
Carbonic oxide 0'96716
Carbonic anhydride 1*52909
Nitrous oxide 1-52951
The value obtained for hydrogen upon the same scale was 0'06960 ; but
the researches of M. Leduc and of Professor Morley appear to show that this
number is a little too high.
[1902. For the absolute densities of air and oxygen, see Vol. IV. p. 51.]
* Boy. Soc. Proc. Vol. LHI. p. 148, 1893 ; Vol. LV. p. 340, 1894 ; Phil. Trans. Vol. CLXXXVI.
p. 189, 1895 ; Roy. Soc. Proc. Vol. LIX. p. 201, 1896. [Vol. iv. pp. 52, 104, 130, 215.]
237.
RONTGEN RAYS AND ORDINARY LIGHT.
[Nature, LVII. p. 607, 1898.]
r
ACCORDING to the theory of the Rontgen rays suggested by Sir G. Stokes*,
and recently developed by Prof. J. J. Thomson f, their origin is to be sought
in impacts of the charged atoms constituting the kathode-stream, whereby
pulses of disturbance are generated in the ether. This theory has certainly
much to recommend it ; but I cannot see that it carries with it some of the
consequences which have been deduced as to the distinction between Rontgen
rays and ordinary luminous and non-luminous radiation. The conclusion of
the authors above mentioned]:, " that the Rontgen rays are not waves of very
short wave-length, but impulses," surprises me. From the fact of their being
highly condensed impulses, I should conclude on the contrary that they are
waves of short wave-length. If short waves are inadmissible, longer waves
are still more inadmissible. What then becomes of Fourier's theorem and
its assertion that any disturbance may be analysed into regular waves ?
Is it contended that previous to resolution (whether merely theoretical,
or practically effected by the spectroscope) the vibrations of ordinary
(e.g. white) light are regular, and thus distinguished from disturbances made
up of impulses ? This view was certainly supported in the past by high
authorities, but it has been shown to be untenable by Gouy§, Schuster ||, and
the present writer 1T. A curve representative of white light, if it were drawn
upon paper, would show no sequences of similar waves.
In the second of the papers referred to, I endeavoured to show in detail
that white light might be supposed to have the very constitution now
ascribed to the Rontgen radiation, except that of course the impulses would
have to be less condensed. The peculiar behaviour of the Rontgen radiation
with respect to diffraction and refraction would thus be attributable merely
to the extreme shortness of the waves composing it.
[1902. In a reply to the above (Nature, LVIII. p. 8), Prof. Thomson
expresses the opinion that "the difference between us is one of terminology."]
* Manchester Memoirs, Vol. XLI. No. 15, 1897.
t Phil. Mag. Vol. XLV. p. 172, 1898.
J See also Prof. S. P. Thompson's Light Visible and Invisible (London, 1897), p. 273.
§ Journ. de Physique, 1886, p. 354.
|| Phil. Mag. Vol. xxxvn. p. 509, 1894.
IT Enc. Brit. " Wave Theory," 1888. [Vol. in. p. 60.] Phil. Mag. Vol. xxvn. p. 461, 1889.
[Vol. HI. p. 270.]
R. iv. 23
238.
NOTE ON THE PRESSURE OF RADIATION, SHOWING AN
APPARENT FAILURE OF THE USUAL ELECTROMAGNETIC
EQUATIONS.
[Philosophical Magazine, XLV. pp. 522—525, 1898.]
FOLLOWING a suggestion of Bartoli, Boltzmann* and W. Wienf have
arrived at the remarkable conclusion that that part of the energy of radiation
from a black body at absolute temperature 6, which lies between wave-lengths
X, and X + d\, has the expression
e^(6\)d\ (1)
where <£ is an arbitrary function of the single variable 0\. The law of
Stefan, according to which the total radiation is as 0*, is therein included.
The argument employed by these authors is very ingenious, and I think
convincing when the postulates are once admitted. The most important of
them relates to the pressure of radiation, supposed to be operative upon the
walls within which the radiation is confined, and estimated at one-third of
the density of the energy in the case when the radiation is alike in all
directions. The argument by which Maxwell originally deduced the pressure
of radiation not being clear to me, I was led to look into the question a
little more closely, with the result that certain discrepancies have presented
themselves which I desire to lay before those who have made a special study
of the electric equations. The criticism which appears to be called for extends
indeed much beyond the occasion which gave rise to it.
A straightforward calculation of the pressure exercised by plane electric
waves incident perpendicularly upon a metallic reflector is given by Prof.
J. J. Thomson :[. The face of the reflector coincides with x = 0, and in the
vibrations under consideration the magnetic force reduces itself to the com-
ponent (/3) parallel to y, and the current to the component (w) parallel to z.
The waves which penetrate the conducting mass die out more or less quickly
according to the conductivity. If the conductivity is great, most of the
energy is reflected, and such part as is propagated into the conductor is
limited to a thin skin at x = 0. According to the usual equations the
* Wied. Ann. Vol. xxn. pp. 31, 291 (1884).
t Berlin. Sitzungsber. Feb. 1893.
J Elements of Electricity and Magnetism, Cambridge, 1895, § 241.
1898] ON THE PRESSURE OF RADIATION. 355
mechanical force exercised upon unit of area of the slice dx of the conductor
is — wbdx, or altogether
I wbdx .................................. (2)
Here b denotes the magnetic induction, and is equal to //,/3, if //, be the per-
meability and /8 the magnetic force. Now
4>7rw = d(3/dx,
so that the integral becomes
where /30 is the value of /3 within the conductor at x = 0, and ftx = 0, if the
conducting slab be sufficiently thick. Since there is no discontinuity of
magnetic force at x = 0, /30 may be taken also to refer to the value at x = 0
just outside the metallic surface.
The expression (3) gives the force at any moment ; but we are concerned
only with the mean value. Since the mean value of ft? is one-half the maxi-
mum value, we have for the pressure
It only remains to compare with the density of the energy outside the
metal, and we may limit ourselves to the case of complete reflexion. The
constant energy of the stationary waves passes alternately between the electric
and magnetic forms. If we estimate it at the moment of maximum magnetic
force, we have
energy = ^jfjpdxdyde ......................... (5)
In (5) ft is variable with x. If j9max. denote the maximum value which
occurs at x = 0, the mean of /32 = l/S^x. Thus
density of energy = 2J53B1 . ^_ ^ ............... (6)
Thus, if the permeability /ju of the metal be unity, (4) and (6) coincide ;
and we conclude that in this case the pressure is equal to the density of the
energy in the neighbourhood of the metal. This is Maxwell's result. When
we consider radiation in all directions, the pressure is expressed as one-third
of the density of energy.
The difficulty that I have to raise relates to the case where //, is not equal
to unity. The conclusion in (4) that the pressure is proportional to p would
make havoc of the theory of Boltzmann and Wien and must, I think, be
rejected. So long as the reflexion is complete — and it may be complete
independently of /* — the radiation is similarly influenced, and (one would
suppose) must exercise a similar force upon the reflector. But if the con-
23—2
356 ON THE PRESSURE OF RADIATION. [238
elusion is impossible, where is the flaw in the process by which it is arrived
at ? Being unable to find any fault with the deduction above given (after
Prof. J. J. Thomson), I was led to scrutinize more closely the fundamental
equation itself; and I will now explain why it appears to me to be incorrect.
For this purpose let us apply it to the very simple case of a wire of
circular section, parallel to z, moving in the direction of x across an originally
uniform magnetic field (yS). The uniformity of the field is disturbed in two
ways : (i) by the operation of the current (w) flowing in the various filaments
of the wire, and (ii) independently of a current, by the magnetic effect of the
material composing the wire whose permeability (fi) is supposed to be great.
In estimating as in (2) the mechanical force parallel to x operative upon the
wire, we should have to integrate wb over the cross-section. In this w is
supposed to be constant, and the local value is everywhere to be attribed to b.
We may indeed, if we please, omit from b the part due to the currents in the
wire, which will in the end contribute nothing to the result ; but we are
directed to use the actual value of 6 as disturbed by the presence of the
magnetic material. In the particular case supposed, where fj, is great, the
value of b within the wire is uniform, and just twice as great as at a distance.
It follows, when the integration is effected, that the force parallel to x acting
upon the wire is greater (in the particular case doubly greater) than it would
be if the value of /* were unity.
But this conclusion cannot be accepted. The force depends upon the
number of lines of force to be crossed when the wire makes a movement
parallel to x. And it is clear that the lines effectively crossed in such a
movement are not the condensed lines due to the magnetic quality of the
wire, but are to be reckoned from the intensity of the undisturbed field. The
mechanical force cannot really depend upon p,, and the formula which leads to
such a result must be erroneous.
As regards the problem of the pressure of radiation, I conclude that in
this case also, and in spite of the formula, the permeability of the reflector is
without effect, and that the consequences deduced by Boltzmann and Wien
remain undisturbed.
Another investigation to which perhaps similar considerations will apply
is that of the mechanical force between parallel slabs conveying rapidly
alternating electric currents. Prof. J. J. Thomson's conclusion* is that the
electromagnetic repulsion is p times the electrostatic attraction, so that a
balance will occur only when p, = 1. It seems more probable that the factor
p, should be omitted, and that balance between the two kinds of force is
realized in every case.
[1902. See Phil. Mag. XLVL p. 154, 1898, where Prof. J. J. Thomson
returns to the consideration of the question above raised.]
* Recent Researches in Electricity and Magnetism, 1893, § 277.
239.
SOME EXPERIMENTS WITH THE TELEPHONE.
[Roy. Inst. Proc. xv. pp. 786—789, 1898; Nature, LVIII.
pp. 429—430, 1898.]
EARLY estimates of the minimum current of suitable frequency audible
in the telephone having led to results difficult of reconciliation with the
theory of the instrument, experiments were undertaken to clear up the
question. The currents were induced in a coil of known construction, either
by a revolving magnet of known magnetic moment, or by a magnetised
tuning-fork vibrating through a measured arc. The connexion with the
telephone was completed through a resistance which was gradually increased
until the residual current was but just easily audible. For a frequency of 512
the current was found to be 7 x 10~8 amperes*. This is a much less degree
of sensitiveness than was claimed by the earlier observers, but it is more in
harmony with what might be expected upon theoretical grounds.
In order to illustrate before an audience these and other experiments
requiring the use of a telephone, a combination of that instrument with a
sensitive flame was introduced. The gas, at a pressure less than that of the
ordinary supply, issues from a pin-hole burner^ into a cavity from which air
is excluded (see figure). Above the cavity, and immediately over the burner,
is mounted a brass tube, somewhat contracted at the top where ignition first
occurs J. In this arrangement the flame is in strictness only an indicator,
the really sensitive organ being the jet of gas moving within the cavity and
surrounded by a similar atmosphere. When the pressure is not too high,
and the jet is protected from sound, the flame is rather tall and burns bluish.
Under the influence of sound of suitable pitch the jet is dispersed. At
first the flame falls, becoming for a moment almost invisible ; afterwards
it assumes a more smoky and luminous appearance, easily distinguishable
from the unexcited flame.
When the sounds to be observed come through the air, they find access
by a diaphragm of tissue paper with which the cavity is faced. This
serves to admit vibration while sufficiently excluding air. To get the best
results the gas pressure must be steady, and be carefully adjusted to the
maximum (about 1 inch) at which the flame remains undisturbed. A hiss
* The details are given in Phil. Mag. Vol. xxxvm. p. 285 (1894). [Vol. iv. p. 109.]
f The diameter of the pin-hole may be 0-03". [inch = 2-54 cm.]
t Camb. Proc. Vol. iv. p. 17, 1880. [Vol, i. p. 500.]
358
SOME EXPERIMENTS WITH THE TELEPHONE.
[239
from the mouth then brings about the transformation, while a clap of the
hands or the sudden crackling of a
piece of paper often causes extinction,
especially soon after the flame has
been lighted.
When the vibrations to be indicated
are electrical, the telephone takes the
place of the disc of tissue paper, and it
is advantageous to lead a short tube
from the aperture of the telephone into
closer proximity with the burner. The
earlier trials of the combination were
comparative failures, from a cause that
could not at first be traced. As applied,
for instance, to a Hughes' induction
balance, the apparatus failed to indicate
with certainty the introduction of a
shilling into one of the cups, and the
performance, such as it was, seemed to
deteriorate after a few minutes' experi-
menting. At this stage an observation
was made which ultimately afforded a
clue to the anomalous behaviour. It
was found that the telephone became
dewed. At first it seemed incredible
that this could come from the water of
combustion, seeing that the lowest part
of the flame was many inches higher.
But desiccation of the gas on its way
to the nozzle was no remedy, and it
was soon afterwards observed that no
dewing ensued if the flame were all
the while under excitation, either from
excess of pressure or from the action
of sound. The dewing was thus con-
nected with the unexcited condition.
Eventually it appeared that the flame
in this condition, though apparently
filling up the aperture from which it
issues, was nevertheless surrounded by
a descending current of air carrying
with it part of the moisture of combus-
tion. The deposition of dew upon the nozzle was thus presumably the source
1898] SOME EXPERIMENTS WITH THE TELEPHONE. 359
of the trouble, and a remedy was found in keeping the nozzle warm by
means of a stout copper wire (not shown) conducting the heat downwards
from the hot tube above.
The existence of the downward current could be made evident to private
observation in various ways, perhaps most easily by projecting little scraps
of tinder into the flame, whereupon bright sparks were seen to pass rapidly
downwards. In this form the experiment could not be shown to an audience,
but the matter was illustrated with the aid of a very delicate ether mano-
meter devised by Professor Dewar. This was connected with the upper part
of the brass tube by means of a small lateral perforation just below the root
of the flame. The influence of sound and consequent passage of the flame
from the unexcited to the excited condition was readily shown by the mano-
meter, the pressure indicated being less in the former state of things.
The downward current is evidently closely associated with the change of
appearance presented by the flame. In the excited state the gas issues
at the large aperture above as from a reservoir at very low pressure. The
unexcited flame rises higher, and must issue at a greater speed, carrying with
it not only the material supplied from the nozzle, and constituting the
original jet, but also some of the gaseous atmosphere in the cavity surround-
ing it. The downward draught thus appears necessary in order to equalise
the total issue from the upper aperture in the two cases.
Although the flame falls behind the ear in delicacy, the combination
is sufficiently sensitive to allow of the exhibition of a great variety of in-
teresting experiments. In the lecture the introduction of a threepenny
piece into one of the cups of a Hughes' induction balance was made evident,
the source of current being three Leclanche cells, and the interrupter being
of the scraping contact type actuated by clockwork.
Among other experiments was shown one to prove that in certain cases
the parts into which a rapidly alternating electric current is divided may
be greater than the whole*. The divided circuit was formed from the three
wires with which, side by side, a large flat coil is wound. One branch is
formed by two of these wires connected in series, the other (in parallel with
the first), by the third wire. Steady currents would traverse all three wires
in the same direction. But the rapidly periodic currents from the interrupter
distribute themselves so as to make the self-induction, and consequently the
magnetic field, a minimum ; and this is effected by the assumption of
opposite values in the two branches, the ratio. of currents being as 2:— 1.
On the same scale the total or main current is + 1. It was shown by means
of the telephone and flame that the current in one branch was about the
same (arithmetically) as in the main, and that the current in the other
branch was much greater.
* See Phil. Mag. Vol. xxn. p, 496 (1886). [Vol. n. p. 575.]
240.
LIQUID AIR AT ONE OPERATION.
[Nature, LVIII. p. 199, 1898.]
IT is to be hoped that personal matters will not divert attention from the
very interesting scientific questions involved. The liquefaction of air at one
operation by Linde and Hampson is indeed a great feat, and a triumph for
the principle of regeneration. But it must not be overlooked that to allow
the air to expand without doing work, or rather to allow the work of ex-
pansion to appear as heat at the very place where the utmost cooling is
desired, is very bad thermodynamics. The work of expansion should not be
dissipated within, but be conducted to the exterior.
I understand that attempts to expand the air under a piston in a cylinder
have led to practical difficulties connected with the low temperature. But
surely a turbine of some sort might be made to work. This would occupy
little space, and even if of low efficiency, would still allow a considerable
fraction of the work of expansion to be conveyed away. The worst turbine
would be better than none, and would probably allow the pressures to be
reduced. It should be understood that the object is not so much to save the
work, as to obviate the very prejudicial heating arising from its dissipation
in the coldest part of the apparatus. It seems to me that the future may
bring great developments in this direction, and that it may thus be possible
to liquefy even hydrogen at one operation.
241.
ON THE CHARACTER OF THE IMPURITY FOUND IN NITRO-
GEN GAS DERIVED FROM UREA [WITH AN APPENDIX
CONTAINING DETAILS OF REFRACTOMETER].
[Proceedings of the Royal Society, LXIV. pp. 95—100, 1898.]
IT has already* been recorded that nitrogen, prepared from urea by
the action of sodium hypobromite or hypochlorite, is contaminated with
an impurity heavier than nitrogen. The weight of pure nitrogen in the
globe employed being 2-299 grams, the gas obtained with hypochlorite was
36 milligrams, or about 1^ per cent., heavier. " A test with alkaline pyro-
gallate appeared to prove the absence from this gas of free oxygen, and only
a trace of carbon could be detected when a considerable quantity of the gas
was passed over red-hot cupric oxide into solution of baryta." Most gases
heavier than nitrogen are excluded from consideration by the thorough treat-
ment with alkali to which the material in question is subjected. In view of
the large amount of the impurity, and of the fact that it was removed by
passage over red-hot iron, I inclined to identify it with nitrous oxide ; but it
appeared that there were strong chemical objections to this explanation, and
so the matter was left open at that time. This summer I have returned to
it ; and although it is difficult to establish by direct evidence the presence of
nitrous oxide, I think there can remain little doubt that this is the true
explanation of the anomaly. I need scarcely say that there is here no
question of argon beyond the minute traces that might be dissolved in the
liquids employed.
In the present experiments hypochlorite has been employed, and the
procedure has been the same as before. The generating bottle, previously
exhausted, is first charged with the full quantity of hypochlorite solution, and
the urea is subsequently fed in by degrees. The gas passes in succession
over cold copper turnings, solid caustic soda, and phosphoric anhydride. In
various experiments the excess of weight was found to be variable, from 23
to 36 milligrams. In order to identify the impurity it was desirable to have
* Eayleigh and Ramsay, Phil. Trans., A (1895), p. 188. [Vol. iv. p. 131.]
362 ON THE CHARACTER OF THE IMPURITY FOUND IN [241
as much of it as possible, and experiments were undertaken to find out the
conditions of maximum weight. A change of procedure to one in which the
urea was first introduced, so that the hypochlorite would always be on the
point of exhaustion, led in the wrong direction, giving an excess of but
7 milligrams. Determinations of refractivity by the apparatus *, which uses
only 12 c.c. of gas, allowed the substitution of a miniature generating vessel,
and showed that the refractivity (and along with it the density) was increased
by a previous heating of the hypochlorite to about 140° F. [60° C.]. Acting
upon this information, arrangements were made for a preliminary heating
of the large generating vessel and its charge, with the result that the
excess of weight was raised to 55 milligrams, or about 2£ per cent, of the
whole. In any case heat is developed during the reaction, and the heavier
weights of some of the earlier trials probably resulted from a more rapid
generation of gas.
In seeking to obtain evidence as to the nature of the impurity, the most
important question is as to the presence or the absence of carbon. The
former experiment has been more than once repeated, with the result that
the baryta showed a slight clouding. Parallel experiments, in which C02 was
purposely introduced, indicated that the whole carbon in a charge of gas
weighing 30 milligrams in excess was about 1 milligram. It is possible
(though scarcely, I think, probable) that this carbon is not to be attributed
to the gas at all, and in any case the amount appears to be too small to afford
an explanation of the 30 milligrams excess of weight. If carbon be excluded,
the range for conjecture is much narrowed. As to oxygen, only traces were
found in most of the samples examined, whereas enormous quantities would
be needed to explain the excessive weight. It should be noted, however, that
the extra heavy sample, showing 55 milligrams excess, gave evidence of con-
taining a more appreciable quantity of oxygen.
It seems difficult to suggest any other impurity than nitrous oxide which
could account for the anomalous weight. Unfortunately there is no direct
test for nitrous oxide, but so far as the examination has been carried, the
behaviour of the gas is consistent with the view that this is the principal
impurity. The gas as collected has no smell. The proportion of nitrous
oxide indicated by the refractometer is nearly the same as that deduced from
the weight. For example, the refractivity was observed of some of the gas
which weighed 55 milligrams in excess. The proportion by volume (a?) of
NaO in the whole required to explain the excess of weight is given by
22 2-299 + 0-055
*X14 + 1-*= 2-299 '
whence x = 0'042.
* Roy. Soc. Proc., Vol. LIX. p. 201, 1896 [Vol. iv. p. 218]; Vol. LX. p. 56, 1896 [Vol. iv.
p. 225]. See also Appendix.
1898] NITROGEN GAS DERIVED FROM UREA. 363
The refractivity (referred to air as unity) of the same gas was deter-
mined by two independent sets of observations as T047, 1*048; mean,
T0475. If we assume that there are only nitrogen and nitrous oxide present,
the proportion (x) of the latter can be deduced from the known refrac-
tivities (/A — 1) of nitrous oxide, nitrogen, and air, which are respectively
0-0005159, 0-0002977, 0'0002927, the number for air being less than for
nitrogen. Thus,
x x 5159 + (l - a?) x 2977 = 1-0475 x 2927,
giving x = 0-0408.
The slight want of agreement can be explained by the presence of a
little oxygen, the recognition of which would lead to a rise in the second
value of x, and a fall in the first. Examination of the gas from the refracto-
meter with alkaline pyrogallate proved that oxygen was actually present.
Evidence may also be obtained by exploding the gas with excess of
hydrogen for which purpose oxy-hydrogen gas must be added. But when
nitrous oxide is in question, operations over water are useless, while for the
more exact procedure with mercury, experience and appliances were somewhat
deficient. The contraction observed was rather in excess of the volume of
nitrous oxide supposed to be present, but of this a good part is readily explained
by a small proportion of free oxygen.
If the impurity is really nitrous oxide, it should admit of concentration
by solution in water. To test this, about 1 litre of water (cooled with ice)
was shaken with the contents of a globe (about 2 litres). The dissolved
gases were then expelled by boiling, and were collected over water rendered
alkaline, in order to guard against the introduction of C02. The quantity
was, of course, too small for weighing, but it could readily be examined in
the refractometer. Of one sample, after desiccation, the refractivity rela-
tively to air was found to be as high as 1*207, although some air was known
to have entered accidentally. The proportion of nitrous oxide in a mixture
with nitrogen which would have this refractivity is 0*255. The impurity thus
agrees with nitrous oxide in being very much more soluble in water than are
the gases of the atmosphere.
In the analytical use of hypobromite for the determination of urea, it
has been noticed* that the nitrogen collected is deficient by about 8 per cent.,
but the matter does not appear to have been further examined. The
deficiency might be attributed to a part of the urea remaining undecomposed,
but more probably to oxidation of nitrogen. In default of analysis any
nitrogen collected as nitrous oxide would not appear anomalous, and the
explanation suggested requires the formation in addition of higher oxides
retained by the alkali.
* Russell and West, Chem. Soc. Journ., Vol. xn. p. 749, 1874.
364 ON THE CHARACTER OF THE IMPURITY FOUND IN [241
There is reason to suspect that nitrogen prepared by the action of chlorine
upon ammonia is also contaminated with nitrous oxide, and this is a matter
of interest, for the contamination in this case cannot well be referred to a
carbon compound. In two trials with distinct samples the refractivities were
decidedly in excess of that of pure nitrogen.
APPENDIX.
Details of Refractometer.
Determinations of refractivity have proved so useful and can be made so
readily and upon such small quantities of gas, that it may be desirable to
give further details of the apparatus employed, referring for explanation of
the principles involved to the former communication already cited.
The optical parts, other than the tubes containing the gases, are mounted
independently of everything else upon a bar of T-iron 90 cm. in length over
all. The telescopes are cheap instruments, of about 3 cm. aperture and
30 cm. focus, from which the eye-pieces are removed. At one end of the
T-iron and in the focus of the collimating telescope the original slit is fixed.
This requires to be rather narrow, and was made by scraping a fine line
upon a piece of silvered glass. At the further end the object-glass of the
observing telescope carries two slits which give passage to the interfering
pencils, and are situated opposite to the axes of the tubes holding the gases.
The sole eye-piece is a short length of glass rod — the same as formerly
described — of about 4 mm. diameter, which serves as horizontal magnifier.
The gas tubes are of brass, about 20 cm. long and 6 mm. in bore. These are
soldered together side by side and are closed at the ends by plates of worked
glass, so cemented as to obstruct as little as possible the passage of light
immediately over the tubes. There are two systems of bands, one formed by
light which has traversed the gases within the tubes, the other by light
which passes independently above; and an observation consists in so adjusting
the pressures within the tubes that the two systems fit one another. Unless
some further provision be made, there is necessarily a dark interval between
the two systems of bands corresponding to the thickness of the walls of the
tubes and any projecting cement. It is, perhaps, an improvement to bring
the two sets of bands into closer juxtaposition. The
interval can be abolished with the aid of a bi-plate
[see figure], formed of worked glass 4 or 5 mm. thick*.
This is placed immediately in front of the object-glass
of the observing telescope, the plane of junction of the
two glasses being horizontal and at the level of the
obstacles which are to be blotted out of the field of view.
* Compare Mascart, Traite d'Optique, Vol. i. p. 495, 1889.
1898] NITROGEN GAS DERIVED FROM UREA. 365
The objects sought in the design of the remainder of the apparatus
were (i) the use of a minimum of gas, and (ii) independence of other
pumping appliances. To this end the glass tubes associated with each
optical tube were arranged so as to serve both as manometer tubes and
as a sort of Geissler pump. The two halves of the apparatus being inde-
pendent and similar, it will suffice to speak of that which contained the gas
to be investigated. The tubes in which the levels of mercury are observed
are about 1 cm. in diameter. The fixed one, corresponding to the " pump-
head" of a Geissler or Topler, is 33 cm. in length, and is surmounted by a'
three-way tap, allowing it to be placed in communication either with the
optical tube or with one of narrow bore ending in a U, drowned in a deep
mercury trough. The bottom of the fixed tube, prolonged by 92 cm. of
narrower bore, is connected through a hose of black rubber with the movable
manometer tube. The latter is 70 cm. long and of one bore (1 cm.) through-
out. It can either be held in the hand or placed in a groove (parallel to the
fixed tube) along which it can slide. The four columns of mercury stand
side by side, and the levels are referred by a cathetometer to a metre scale
which occupies the central position. It is not proposed to describe the cathe-
tometer in detail, but it may be mentioned that it is of home construction,
and is mounted on centres attached to the floor and ceiling of the room. It
sufficed to record the levels to tenths of millimetres. The whole apparatus
was constructed by Mr Gordon.
If the glasses closing the optical tubes were perfect, there would be coin-
cidence of bands corresponding to complete exhaustion of both optical tubes.
A correction could be made for the residual error once for all determined, but
it is safer to make two independent settings, one at pressures as nearly atmo-
spheric as the case admits, and a second at minimum pressures. There are
then in all eight readings to be combined. An example may be taken from a
case already referred to : —
I. II. III. IV.
9770 9371 9749 9790
7272 2165 2469 7445
Columns I, II refer to the anomalous nitrogen, III and IV to the dried
air used as a standard of comparison. I and IV are the fixed manometer
tubes in communication with the optical tubes. The reduction may be
effected by subtraction of the rows :
2498 7206 7280 2345
Thus 4708, the difference between II and I, of the nitrogen balances
4935, the difference between III and IV, of air. The refractivity referred to
air is accordingly ffff, or T048.
366 THE CHARACTER OF THE IMPURITY FOUND IN NITROGEN GAS. [241
In this example the range of pressures for the air is 493'5 mm., or about
two-thirds of an atmosphere.
Great care is sometimes required to ensure matching the same bands in
the two settings. A mistake of one band in the above example would entail
nearly 2 per cent, error in the final result, inasmuch as the whole number of
bands concerned is about 96 per atmosphere of air, or about 62 over the
range actually used. It is wise always to include a match with pressures
about midway between the extremes. If the results harmonise, an error of
a single band is excluded ; and it is hardly possible to make a mistake of
two bands.
As regards accuracy, independent final results usually agree to one-
thousandth part.
242.
ON ISO-PERIODIC SYSTEMS.
[Philosophical Magazine, XLVI. pp. 567—569, 1898.]
IN general a system with m degrees of freedom vibrating about a con-
figuration of equilibrium has m distinct periods, or frequencies, of vibration,
but in particular cases two or more of these frequencies may be equal. The
simple spherical pendulum is an obvious example of two degrees of freedom
whose frequencies are equal. It is proposed to point out the properties
of vibrating systems of such a character that all the frequencies are equal.
In the general case when a system is referred to its normal coordinates
</>!, </>2, ... we have for the kinetic and potential energies*,
-*tflk...
and for the vibrations
4>1 = Acos(n1t-a), </>2 = B cos(n2t -/3), &c .......... (2)
where A, B, ... a, @ ... are arbitrary constants and
n^c,/^, n22=c2/a2, &c ...................... (3)
If «!, n^, &c., are all equal, T and V are of the same form except as
to a constant multiplier. By supposing a, ft . . . equal, we see that any
prescribed ratios may be assigned to fa, <£2 ..., so that vibrations of arbitrary
type are normal and can be executed without constraint. In particular any
parts of the system may remain at rest.
If x, y, z be the space coordinates (measured from the equilibrium position)
of any point of the system, the most general values are given by
x = X± cos nt + X2 sin nt \
y= Fjcos nt+ Y^sinnt L ........................ (4)
z = Z1 cos nt + Z2 sin nt }
* See, for example, Theory of Sound, § 87.
368 ON ISO-PERIODIC SYSTEMS. [242
where Xl} X2, &c. are constants for each point. These equations indicate
elliptic motion in the plane
x(Y1Z2-Z1Y2) + y(ZlX2-X1Z2) + z(XlY2-Y1XJ = 0 (5)
Thus every point of the system describes an elliptic orbit in the same periodic
time.
An interesting case is afforded by a line of similar bodies of which each
is similarly connected to its neighbours*. The general formula for w2 is
_ C, - 2fl cos ka - 2 C2 cos 2ka - . . .
~ A0- 2A, cos ka - 2A2 cos 2ka - . . . '
in which the constants CQ, C^ ... refer to the potential, and Al} A^ ... to
the kinetic energy. Here C1} A^ represent the influence of immediate
neighbours distant a from one another, C2, A2 the influence of neighbours
distant 2a, and so on. Further, k denotes 2-Tr/A,, X being the wave-length.
If C-i, C2 ... , Alt A2 ... vanish, each body is uninfluenced by its neighbours,
and the case is one considered by Reynolds of a number of similar and
disconnected pendulums hanging side by side at equal distances. It is
obvious that a vibration of any type is normal and is executed in the same
time. If we consider a progressive wave, its velocity is proportional to A,.
A disturbance communicated to any region has no tendency to propagate
itself ; the " group velocity " is zero.
Although the line of disconnected pendulums is interesting and throws
light upon the general theory of wave and group propagation, one can hardly
avoid the feeling that it is only by compliment that it is regarded as a single
system. It is therefore not without importance to notice that there are other
cases for which n assumes a constant, and the group-velocity a zero, value.
To this end it is only necessary that
C0:C1:Ca:...=A0:A1:A,: (7)
If this condition be satisfied, the connexion of neighbouring bodies does not
entail the propagation of disturbance. Any number of the bodies may remain
at rest, and all vibrations have the same period.
We might consider particular systems for which C2, C3 ... A.2, A3... vanish,
while CJ/CQ = A1/A0 ; but it is perhaps more interesting to draw an illustra-
tion from the case of continuous linear bodies. Consider a wire stretched
with tension T1} each element dx of which is urged to its position of equili-
brium (y = 0) by a force equal to pydx. The potential energyf is given by
(8)
* Phil. Mag. Vol. XLIV. p. 356, 1897. [Vol. iv. p. 340.]
t See Theory of Sound, §§ 122, 162, 188.
1898] ON ISO-PERIODIC SYSTEMS. 369
If the "rotatory inertia" be included, the corresponding expression for the
kinetic energy is
in which p is the volume density, to the area of cross section, and K the radius
of gyration of the cross section about an axis perpendicular to the plane
of bending. In waves along an actual wire vibrating transversely the second
term would be relatively unimportant, but there is no contradiction in the
supposition that the rotatory term is predominant. The differential equation
derived from (8) and (9) is
a2S+c22/=°> (10)
where a*=T1/pa>, c2 = /i//xo ...................... (11)
If we suppose that there is no tension and no rotatory inertia, a = 0, K = 0,
and the solution of (10) may be written
y = cos ct . yl + sinc£ . yZi .................. . ..... (12)
2/i > 2/2 being arbitrary functions of x. If yl = cos mx, yz— sin mx, (12) becomes
y = cos (ct — mx), ........................... (13)
and the velocity of propagation (elm) is proportional to \, equal to 27r/m.
This is the case of the disconnected pendulums.
On the other hand we may equally well suppose that c is zero and that
the rotatory inertia is paramount, so that (10) reduces to
The periodic part of the solution is again of the form (12), and has the same
peculiar properties as before.
In the general case we have the solution for stationary vibrations
y = sin mx cos nt, .............................. (14)
where m= ITT /I, i being an integer, and
This gives the frequencies for the various modes of vibration of a wire of
length I fastened at the ends.
If /e2 = a2/c2, n becomes independent of m as before.
If K2 < a2/c2,w2 increases, as i and m increase, and approaches a finite upper
limit a2//c2. The series of frequencies is thus analogous to those met with in
the spectra of certain bodies*.
* Compare Schuster, Nature, Vol. LV. p. 200 (1890).
R. iv. 24
243.
ON JAMES BERNOULLI'S THEOREM IN PROBABILITIES.
[Philosophical Magazine, XLVII. pp. 246 — 251, 1899.]
IF p denote the probability of an event, then the probability that in p,
trials the event will happen in times and fail n times is equal to a certain
term in the expansion of (p + q)*, namely,
m\n\r
where p + q=l, m + n = fjt,.
" Now it is known from Algebra that if m and n vary subject to the
condition that m + n is constant, the greatest value of the above term is
when m/n is as nearly as possible equal to p/q, so that m and n are as nearly
as possible equal to pp and pq respectively. WTe say as nearly as possible,
because p.p is not necessarily an integer, while m is. We may denote the
value of m by up + z, where z is some proper fraction, positive or negative ;
and then n = p,q — z"
The rth term, counting onwards, in the expansion of (p + q)* after (1) is
— r\
(2)
The approximate value of (2) when in and n are large numbers may be
obtained with the aid of Stirling's theorem, viz.
(3)
The process is given in detail after Laplace in Todhunter's History of the
Theory of Probability, p. 549, from which the above paragraph is quoted.
The expression for the rth term after the greatest is
n</pLprzr(n-m)_T* , 1* } .
rmn} ( mn 2mn 6m2 T 6w2) '
1899] ON JAMES BERNOULLI'S THEOREM IN PROBABILITIES. 371
and that for the rth term before the greatest may be deduced by changing
the sign of r in (4).
It is assumed that r2 does not surpass p, in order of magnitude, and
fractions of the order I//JL are neglected.
There is an important case in which the circumstances are simpler than
in general. It arises when p = q = £ , and //, is an even number, so that
m = n= £//,. Here z disappears ab initio, and (4) reduces to
representing (2), which now becomes
(6)
An important application of (5) is to the theory of random vibrations.
If /A vibrations are combined, each of the same phase but of amplitudes which
are at random either +1 or — 1, (5) represents the probability of \p + r of
them being positive vibrations, and accordingly \^—r being negative. In
this case, and in this case only, is the resultant + 2r. Hence if x represent
the resultant, the chance of x, which is necessarily an even integer, is
The next greater resultant is (x + 2); so that when x is great the above
expression may be supposed to correspond to a range for x equal to 2. If we
represent the range by dx, the chance of a resultant lying between x and
x + dx is given by
Another view of this matter, leading to (5) or (7) without the aid of
Stirling's theorem, or even of formula (1), is given (somewhat imperfectly) in
Theory of Sound, 2nd ed. § 42 a. It depends upon a transition from an
equation in finite differences .to the well-known equation for the conduction
of heat and the use of one of Fourier's solutions of the latter. Let/(/*, r)
denote the chance that the number of events occurring (in the special ap-
plication positive vibrations) is \p + r, so that the excess is r. Suppose that
each random combination of /* receives two more random contributions — two
in order that the whole number may remain even, — and inquire into the
chance of a subsequent excess r, denoted by /(ft + 2, r). The excess after the
addition can only be r if previously it were r — 1, r, or r + 1. In the first
case the excess becomes r by the occurrence of both of the two new events,
* Phil. Mag. Vol. x. p. 75 (1880). [Vol. i. p. 491.]
372 ON JAMES BERNOULLI'S THEOREM IN PROBABILITIES. [243
of which the chance is \ . In the second case the excess remains r in conse-
quence of one event happening and the other failing, of which the chance is
£; and in the third case the excess becomes r in consequence of the failure
of both the new events, of which the chance is \. Thus
/(/* + 2, r) = If (p., r - 1) + i/0*. r) + £/(,., r + 1) ....... (8)
According to the present method the limiting form of f is to be derived from
(8). We know, however, that/ has actually the value given in (6), by means
of which (8) may be verified.
Writing (8) in the form
/(/* + 2, r) -f(p., r) = i/0*. r - 1) - 1/0*. r) + £/(,*, r + 1), ...(9)
we see that when p. and r are infinite the left-hand member becomes Zdf/dfj,,
and the right-hand member becomes ^d^f/dr2, so that (9) passes into the
differential equation
In (9), (10) r is the excess of the actual occurrences over |/z. If we take #
to represent the difference between the number of occurrences and the number
of failures, x = 2r and (10) becomes
#-*#' (11)
dp, 2dx*'
In the application to vibrations /(/A, #) then denotes the chance of a resultant
+ x from a combination of p, unit vibrations which are positive or negative
at random.
In the formation of (10) we have supposed for simplicity that the addition
to p, is 2, the lowest possible consistently with the total number remaining
even. But if we please we may suppose the addition to be any even number
//. The analogue of (8) is then
2*' ./(/, + /, r) = /(,*, r - I,*') -f p.' /(p., r - ^ + 1)
+ /-i)/(^ r _ ^ + 2) + _
and when /A is treated as very great the right-hand member becomes
*' (/*' - 2)2 + 1 . //2 .
1899] ON JAMES BERNOULLI'S THEOREM IN PROBABILITIES. 373
The series which multiplies f is (1 + \Y'> or 2M/. The second series is
equal to jjf . 2** ', as may be seen by comparison of coefficients of #2 in the
equivalent forms
(e* + e~x)n = 2" (1 + %x* + . . .)»
.
The value of the left-hand member becomes simultaneously
so that we arrive at the same differential equation (10) as before.
This is the well-known equation for the conduction of heat, and the
solution developed by Fourier is at once applicable. The symbol /JL corre-
sponds to time and r to a linear coordinate. The special condition is that
initially — that is when /* is relatively small — /must vanish for all values of r
that are not small. We take therefore
which may be verified by differentiation.
The constant A may be determined by the understanding that/(/ci, r)dr
is to represent the chance of an excess lying between r and r + dr, and that
accordingly
+)rfr = l ............................ (13)
r+oo
Since I e~lftdz = *Jir, we have
£
and, finally, as the chance that the excess lies between r and r + dr,
Another method by which A in (12) might be determined would be by
comparison with (6) in the case of r = 0. In this way we find
A til 1.3.5...0*-!)
\ 2.4.6
J (
—} by Wallis' theorem.
374 ON JAMES BERNOULLI'S THEOREM IN PROBABILITIES. [243
If, as is natural in the problem of random vibrations, we replace r
by x, denoting the difference between the number of occurrences and
the number of failures, we have as the chance that x lies between x and
x + dx
identical with (7).
In the general case when p and q are not limited to the values £,
it is more difficult to exhibit the argument in a satisfactory form,
because the most probable numbers of occurrences and failures are no
longer definite, or at any rate simple, fractions of /i. But the general
idea is substantially the same. The excess of occurrences over the most
probable number is still denoted by r, and its probability by /(/*, r}. We
regard r as continuous, and we then suppose that p increases by unity.
If the event occurs, of which the chance is p, the total number of occurrences
is increased by unity. But since the most probable number of occurrences
is increased by p, r undergoes only an increase measured by 1 — p or q.
In like manner if the event fails, r undergoes a decrease measured by p.
Accordingly
(17)
On the right of (17) we expandy(/A, r — q), f([i, r + p) in powers of p and q.
Thus
so that the right-hand member is
The left-hand member may be represented by/+ df/d/j,, so that ultimately
Accordingly by the same argument as before the chance of an excess r lying
between r and r + dr is given by
(19)
We have already considered the case of p = q = |. Another particular case
of importance arises when p is very small, and accordingly q is nearly equal
to unity. The whole number /* is supposed to be so large that pjj,, or m,
1899] ON JAMES BERNOULLI'S THEOREM IN PROBABILITIES. 375
representing the most probable number of occurrences, is also large. The
general formula now reduces to
1
_r2/2r/iJr. /9ft \
V(2^)e
which gives the probability that the number of occurrences shall lie between
m + r and m + r + dr. It is a function of m and r only.
The probability of the deviation from m lying between + r
(21)
where r = r/\/(2m). This is equal to '84 when r = TO, or r = ^(2m) ; so that
the chance is comparatively small of a deviation from m exceeding + V(2w).
For example, if m is 50, there is a rather strong probability that the actual
number of occurrences will lie between 40 and 60.
The formula (20) has a direct application to many kinds of statistics.
244.
ON THE COOLING OF AIR BY RADIATION AND CONDUCTION,
AND ON THE PROPAGATION OF SOUND.
[Philosophical Magazine, XLVII. pp. 308—314, 1899.]
ACCORDING to Laplace's theory of the propagation of Sound the expansions
(and contractions) of the air are supposed to take place without transfer of
heat. Many years ago Sir G. Stokes* discussed the question of the influence
of radiation from the heated air upon the propagation of sound. He showed
that such small radiating power as is admissible would tell rather upon the
intensity than upon the velocity. If a; be measured in the direction of
propagation, the factor expressing the diminution of amplitude is e~mx, where
m = Tl±£m ...(i)
7 2a
In (1) 7 represents the ratio of specific heats (1'41), a is the velocity of sound,
and q is such that e~qt represents the law of cooling by radiation of a small
mass of air maintained at constant volume. If r denote the time required to
traverse the distance x, r = x/a, and (1) may be taken to assert that the
amplitude falls to any fraction, e.g. one-half, of its original value in 7 times
the interval of time required by a mass of air to cool to the same fraction
of its original excess of temperature. " There appear to be no data by which
the latter interval can be fixed with any approach to precision ; but if we
take it at one minute, the conclusion is that sound would be propagated for
(seven) minutes, or travel over about (80) miles, without very serious loss from
this cause f." We shall presently return to the consideration of the probable
value of q.
Besides radiation there is also to be considered the influence of conductivity
in causing transfer of heat, and further there are the effects of viscosity.
« Phil. Mag. [4] i. p. 305, 1851 ; Theory of Sound, § 247.
t Proc. Roy. Inst. April 9, 1897. [Vol. iv. p. 298.]
1899] ON THE COOLING OF AIR BY RADIATION AND CONDUCTION. 377
The problems thus suggested have been solved by Stokes and Kirchhoff*.
If the law of propagation be
U = e-m'*co8(nt-as/a), (2)
then
in which the frequency of vibration is w/2-Tr, /jf is the kinematic viscosity, and
v the thermometric conductivity. In c.G.S. measure we may take // = "14,
v = '26, so that
To take a particular case, let the frequency be 256 ; then since a = 33200,
we find for the time of propagation during which the amplitude diminishes
in the ratio of e : 1,
(ma)-1 = 3560 seconds.
Accordingly it is only very high sounds whose propagation can be ap-
preciably influenced by viscosity and conductivity.
If we combine the effects of radiation with those of viscosity and conduction,
we have as the factor of attenuation
Q— (m+m')x
where m + m' = "14< (q / a) + !12(n9/a*) ...................... (4)
In actual observations of sound we must expect the intensity to fall off
in accordance with the law of inverse squares of distances. A very little
experience of moderately distant sounds shows that in fact the intensity is in
a high degree uncertain. These discrepancies are attributable to atmospheric
refraction and reflexion, and they are sometimes very surprising. But the
question remains whether in a uniform condition of the atmosphere the
attenuation is sensibly more rapid than can be accounted for by the law of
inverse squares. Some interesting experiments towards the elucidation of
this matter have been published by Mr Wilmer Duff -f-, who compared the
distances of audibility of sounds proceeding respectively from two and from
eight similar whistles. On an average the eight whistles were audible only
about one-fourth further than a pair of whistles ; whereas, if the sphericity of
the waves had been the only cause of attenuation, the distances would have
been as 2 to 1. Mr Duff considers that in the circumstances of his experi-
ments there was little opportunity for atmospheric irregularities, and he
attributes the greater part of the falling off to radiation. Calculating from
(1) he deduces a radiating power such that a mass of air at any given excess
of temperature above its surroundings will (if its volume remain constant)
fall by radiation to one-half of that excess in about one-twelfth of a second.
* Fogg. Ann. Vol. cxxxiv. p. 177, 1868 ; Theory of Sound, 2nd ed. § 348.
t Phys. Review, Vol. vi. p. 129, 1898.
378 ON THE COOLING OF AIB BY RADIATION AND CONDUCTION, [244
In this paper I propose to discuss further the question of the radiating
power of air, and I shall contend that on various grounds it is necessary to
restrict it to a value hundreds of times smaller than that above mentioned.
On this view Mr Duff's results remain unexplained. For myself I should
still be disposed to attribute them to atmospheric refraction. If further
experiment should establish a rate of attenuation of the order in question
as applicable in uniform air, it will I think be necessary to look for a cause
not hitherto taken into account. We might imagine a delay in the equaliza-
tion of the different sorts of energy in a gas undergoing compression, not
wholly insensible in comparison with the time of vibration of the sound. If
in the dynamical theory we assimilate the molecules of a gas to hard smooth
bodies which are nearly but not absolutely spherical, and trace the effect of a
rapid compression, we see that at the first moment the increment of energy is
wholly translational and thus produces a maximum effect in opposing the
compression. A little later a due proportion of the excess of energy will
have passed into rotational forms which do not influence the pressure, and
this will accordingly fall off. Any effect of the kind must give rise to
dissipation, and the amount of it will increase with the time required for the
transformations, i.e. in the above mentioned illustration with the degree of
approximation to the spherical form. In the case of absolute spheres no
transformation of translatory into rotatory energy, or vice versa, would
occur in a finite time. There appears to be nothing in the behaviour of
gases, as revealed to us by experiment, which forbids the supposition of
a delay capable of influencing the propagation of sound.
Returning now to the question of the radiating power of air, we may
establish a sort of superior limit by an argument based upon the theory of
exchanges, itself firmly established by the researches of B. Stewart. Consider
a spherical mass of radius r, slightly and uniformly heated. Whatever may
be the radiation proceeding from a unit of surface, it must be less than the
radiation from an ideal black surface under the same conditions. Let us,
however, suppose that the radiation is the same in both cases and inquire
what would then be the rate of cooling. According to Bottomley* the
emissivity of a blackened surface moderately heated is '0001. This is the
amount of heat reckoned in water-gram-degree units emitted in one second
from a square centimetre of surface heated 1° C. If the excess of temperature
be 6, the whole emission is
0 x 47rr2 x -0001
On the other hand, the capacity for heat is
fur3 x -0013 x -24,
the first factor being the volume, the second the density, and the third the
* Everett, C.G.S. Units, 1891, p. 134.
1899] AND ON THE PROPAGATION OF SOUND. 379
specific heat of air referred, as usual, to water. Thus for the rate of cooling,
d6 '0003 1
whence 0 = 00ertlr, ................................. (5)
00 being the initial value of 0. The time in seconds of cooling in • the
ratio of e : 1 is thus represented numerically by r expressed in centims.
When r is very great, the suppositions on which (5) is calculated will
be approximately correct, and that equation will then represent the actual
law of cooling of the sphere of air, supposed to be maintained uniform by
mixing if necessary. But ordinary experience, and more especially the
observations of Tyndall upon the diathermancy of air, would lead us to
suppose that this condition of things would not be approached until r
reached 1000 or perhaps 10,000 centims. For values of r comparable with
the half wave-length of ordinary sounds, e.g. 30 centim., it would seem that
the real time of cooling must be a large multiple of that given by (5).
At this rate the time of cooling of a mass of air must exceed, and probably
largely exceed, 60 seconds. To suppose that this time is one-twelfth of a
second would require a sphere of air 2 millim. in diameter to radiate as much
heat as if it were of blackened copper at the same temperature.
Although, if the above argument is correct, there seems little likelihood
of the cooling of moderate masses of air being sensibly influenced by radiation,
1 thought it would be of interest to inquire whether the observed cooling (or
heating) in an experiment on the lines of Clement and Desormes could be
adequately explained by the conduction of heat from the walls of the vessel
in accordance with the known conductivity of air. A nearly spherical vessel
of glass of about 35 centim. diameter, well encased, was fitted, air-tight, with
two tubes. One of these led to a manometer charged with water or sulphuric
acid; the other was provided with a stopcock and connected with an air-
pump. In making an experiment the stopcock was closed and a vacuum
established in a limited volume upon the further side. A rapid opening and
reclosing of the cock allowed a certain quantity of air to escape suddenly, and
thus gave rise to a nearly uniform cooling of that remaining behind in the
vessel. At the same moment the liquid rose in the manometer, and the
observation consisted in noting the times (given by a metronome beating
seconds) at which the liquid in its descent passed the divisions of a scale,
as the air recovered the temperature of the containing vessel. The first
record would usually be at the third or fourth second from the turning of the
cock, and the last after perhaps 120 seconds. In this way data are obtained
for a plot of the curve of pressure ; and the part actually observed has to
be supplemented by extrapolation, so as to go back to the zero of time (the
moment of turning the tap) and to allow for the drop which might occur
380 ON THE COOLING OF AIR BY RADIATION AND CONDUCTION, [244
subsequent to the last observation. An estimate, which cannot be much in
error, is thus obtained of the whole rise in pressure during the recovery of
temperature, and for the time, reckoned from the commencement, at which
the rise is equal to one-half of the total.
In some of the earlier experiments the whole rise of pressure (fall in the
manometer) during the recovery of temperature was about 20 millim. of
water, and the time of half recovery was 15 seconds. I was desirous of
working with the minimum range, since only in this way could it be hoped
to eliminate the effect of gravity, whereby the interior and still cool parts
of the included air would be made to fall and so come into closer proximity
to the walls, and thus accelerate the mean cooling. In order to diminish
the disturbance due to capillarity, the bore of the manometer-tube, which
stood in a large open cistern, was increased to about 18 millim.*, and suitable
optical arrangements were introduced to render small movements easily
visible. By degrees the range was diminished, with a prolongation of the
time of half recovery to 18, 22, 24, and finally to about 26 seconds. The
minimum range attained was represented by 3 or 4 millim. of water, and at
this stage there did not appear to be much further prolongation of cooling
in progress. There seemed to be no appreciable difference whether the
air was artificially dried or not, but in no case was the moisture sufficient
to develop fog under the very small expansions employed. The result of the
experiments may be taken to be that when the influence of gravity was,
as far as practicable, eliminated, the time of half recovery of temperature was
about 26 seconds.
It may perhaps be well to give an example of an actual experiment.
Thus in one trial on Nov. 1, the recorded times of passage across the divisions
of the scale were 3, 6, 11, 18, 26, 35, 47, 67, 114 seconds. The divisions
themselves were millimetres, but the actual movements of the meniscus were
less in the proportion of about 2£ : 1. A plot of these numbers shows that
one division must be added to represent the movement between 0s and 3s,
and about as much for the movement to be expected between 114s and oo .
The whole range is thus 10 divisions (corresponding to 4 millim. at the
meniscus), and the mid-point occurs at 26s. On each occasion 3 or 4
sets of readings were taken under given conditions with fairly accordant
results.
It now remains to compare with the time of heating derived from theory.
The calculation is complicated by the consideration that when during the
process any part becomes heated, it expands and compresses all the other
parts, thereby developing heat in them. From the investigation which
* It must not be forgotten that too large a diameter is objectionable, as leading to an
augmentation of volume during an experiment, as the liquid falls.
1899] AND ON THE PROPAGATION OF SOUND. 381
follows *, we see that the time of half recovery t is given by the formula
in which a is the radius of the sphere, 7 the ratio of specific heats (1'41), and
v is the thermometric conductivity, found by dividing the ordinary or calori-
metric conductivity by the thermal capacity of unit volume. This thermal
capacity is to be taken with volume constant, and it will be less than the
thermal capacity with pressure constant in the ratio of 7 : 1. Accordingly v/y
in (6) represents the latter thermal capacity, of which the experimental value
is '00128 x '239, the first factor representing the density of air referred to
water. Thus, if we take the calorimetric conductivity at '000056, we have in
C.G.s. measure
i> = -258, i;/7 = 183;
and thence
t = '102a2.
In the present apparatus a, determined by the contents, is 16'4 centim.,
whence
t = 2 7 '4 seconds.
The agreement of the observed and calculated values is quite as close
as could have been expected, and confirms the view that the transfer of heat
is due to conduction, and that the part played by radiation is insensible.
From a comparison of the experimental and calculated curves, however,
it seems probable that the effect of gravity was not wholly eliminated, and
that the later stages of the phenomenon, at any rate, may still have been
a little influenced by a downward movement of the central parts.
* See next paper.
245.
ON THE CONDUCTION OF HEAT IN A SPHERICAL MASS
OF AIR CONFINED BY WALLS AT A CONSTANT
TEMPERATURE.
[Philosophical 'Magazine, XLVII. pp. 314 — 325, 1899.]
IT is proposed to investigate the subsidence to thermal equilibrium of
a gas slightly disturbed therefrom and included in a solid vessel whose
walls retain a constant temperature. The problem differs from those con-
sidered by Fourier in consequence of the mobility of the gas, which may give
rise to two kinds of complication. In the first place gravity, taking ad-
vantage of the different densities prevailing in various parts, tends to produce
circulation. In many cases the subsidence to equilibrium must be greatly
modified thereby. But this effect diminishes with the amount of the
temperature disturbance, and for infinitesimal disturbances the influence
of gravity disappears. On the other hand, the second complication remains,
even though we limit ourselves to infinitesimal disturbances. When one
part of the gas expands in consequence of reception of heat by radiation
or conduction, it compresses the remaining parts, and these in their turn
become heated in accordance with the laws of gases. To take account of
this effect a special investigation is necessary.
But although the fixity of the boundary does not suffice to prevent local
expansions and contractions and consequent motions of the gas, we may
nevertheless neglect the inertia of these motions since they are very slow
in comparison with the free oscillations of the mass regarded as a resonator.
Accordingly the pressure, although variable with time, may be treated as
uniform at any one moment throughout the mass.
In the usual notation*, if s be the condensation and 6 the excess of
temperature, the pressure p is given by
(1)
* Theory of Sound, § 247.
1899] ON THE CONDUCTION OF HEAT IN A SPHERICAL MASS OF AIR. 383
The effect of a small sudden condensation s is to produce an elevation of
temperature, which may be denoted by fts. Let dQ be the quantity of heat
entering the element of volume in the time dt, measured by the rise of
temperature which it would produce, if there were no " condensation."
Then
dO ds d
and, if the passage of dQ be the result of radiation and conduction, we have
f = vw-qe .............................. .(3)
In (3) v represents the " therrnometric conductivity " found by dividing the
conductivity by the thermal capacity of the gas (per unit volume), at constant
volume. Its value for air at 0° and atmospheric pressure may be taken to be
•26 cm2. /sec. Also q represents the radiation, supposed to depend only upon
the excess of temperature of the gas over that of the enclosure.
If dQ = 0, 0 = /3s, and in (1)
so that
l + «/9 = 7, ................................. (4)
where 7 is the well-known ratio of specific heats, whose value for air and
several other gases is very nearly 1/41.
In general from (2) and (3)
In order to find the normal modes into which the most general subsidence
may be analysed, we are to assume that s and 6 are functions of the time
solely through the factor e~ht. Since p is uniform, s + a.6 must by (1) be of
the form He~ht, where H is some constant ; so that if for brevity the factor
e~ht be dropped,
s + a0 = H; ................................. (6)
while from (5)
q)e = hps ......................... (7)
Eliminating s between (5) and (7), we get
V20 + m* (6 - C) = 0, ........................... (8)
where
m, = h-_q 0_Wff ......................
v hj — q
These equations are applicable in the general case, but when radiation
and conduction are both operative the equation by which ra is determined
384 ON THE CONDUCTION OF HEAT IN A SPHERICAL MASS OF AIR [245
becomes rather complicated. If there be no conduction, v = 0. The solution
is then very simple, and may be worth a moment's attention.
Equations (6) and (7) give
hftH
.(10)
hy-q'
Now the mean value of s throughout the mass, which does not change with
the time, must be zero ; so that from (10) we obtain the alternatives
(i) h = q, (ii) H = 0.
Corresponding to (i) we have with restoration of the time-factor
«=0 ...................... (11)
In this solution the temperature is uniform and the condensation zero
throughout the mass. By means of it any initial mean temperature may be
provided for, so that in the remaining solutions the mean temperature may
be considered to be zero.
In the second alternative H— 0, so that s = - aO. Using this in (7) with
v evanescent, we get
07-00 = 0 ............................... (12)
The second solution is accordingly
......... (13)
where <f> denotes a function arbitrary throughout the mass, except for the
restriction that its mean value must be zero.
Thus if © denote the initial value of 0 as a function of x, y, z, and ©0 its
mean value, the complete solution may be written
e = ®0e-<it + (®-G0)e-#iY \
k .................. (14)
8= _a(e-@0)e-9'/yJ
giving
s + a0=a®Qe-# ............................ (15)
It is on (15) that the variable part of the pressure depends.
When the conductivity v is finite, the solutions are less simple and involve
the form of the vessel in which the gas is contained. As a first example
we may take the case of gas bounded by two parallel planes perpendicular
to x, the temperature and condensation being even functions of x measured
from the mid-plane. In this case V2 = d?/da?, and we get
6 = C + A cos mx, -s/a = D + Acosmx, ............ (16)
<*C-aD = H. ........................ (17)
1899] CONFINED BY WALLS AT A CONSTANT TEMPERATURE. 385
By (9), (17)
y-q
There remain two conditions to be satisfied. The first is simply that 6 = 0
when x = ± a, 2a being the distance between the walls. This gives
0 + Acosma=0 ............................ (19)
The remaining condition is given by the consideration that the mean value
of s, proportional to jsdx, must vanish. Accordingly
ma.D + sinma.A=Q ......................... (20)
From (18), (19), (20) we have as the equation for the admissible values
of m,
tan ma _ a@q — vm?
ma ~ z '
reducing for the case of evanescent q to
ma a/3'
The general solution may be expressed in the series
}
)
(23)
where h1} h2>... are the values of h corresponding according to (9) with the
various values of m, and 0l} 02 ... are of the form
0l = cos TOI# — cos TO!«. )
I (24)
It only remains to determine the arbitrary constants Alt A2, ... to suit
prescribed initial conditions. We will limit ourselves to the simpler case
of q = 0, so that the values of m are given by (22). With use of this relation
and putting for brevity a = 1, we find from (24)
r1 a/3 + 1
J— -5 — cos TO! cos ra2,
a/3 + 1
s^dsc = ^7^ — cos TO! cos m^;
so that
0, (25)
'o Jo
?,, 02 being any (different) functions of the form (24). Also
E. jv. 25
386 ON THE CONDUCTION OF HEAT IN A SPHERICAL MASS OF AIR [245
There is now no difficulty in finding Alt Az, ... to suit arbitrary initial
values of 6 and its associated s, i.e. so that
& = A10l + A,0«+... }
......................... (27)
S=AISI + A*SS + ... J
Thus to determine Al}
\l(%0l + /3/a . SSl) dx = A, P(0f + y3/a . O dx
o Jo
in which the coefficients of A2, As ... vanish by (25); so that by (26)
An important particular case is that in which 0 is constant, and accordingly
S = 0. Since
f1 „
I 6l
Jo
sin m, 1 4- a/3
— -- cos 7^1 = -- 7r- cos ???i ,
ap
we have
For the pressure we have
-
a/3
or in the particular case of (29),
cos w, .
a
(30)
If /3 = 0, we fall back upon a problem of the Fourier type. By (22) in
that case
ma = |TT (1, 3, 5, . . . ) and cos2 ma = a-fi2/
so that (30) becomes
or initially
80 n 1 !_
The values of h are given by
...(32)
1899] CONFINED BY WALLS AT CONSTANT TEMPERATURE. 387
We will now pass on to the more important practical case of a spherical
envelope of radius a. The equation (8) for (6 — C) is identical with that
which determines the vibrations of air* in a spherical case, and the solution
may be expanded in Laplace's series. The typical term is
(mr).Yn, ..................... (33)
Yn being the surface spherical harmonic of order n where n = 0, 1, 2, 3 ... ,
and J the symbol of Bessel's functions. In virtue of (6) we may as before
equate - s/a - D, where D is another constant, to the right-hand member of
(33). The two conditions yet to be satisfied are that 6 = 0 when r = a, and
that the mean value of s throughout the sphere shall vanish.
When the value of n is greater than zero, the first of these conditions
gives (7=0 and the second D — 0 ; so that
0 = -s(* = (mr)-Un+i(mr).Yn, .................. (34)
and s + ad = 0. Accordingly these terms contribute nothing to the pressure.
It is further required that
Jn+l(ma) = 0, .............................. (35)
by which the admissible values of m are determined. The roots of (35)
are discussed in Theory of Sound, § 206... ; but it is not necessary to go
further into the matter here, as interest centres rather upon the case n = 0.
If we assume symmetry with respect to the centre of the sphere, we may
1 d2
replace V2 in (8) by - r~z r, thus obtaining
(36)
of which the general solution is
But for the present purpose the term r~l cos mr is excluded, so that we may
write
, ......... (37)
mr mr
giving
s + a0 = a(C-D)=H. ..................... (37 bis)
The first special condition gives
maC + B sin ma = 0 ......................... (38)
The second, that the mean value of s shall vanish, gives on integration
^m3a?D + B (sin ma — ma cos ma) = 0 ................ (39)
* Theory of Sound, Vol. IT. ch. xvii.
25—2
388 ON THE CONDUCTION OF HEAT IN A SPHERICAL MASS OF AIR [245
Equations (18), derived from (9) and (37 bis), giving C and D in terms
of H, hold good as before. Thus
*
G~ haft aft(q+Vm*)'
Equating this ratio to that derived from (38), (39), we find
3 ma cos ma — sin ma _ vmz — aftq . - .
m2a2 sin ma aft (vmz + q) '
This is the equation from which m is to be found, after which h is given
by (9).
In the further discussion we will limit ourselves to the case of q = 0,
when (41) reduces to
l), ........................ (42)
in which a has been put equal to unity. Here by (40)
D = -C/aft.
Thus we may set, as in (23),
6 = B1e-h>t0l + Bze-h*t02+ ...... )
k ....' ........... (43)
s=Ble-h*tsl+B9erUs2 + ...... j
in which
.. sin ??ij?' sin m^a sin w,r 1 sin VIM
0i= — --- — , «! = — «— — ...(44)
m{r m^a, my ft mta
and by (9) J^—vm^/y. Also
The process for determining B1} B2, ... follows the same lines as before.
By direct integration from (44) we find
_ sin (m-i — m^) _ sin (m^ + m2) 2 sin ml sin m2
Wj — 77^2 ml + m,t 3«/3
a being put equal to unity. By means of equation (42) satisfied by m±
and ra2 we may show that the quantity on the right in the above equation
vanishes. For the sum of the first two fractions is
2m2 sin ml cos ra2 — 2m1 sin w2 cos m^
of which the denominator by (42) is equal to
3a/3 (nh cot ml — m2 cot r?i2).
1899] CONFINED BY WALLS AT CONSTANT TEMPERATURE. 389
Accordingly f (010., + ^/a.s1s.z)r2dr = 0 ...................... (46)
Jo
Also
To determine the arbitrary constants Bl ... from the given initial values
of 9 and s, say ® and 8, we proceed as usual. We limit ourselves to the
term of zero order in spherical harmonics, i.e. to tne supposition that 6, s
are functions of r only. The terms of higher order in spherical harmonics, if
present, are treated more easily, exactly as in the ordinary theory of the
conduction of heat. By (43)
and thus I \0 6l + 01 a . SsJ f2dr = B, fW* + £/ a . 6V2) i*dr
Jo Jo
z !\0102 + /3/a . 8,8,} r*dr + ...... ,
Jo
in which the coefficients of B«, B3> ... vanish by (46). The coefficient of Bt
is given by (47). Thus
by which Bl is determined.
An important particular case is that where ® is constant and accordingly
S vanishes. Now with use of (42)
f1 sin ml — m1 cos m1 sin m1 _ (1 + a/3) sin ml
Jo1 mf ~~3m^~~
so that
sin 2m1 2sin2m!] 2m^ sin m^ . ® /Kn,
"2^" "S^")" 3«/3
Bl, B.,, ... being thus known, 0 and s are given as functions of the time and
of the space coordinates by (43), (44).
To determine the pressure in this case we have from (45)
0 + s/a. _ I +a/3 _ sin2 m . e~ht ( ,
~~ sin 2m\ '
the summation extending to all the values of m in (42). Since (for each
term) the mean value of s is zero, the right-hand member of (51) represents
also 0/®, where 0 is the mean value of 0.
If in (51) we suppose /3 = 0, we fall back upon a known Fourier solution,
390 ON THE CONDUCTION OF HEAT IN A SPHERICAL MASS OF AIR [245
relative to the mean temperature of a spherical solid which, having been
initially at uniform temperature ® throughout, is afterwards maintained
at zero all over the surface. From (42) we see that in this case sin in is
small and of order /3. Approximately
sin m = 3a/3lm ;
and (51) reduces to
0 6 ,e-M e-M e-/M
of which by a known formula the right-hand member identifies itself with
unity when t = 0. By (9) with restoration of a,
h = (I2, 32, 52, ...)*/7r2/a2 (53)
In the general case we may obtain from (42) an approximate value
applicable when m is moderately large. The first approximation is m = ITT,
i denoting an integer. Successive operations give
3a£ ISa-ft2 + 9a3/33
m = 17T + — ; ; • (54)
ITT i 77"^
In like manner we find approximately in (51)
sin2 m (1 + qff)/a/3 = 6 (1 + a/8) L 15ay8 + 9a2y8-- }
. 3a n
sin2 m -\ — -,
sin
ftH
• • -(55)
showing that the coefficients of the terms of high order in (51) differ from the
corresponding terms in (52) only by the factor (1 + a/3) or 7.
In the numerical computation we take 7 = 1*41, a/3 = '41. The series (54)
suffices for finding m when i is greater than 2. The first two terms are
found by trial and error with trigonometrical tables from (42). In like
manner the approximate value of the left-hand member of (51) therein given
suffices when i is greater than 3. The results as far as i = 12 are recorded in
the annexed table.
i
mjw
Left-hand
member
of (55)
i
m/T
Left-hand
member
of (55)
1
1-0994
•4942
7
7-0177
•0175
2
2-0581
•1799
8
8-0156
•0134
3
3-0401
•0871
9
9-0138
•0106
4
4-0305
•0510
10
10-0125
•0086
5
5-0246
•0332
11
11-0113
•0071
6
6-0206
•0233
12
12-0104
•0060
Thus the solution (51) of our problem is represented by
0/0 = •4942e-(1-°9!M>!i<'-l--l799e-(2-0581)2t'+ ...
.(56)
1899]
CONFINED BY WALLS AT CONSTANT TEMPERATURE.
391
where by (9), with omission of q and restoration of a,
t'/t = Tr'vlyct? ............................... (57)
The numbers entered in the third column of the above table would
add up to unity if continued far enough. The verification is best made
by a comparison with the simpler series (52). If with t zero we call this
series 2' and the present series 2, both 2 and 2' have unity for their sum,
and accordingly 7^' — 2 = 7 — 1, or
Here Qy/tr2 = '8573, and the difference between this and the first term of
S, i.e. '4942, is '3631. The differences of the second, third, &c. terms are
•0344, -0082, -0026, '0011, '0005, '0000, &c., making a total of '4099.
We are now in a position to compute the right-hand member of (56)
as a function of t'. The annexed table contains sufficient to give an idea
t'
(56)
t'
(56)
t'
(56)
•oo
1-0000
•40
•3401
•90
•1705
•05
•7037
•50
•2926
1-00
•1502
•10
•6037
•60
•2538
1-50
•0809
•20
•4811
•70
•2215
2-00
•0441
•30
•4002
•80
•1940
of the course of the function. It is plotted in the figure. The second entry
(t' = -05) requires the inclusion of 9 terms of the series. After t' = '7 two
terms suffice ; and after t' = 2'0 the first term represents the series to four
places of decimals.
By interpolation we find that the series attains the value '5 when
(58)
246.
TRANSPARENCY AND OPACITY.
[Proc. Roy. Inst. xvi. pp. 116—119, 1899; Nature, LX. pp. 64, 65, 1899.]
ONE kind of opacity is due to absorption; but the lecture dealt rather
with that deficiency of transparency which depends upon irregular reflections
and refractions. One of the best examples is that met with in Christiansen's
experiment. Powdered glass, all from one piece and free from dirt, is placed
in a bottle with parallel flat sides. In this state it is quite opaque ; but
if the interstices between the fragments are filled up with a liquid mixture
of bisulphide of carbon and benzole, carefully adjusted so as to be of equal
refractivity with the glass, the mass becomes optically homogeneous, and
therefore transparent. In consequence, however, of the different dispersive
powers of the two substances, the adjustment is good for one part only of the
spectrum, other parts being scattered in transmission much as if no liquid
were employed, though, of course, in a less degree. The consequence is that
a small source of light, backed preferably by a dark ground, is seen in its
natural outlines but strongly coloured. The colour depends upon the precise
composition of the liquid, and further varies with the temperature, a few
degrees of warmth sufficing to cause a transition from red through yellow to
green.
The lecturer had long been aware that the light regularly transmitted
through a stratum from 15 to 20 mm. thick was of a high degree of purity,
but it was only recently that he found to his astonishment, as the result of a
more particular observation, that the range of refrangibility included was but
two and a half times that embraced by the two D-lines. The poverty of
general effect, when the darkness of the background is not attended to, was
thus explained; for the highly monochromatic and accordingly attenuated
light from the special source is then overlaid by diffused light of other
colours.
1899] TRANSPARENCY AND OPACITY. 393
More precise determinations of the range of light transmitted were
subsequently effected with thinner strata of glass powder contained in cells
formed of parallel glass. The cell may be placed between the prisms of the
spectroscope and the object-glass of the collimator. With the above mentioned
liquids a stratum 5 mm. thick transmitted, without appreciable disturbance, a
range of the spectrum measured by 11 '3 times the interval of the D's. In
another cell of the same thickness an effort was made to reduce the difference
of dispersive powers. To this end the powder was of plate glass and the
liquid oil of cedar- wood adjusted with a little bisulphide of carbon. The
general transparency of this cell was the highest yet observed. When it
was tested upon the spectrum, the range of refrangibility transmitted was
estimated at 34 times the interval of the D's.
As regards the substitution of other transparent solid material for glass,
the choice is restricted by the presumed necessity of avoiding appreciable
double refraction. Common salt is singly refracting, but attempts to use
it were not successful. Opaque patches always interfered. With the idea
that these might be due to included mother- liquor, the salt was heated to
incipient redness, but with little advantage. Transparent rock-salt artificially
broken may, however, be used with good effect, but there is some difficulty in
preventing the approximately rectangular fragments from arranging them-
selves too closely.
The principle of evanescent refraction may also be applied to the spectro-
scope. Some twenty years ago, an instrument had been constructed upon
this plan*. Twelve 90° prisms of Chance's "dense flint" were cemented in a
row upon a strip of glass (Fig. 1), and the whole was immersed in a liquid
mixture of bisulphide of carbon with a little benzole. The dispersive power
of the liquid exceeds that of the solid, and the difference amounts to about
three-quarters of the dispersive power of Chance's " extra dense flint." The
Fig. 1.
resolving power of the latter glass is measured by the number of centimetres
of available thickness, if we take the power required to resolve the jD-lines as
unity. The compound spectroscope had an available thickness of 12 inches
or 30 cm., so that its theoretical resolving power (in the yellow region of the
spectrum) would be about 22. With the aid of a reflector the prism could be
used twice over, and then the resolving power is doubled.
* [Vol. i. p. 456.]
394 TRANSPARENCY AND OPACITY. [246
One of the objections to a spectroscope depending upon bisulphide of
carbon is the sensitiveness to temperature. In the ordinary arrangement of
prisms the refracting edges are vertical. If, as often happens, the upper part
of a fluid prism is warmer than the lower, the definition is ruined, one degree
(Centigrade) of temperature making nine times as great a difference of
refraction as a passage from Z^ to D.2. The objection is to a great extent
obviated by so mounting the compound prism that the refracting edges are
horizontal, which of course entails a horizontal slit. The disturbance
due to a stratified temperature is then largely compensated by a change
of focus.
In the instrument above described the dispersive power is great — the
D-lines are seen widely separated with the naked eye — but the aperture is
inconveniently small (|-inch). In the new instrument exhibited the prisms
(supplied by Messrs Watson) are larger, so that a line of ten prisms occupies
20 inches. Thus, while the resolving power is much greater, the dispersion
is less than before*.
In the course of the lecture the instrument was applied to show the
duplicity of the reversed soda lines. The interval on the screen between the
centres of the dark lines was about half an inch.
It is instructive to compare the action of the glass powder with that of
the spectroscope. In the latter the disposition of the prisms is regular, and
in passing from one edge of the beam to the other there is complete substitu-
tion of liquid for glass over the whole length. For one kind of light there is
no relative retardation ; and the resolving power depends upon the question
of what change of wave-length is required in order that its relative retardation
may be altered from zero to the quarter wave-length. All kinds of light for
which the relative retardation is less than this remain mixed. In the case
of the powder we have similar questions to consider. For one kind of light
the medium is optically homogeneous, i.e. the retardation is the same along
all rays. If we now suppose the quality of the light slightly varied, the
retardation is no longer precisely the same along all rays ; but if the variation
from the mean falls short of the quarter wave-length, it is without importance,
and the medium still behaves practically as if it were homogeneous. The
difference between the action of the powder and that of the regular prisms in
the spectroscope depends upon this, that in the latter there is complete
substitution of glass for liquid along the extreme rays, while in the former the
paths of all the rays lie partly through glass and partly through liquid in
nearly the same proportions. The difference of retardations along various
rays is thus a question of a deviation from an average.
* [1902. When carefully used this instrument gives about as good definition in the greeii
as a first-rate Rowland grating.]
1899] TRANSPARENCY AND OPACITY. 395
It is true that we may imagine a relative distribution of glass and liquid
that would more nearly assimilate the two cases. If, for example, the glass
consisted of equal spheres resting against one another in cubic order, some
rays might pass entirely through glass and others entirely through liquid,
and then the quarter wave-length of relative retardation would enter at the
same total thickness in both cases. But such an arrangement would be
highly unstable; and, if the spheres be packed in close order, the extreme
relative retardation would be much less. The latter arrangement, for which
exact results could readily be calculated, represents the glass powder more
nearly than does the cubic order.
A simplified problem, in which the element of chance is retained, may
be constructed by supposing the particles of glass replaced by thin parallel
discs which are distributed entirely at random over a certain stratum. We
may go further and imagine the discs limited to a particular plane. Each
disc is supposed to exercise a minute retarding influence on the light which
traverses it, and they are supposed to be so numerous that it is improbable
that a ray can pass the plane without encountering a large number. A
certain number (m) of encounters is more probable than any other, but if
every ray encountered the same number of discs, the retardation would be
uniform and lead to no disturbance.
It is a question of Probabilities to determine the chance of a prescribed
number of encounters, or of a prescribed deviation from the mean. In the
notation of the integral calculus the chance of the deviation from in lying
between ± r is*
where r = r/\/(2w). This is equal to '84 when r=l*0, or r=\f(2m); so
that the chance is comparatively small of a deviation from m exceeding
± V(2w).
To represent the glass powder occupying a stratum of 2 cm. thick, we may
perhaps suppose that m = 72. There would thus be a moderate chance of a
difference of retardations equal to, say, one-fifth of the extreme difference
corresponding to a substitution of glass for liquid throughout the whole
thickness. The range of wave-lengths in the light regularly transmitted by
the powder would thus be about five times the range of wave-lengths still
unseparated in a spectroscope of equal (2cm.) thickness. Of course, no
calculation of this kind can give more than a rough idea of the action of the
powder, whose disposition, though partly a matter of chance, is also influenced
by mechanical considerations ; but it appears, at any rate, that the character
* See Phil. Mag. 1899, Vol. XLVII. p. 251. [Vol. zv. p. 375.]
396 TRANSPARENCY AND OPACITY. [246
of the light regularly transmitted by the powder is such as may reasonably
be explained.
As regards the size of the grains of glass, it will be seen that as great or a
greater degree of purity may be obtained in a given thickness from coarse
grains as from fine ones, but the light not regularly transmitted is dispersed
through smaller angles. Here again the comparison with the regularly
disposed prisms of an actual spectroscope is useful.
At the close of the lecture the failure of transparency which arises from
the presence of particles small compared to the wave-length of light was
discussed. The tints of the setting sun were illustrated by passing the
light from the electric lamp through a liquid in which a precipitate of
sulphur was slowly forming*. The lecturer gave reasons for his opinion
that the blue of the sky is not wholly, or even principally, due to particles
of foreign matter. The molecules of air themselves are competent to dis-
perse a light not greatly inferior in brightness to that which we receive
from the sky.
* Op. cit. 1881, Vol. xn. p. 96. [Vol. i. p. 531.]
247.
ON THE TRANSMISSION OF LIGHT THROUGH AN ATMO-
SPHERE CONTAINING SMALL PARTICLES IN SUSPENSION,
AND ON THE ORIGIN OF THE BLUE OF THE SKY.
[Philosophical Magazine, XLVII. pp. 375—384, 1899.]
THIS subject has been treated in papers published many years ago*.
I resume it in order to examine more closely than hitherto the attenuation
undergone by the primary light on its passage through a medium containing
small particles, as dependent upon the number and size of the particles.
Closely connected with this is the interesting question whether the light
from the sky can be explained by diffraction from the molecules of air
themselves, or whether it is necessary to appeal to suspended particles
composed of foreign matter, solid or liquid. It will appear, I think, that
even in the absence of foreign particles we should still have a blue skyf.
The calculations of the present paper are not needed in order to explain
the general character of the effects produced. In the earliest of those above
* Phil. Mag. XLI. pp. 107, 274, 447 (1871); xn. p. 81 (1881). [Vol. i. pp. 87, 104, 518.]
f My attention was specially directed to this question a long while ago by Maxwell in a
letter which I may be pardoned for reproducing here. Under date Aug. 28, 1873, he wrote : —
"I have left your papers on the light of the sky, &c. at Cambridge, and it would take me, even
if I had them, some time to get them assimilated sufficiently to answer the following question,
which I think will involve less expense to the energy of the race if you stick the data into your
formula and send me the result....
" Suppose that there are N spheres of density p and diameter s in unit of volume of the
medium. Find the index of refraction of the compound medium and the coefficient of extinction
of light passing through it.
" The object of the enquiry is, of course, to obtain data about the size of the molecules of air.
Perhaps it may lead also to data involving the density of the aether. The following quantities
are known, being combinations of the three unknowns,
M=ms.ss of molecule of hydrogen ;
N= number of molecules of any gas in a cubic centimetre at 0° C. and 760 B.
s = diameter of molecule in any gas :—
398 ON THE TRANSMISSION OF LIGHT THROUGH AN [247
referred to I illustrated by curves the gradual reddening of the transmitted
light by which we see the sun a little before sunset. The same reasoning
proved, of course, that the spectrum of even a vertical sun is modified by the
atmosphere in the direction of favouring the waves of greater length.
For such a purpose as the present it makes little difference whether
we speak in terms of the electromagnetic theory or of the elastic solid
theory of light ; but to facilitate comparison with former papers on the light
from the sky, it will be convenient to follow the latter course. The small
particle of volume T is supposed to be small in all its dimensions in comparison
with the wave-length (X), and to be of optical density D' differing from that
(D) of the surrounding medium. Then, if the incident vibration be taken
as unity, the expression for the vibration scattered from the particle in a
direction making an angle 6 with that of primary vibration is
- irT . fa ,,, ,*
(6(-r)», .................. U)
r being the distance from T of any point along the secondary ray.
In order to find the whole emission of energy from T we have to integrate
the square of (1) over the surface of a sphere of radius r. The element
of area being far3 sin Odd, we have
r *™^ far- sin 0d0 = 47T f '"sin" OdO = ^ ;
Jo r* Jo o
o r o
so that the energy emitted from T is represented by
Known Combinations.
M N= density.
A/s2 from diffusion or viscosity.
Conjectural Combination.
• —3 = density of molecule.
" If you can give us (i) the quantity of light scattered in a given direction by a stratum of a
certain density and thickness ; (ii) the quantity cut out of the direct ray ; and (iii) the effect of
the molecules on the index of refraction, which I think ought to come out easily, we might get
a little more information about these little bodies.
" You will see by Nature, Aug. 14, 1873, that I make the diameter of molecules about j^Vu of
a wave-length.
" The enquiry into scattering must begin by accounting for the great observed transparency of
air. I suppose we have no numerical data about its absorption.
"But the index of refraction can be numerically determined, though the observation is of
a delicate kind, and a comparison of the result with the dynamical theory may lead to some new
information."
Subsequently he wrote, "Your letter of Nov. 17 quite accounts for the observed transparency
of any gas." So far as I remember, my argument was of a general character only.
* The factor TT was inadvertently omitted in the original memoir.
1899] ATMOSPHERE CONTAINING SMALL PARTICLES IN SUSPENSION. 399
on such a scale that the energy of the primary wave is unity per unit of
wave-front area.
The above relates to a single particle. If there be n similar particles per
unit volume, the energy emitted from a stratum of thickness dx and of unit
area is found from (2) by introduction of the factor ndx. Since there is
no waste of energy on the whole, this represents the loss of energy in the
primary wave. Accordingly, if E be the energy of the -primary wave,
1 dE 87rsn(D'-D)*T*
Edx = ~3 -- W~V> ..................... (3)
whence
where
E = Ene~hx,
8>rr3n(D'-D)*
-
(4)
If we had a sufficiently complete expression for the scattered light, we
might investigate (5) somewhat more directly by considering the resultant
of the primary vibration and of the secondary vibrations which travel in the
same direction. If, however, we apply this process to (1), we find that it
fails to lead us to (5), though it furnishes another result of interest. The
combination of the secondary waves which travel in the direction in question
has this peculiarity, that the phases are no more distributed at random.
The intensity of the secondary light is no longer to be arrived at by addition
of individual intensities, but must be calculated with consideration of the
particular phases involved. If we consider a number of particles which all
lie upon a primary ray, we see that the phases of the secondary vibrations
which issue along this line are all the same.
The actual calculation follows a similar course to that by which Huygens'
conception of the resolution of a wave into components
corresponding to the various parts of the wave-front
is usually verified. [See for example Vol. in. p. 74.]
Consider the particles which occupy a thin stratum dx
perpendicular to the primary ray x. Let AP (Fig. 1) be
this stratum and 0 the point where the vibration is to
be estimated. If AP = p, the element of volume is
dx.^Trpdp, and the number of particles to be found in
it is deduced by introduction of the factor n. Moreover,
if OP = r, A0 = x, r* = x* + p\ and pdp = rdr. The
resultant at 0 of all the secondary vibrations which issue
from the stratum dx is by (1), with sin 6 equal to unity,
ndx
or
c^D'-DirT Sir ,,,
. I — j; ---- — cos — - (bt —
J x U 7* A* A*
, D'-DirT . 27r/r
ndx.—j. --- — sm — (bt — x)
.L/ A* A*
(6)
400 ON THE TRANSMISSION OF LIGHT THROUGH AN [247
To this is to be added the expression for the primary wave itself, supposed
to advance undisturbed, viz., cos -^ (bt - x\ and the resultant will then
A.
represent the whole actual disturbance at 0 as modified by the particles
in the stratum da.
It appears, therefore, that to the order of approximation afforded by (1)
the effect of the particles in dec is to modify the phase, but not the intensity,
of the light which passes them. If this be represented by
cos ^ (fa- a: -S), (7)
8 is the retardation due to the particles, and we have
If fi be the refractive index of the medium as modified by the particles,
that of the original medium being taken as unity, 8 = (/u, — 1) dx, and
p.- 1 =nT(D' — D)/2D (9)
If ft denote the refractive index of the material composing the particles
regarded as continuous, D'/D = /*'*, and
reducing to
in the case where p! — 1 can be regarded as small.
It is only in the latter case that the formulae of the elastic-solid theory
are applicable to light. In the electric theory, to be preferred on every
ground except that of easy intelligibility, the results are more complicated
in that when (// — 1) is not small, the scattered ray depends upon the shape
and not merely upon the volume of the small obstacle. In the case of spheres
we are to replace (D' - D)/D by 3 (K' - K)/(K'+ 2K), where K, K' are
the dielectric constants proper to the medium and to the obstacle respectively*;
so that instead of (10)
onf //,2 — 1 , .
/*-! = "y^pq^ (12)
On the same suppositions (5) is replaced by
On either theory
* Phil. Mag. xn. p. 98 (1881). [Vol. i. p. 533.] For the corresponding theory in the case of
an ellipsoidal obstacle, see Phil. Map. Vol. xuv. p. 48 (1897). [Vol. iv. p. 305.]
1899] ATMOSPHERE CONTAINING SMALL PARTICLES IN SUSPENSION. 401
a formula giving the coefficient of transmission in terms of the refraction,
and of the number of particles per unit volume.
We have seen that when we attempt to find directly from (1) the effect
of the particles upon the transmitted primary wave, we succeed only so far
as regards the retardation. In order to determine the attenuation by this
process it would be necessary to supplement (1) by a terra involving
sin 2?r (6* - r)/\;
but this is of higher order of smallness. We could, however, reverse the
process and determine the small term in question a posteriori by means of
the value of the attenuation obtained indirectly from (1), at least as far as
concerns the secondary light emitted in the direction of the primary ray.
The theory of these effects may be illustrated by a completely worked
out case, such as that of a small rigid and fixed spherical obstacle (radius c)
upon which plane waves of sound impinge*. It would take too much space
to give full details here, but a few indications may be of use to a reader
desirous of pursuing the matter further.
The expressions for the terms of orders 0 and 1 in spherical harmonics of
the velocity-potential of the secondary disturbance are given in equations
(16), (17), § 334. With introduction of approximate values of 70 and 7^ viz.
70 + kc = %k?c3, 7x + kc = \ir
we get
[*.] + [*J = - ^ (l + y) cos k (at - r) + ^ (l - ^) sin k (at - r), . . .(15)f
in which c is the radius of the sphere, and k = 27T/X.. This corresponds to
the primary wave
[</>] = cos k (at + x), ........................... (16)
and includes the most important terms from all sources in the multipliers
of cos k (at - r), sin k (at — r). Along the course of the primary ray (JM = — 1)
it reduces to
~ r) ....... (17)
We have now to calculate by the method of Fresnel's zones the effect
of a distribution of n spheres per unit volume. We find, corresponding
to (6), for the effect of a layer of thickness dx,
2-rrndx {%kc* sin k (at + x) - ^JfcV cos k (at + x)} .......... (18)
* Theory of Sound, 2nd ed. § 334.
t [1902. n here denotes the sine of the latitude.]
B. IV. 26
402 ON THE TRANSMISSION OF LIGHT THROUGH AN [247
To this is to be added the expression (16) for the primary wave. The
coefficient of cos k (at + x) is thus altered by the particles in the layer dx
from unity to (1 — ^T^c^Trndx), and the coefficient of sink(at + x) from 0
to \k<?Trndx. Thus, if E be the energy of the primary wave,
dEj E = - ^kWirndx ;
so that if, as in (4), E=E0e~hx,
(19)
The same result may be obtained indirectly from the first term of (15).
For the whole energy emitted from one sphere may be reckoned as
(20)
unity representing the energy of the primary wave per unit area of wave-
front. From (20) we deduce the same value of h as in (19).
The first term of (18) gives the refractivity of the medium. If 8 be the
retardation due to the spheres of the stratum dx,
or & = %7rn(?dx ............................... (21)
Thus, if jj, be the refractive index as modified by the spheres, that of the
original medium being unity,
ip, ........................... (22)
where p denotes the (small) ratio of the volume occupied by the spheres
to the whole volume. This result agrees with equations formerly obtained
for the refractivity of a medium containing spherical obstacles disposed in
cubic order*.
Let us now inquire what degree of transparency of air is admitted by its
molecular constitution, i.e., in the absence of all foreign matter. We may
take X = 6 x 10~8 centim., p — 1 = '0003 ; whence from (14) we obtain as
the distance x, equal to I/ h, which light must travel in order to undergo
attenuation in the ratio e:I,
x = 4>-4> x 10~13 x n ............................ (23)
The completion of the calculation requires the value of n. Unfortunately
this number — according to Avogadro's law the same for all gases — can
hardly be regarded as known. Maxwell f estimates the number of molecules
under standard conditions as 19 x 1018 per cub. centim. If we use this value
of n, we find
x = 8'3 x 10" cm. = 83 kilometres,
* Phil. Mag. Vol. xxxiv. p. 499 (1892). [Vol. iv. p. 35.] Suppose m=o> , <r=cc .
t "Molecules," Nature, vni. p. 440 (1873).
1899] ATMOSPHERE CONTAINING SMALL PARTICLES IN SUSPENSION. 403
as the distance through which light must pass in air at atmospheric pres-
sure before its intensity is reduced in the ratio of 2*7 : 1.
Although Mount Everest appears fairly bright at 100 miles distance
as seen from the neighbourhood of Darjeeling, we cannot suppose that
the atmosphere is as transparent as is implied in the above numbers;
and of course this is not to be expected, since there is certainly suspended
matter to be reckoned with. Perhaps the best data for a comparison are
those afforded by the varying brightness of stars at various altitudes. Bouguer
and others estimate about '8 for the transmission of light through the entire
atmosphere from a star in the zenith. This corresponds to 8'3 kilometres
of air at standard pressure. At this rate the transmission through 83 kilo-
metres would be (-8)10, or '11, instead of l/e or '37. It appears then that
the actual transmission through 83 kilometres is only about 3 times less
than that calculated (with the above value of n) from molecular diffraction
without any allowance for foreign matter at all. And we may conclude
that the light scattered from the molecules would suffice to give us a blue
sky, not so very greatly darker than that actually enjoyed.
If n be regarded as altogether unknown, we may reverse our argument,
and we then arrive at the conclusion that n cannot be greatly less than
was estimated by Maxwell. A lower limit for n, say 7 x 1018 per cubic centi-
metre, is somewhat sharply indicated. For a still smaller value, or rather
the increased individual efficacy which according to the observed refraction
would be its accompaniment, must lead to a less degree of transparency than
is actually found. When we take into account the known presence of foreign
matter, we shall probably see no ground for any reduction of Maxwell's
number.
The results which we have obtained are based upon (14), and are as true
as the theories from which that equation was derived. In the electromagnetic
theory we have treated the molecules as spherical continuous bodies differing
from the rest of the medium merely in the value of their dielectric constant.
If we abandon the restriction as to sphericity, the results will be modified in
a manner that cannot be precisely defined until the shape is specified. On
the whole, however, it does not appear probable that this consideration would
greatly affect the calculation as to transparency, since the particles must be
supposed to be oriented in all directions indifferently. But the theoretical
conclusion that the light diffracted in a direction perpendicular to the primary
rays should be completely polarized may well be seriously disturbed. If the
view, suggested in the present paper, that a large part of the light from
the sky is diffracted from the molecules themselves, be correct, the observed
incomplete polarization at 90° from the Sun may be partly due to the
molecules behaving rather as elongated bodies with indifferent orientation
than as spheres of homogeneous material,
26—2
404 ON THE TRANSMISSION OF LIGHT THROUGH AN [247
Again, the suppositions upon which we have proceeded give no account
of dispersion. That the refraction of gases increases as the wave-length
diminishes is an observed fact ; and it is probable that the relation between
refraction and transparency expressed in (14) holds good for each wave-
length. If so, the falling off of transparency at the blue end of the spectrum
will be even more marked than according to the inverse fourth power of the
wave-length.
An interesting question arises as to whether (14) can be applied to
highly compressed gases and to liquids or solids. Since approximately
(p — 1) is proportional to n, so also is h according to (14). We have no
reason to suppose that the purest water is any more transparent than (14)
would indicate; but it is more than doubtful whether the calculations are
applicable to such a case, where the fundamental supposition, that the phases
are entirely at random, is violated. When the volume occupied by the
molecules is no longer very small compared with the whole volume, the fact
that two molecules cannot occupy the same space detracts from the random
character of the distribution. And when, as in liquids and solids, there is
some approach to a regular spacing, the scattered light must be much less
than upon a theory of random distribution.
Hitherto we have considered the case of obstacles small compared to the
wave-length. In conclusion it may not be inappropriate to make a few
remarks upon the opposite extreme case and to consider briefly the obstruction
presented, for example, by a shower of rain, where the diameters of the
drops are large multiples of the wave-length of light.
The full solution of the problem presented by spherical drops of water
would include the theory of the rainbow, and if practicable at all would be
a very complicated matter. But so far as the direct light is concerned, it
would seem to make little difference whether we have to do with a spherical
refracting drop, or with an opaque disk of the same diameter. Let us suppose
then that a large number of small disks are distributed at random over a
plane parallel to a wave-front, and let us consider their effect upon the direct
light at a great distance behind. The plane of the disks may be divided
into a system of Fresnel's zones, each of which will by hypothesis include
a large number of disks. If a be the area of each disk, and v the number
distributed per unit of area of the plane, the efficiency of each zone is
diminished in the ratio 1 : 1 — vet, and, so far as the direct wave is concerned,
this is the only effect. The amplitude of the direct wave is accordingly
reduced in the ratio 1 : 1 — va, or, if we denote the relative opaque area by ra,
in the ratio 1 : 1 — m*. A second operation of the same kind will reduce the
* The intensity of the direct wave is l-2m, and that of the scattered light m, making
altogether 1 — m.
1899] ATMOSPHERE CONTAINING SMALL PARTICLES IN SUSPENSION. 405
amplitude to (1 — ra)2, and so on. After x passages the amplitude is (1 — ra)*,
which if m be very small may be equated to e~mx. Here mx denotes the
whole opaque area passed, reckoned per unit area of wave-front; and it
would seem that the result is applicable to any sufficiently sparse random
distribution of obstacles.
It may be of interest to give a numerical example. If the unit of length
be the centimetre and x the distance travelled, m will denote the projected
area of the drops situated in one cubic centimetre. Suppose now that a is
the radius of a drop, and n the number of drops per cubic centimetre, then
m = WTra2. The distance required to reduce the amplitude in the ratio e : 1
is given by
a; = l/W7ra2.
Suppose that a = -^ centim., then the above-named reduction will occur
in a distance of one kilometre (x= 105) when n is about 10~3, i.e. when there
is about one drop of one millimetre diameter per litre.
It should be noticed that according to this theory a distant point of light
seen through a shower of rain ultimately becomes invisible, not by failure
of definition, but by loss of intensity either absolutely or relatively to the
scattered light.
248.
THE INTERFEROMETER.
[Nature, LIX. p. 533, 1899.]
THE questions raised by Mr Preston (Nature, March 23) can only be fully
answered by Prof. Michelson himself; but as one of the few who have used
the interferometer in observations involving high interference, I should
like to make a remark or two. My opportunity was due to the kindness
of Prof. Michelson, who some years ago left in my hands a small instrument
of his model.
I do not understand in what way the working is supposed to be prejudiced
by " diffraction." My experience certainly suggested nothing of the sort, and
I do not see why it is to be expected upon theoretical grounds.
The estimation of the "visibility" of the bands, and the deduction of
the structure of the spectrum line from the visibility curve, are no doubt
rather delicate matters. I have remarked upon a former occasion (Phil. Mag.
November, 1892)* that, strictly speaking, the structure cannot be deduced
from the visibility curve without an auxiliary assumption. But in the
application to radiation in a magnetic field the assumption of symmetry
would appear to be justified.
My observations were made with a modification of the original apparatus,
which it may be worth while briefly to describe. In order to increase the
retardation it is necessary to move backwards, parallel to itself, one of the
perpendicularly reflecting mirrors. Unless the ways upon which the sliding
piece travels are extremely true, this involves a troublesome readjustment
of the mirror after each change of distance. The difficulty is avoided by
the use of a fluid surface as reflector, which after each movement automatically
sets itself rigorously horizontal. If mercury be contained in a glass dish,
the depth must be considerable, and then the surface is inconveniently
mobile. A better plan is to use a thin layer standing on a piece of copper
plate carefully amalgamated. A screw movement for raising and lowering
the mercury reflector is still desirable, though not absolutely necessary.
* [Vol. iv. p. 15.]
249.
ON THE CALCULATION OF THE FREQUENCY OF VIBRATION
OF A SYSTEM IN ITS GRAVEST MODE, WITH AN
EXAMPLE FROM HYDRODYNAMICS.
[Philosophical Magazine, XLVII. pp. 566 — 572, 1899.]
WHEN the expressions for the kinetic (T) and potential (F) energy of a
system moving about a configuration of stable equilibrium are given, the
possible frequencies of vibration are determined by an algebraic equation
of degree (in the square of the frequency) equal to the number of independent
motions of which the system is capable. Thus in the case of a system whose
position is defined by two coordinates q1 and q2, we have
and if in a free vibration the coordinates are proportional to cos pt, the
determinantal equation is
A — V&J. R — rfM
= 0, (2)
C-p*N
(3)
And whatever be the number of coordinates, the possible frequencies are
given by a determinantal equation analogous to (2).
When the determinantal equation is fully expressed, the smallest root,
or indeed any other root, can be found by the ordinary processes of successive
approximation. In many of the most interesting cases, however, the number
of coordinates is infinite, and the inclusion of even a moderate number of
them in the expressions for T and V would lead to laborious calculations.
We may then avail ourselves of the following method of approximating to
the value of the smallest root.
408 ON THE CALCULATION OF THE FREQUENCY OF [249
The method is founded upon the principle* that the introduction of a
constraint can never lower, and must in general raise, the frequency of any
mode of a vibrating system. The first constraint that we impose is the
evanescence of one coordinate, say the last. The lowest frequency of the
system thus constrained is higher than the lowest frequency of the uncon-
strained system. Next impose as an additional constraint the evanescence
of the last coordinate but one. The lowest frequency is again raised. If
we continue this process until only one coordinate is left free to vary, we
obtain a series of continually increasing quantities as the lowest frequencies
of the various systems. Or, if we contemplate the operations in the reverse
order, we obtain a series of decreasing quantities ending in the precise
quantity sought. The first of the series, resulting from the sole variation
of the first coordinate, is given by an equation of the first degree, viz.
A — p*L = 0. The second is the lower root of the determinant (2) of the
second order. The third is the lowest root of a determinant of the third
order formed by the addition of one row and one column to (2), and so on.
This series of quantities may accordingly be regarded as successive approxi-
mations to the value required. Each is nearer than its predecessor to the
truth, and all (except of course the last itself) are too high.
The practical success of the method must depend upon the choice of
coordinates and of the order in which they are employed. The object is
so to arrange matters that the variation of the first two or three coordinates
shall allow a good approximation to the actual mode of vibration.
The example by which I propose to illustrate the method is one already
considered by Prof. Lamb. It is that of the transverse vibration of a liquid
mass, contained in a horizontal cylindrical vessel, and of such quantity that
the free surface contains the axis of the cylinder (r = 0). If we measure 9
vertically downwards, the fluid is limited by r = 0, r = c, and by 0 = — l^rr,
0=+ \TT. Between the above limits of 6 and when r = c the motion must
be exclusively tangential.
In the gravest mode of vibration the fluid swings from one side to the
other in such a manner that the horizontal motions are equal and the vertical
motions opposite at any two points which are images of one another in the
line 0 = 0. This relation, which holds also at the two halves of the free
surface, implies a stream-function ty which is symmetrical with respect
to 0 = 0.
Let ?/, denoting the elevation of the surface at a distance r from the
centre on the side for which d = \TT, be expressed by
/c)3-6^(r/c)5+...; ............ (4)
* Theory of Sound, §§ 88, 89. [See Vol. i. p. 170.]
1899] VIBRATION OF A SYSTEM IN ITS GRAVEST MODE. 409
then the potential energy for the whole mass (supposed to be of unit density)
is given by
° ...) ............. (5)
The more difficult part of the problem lies in determining the motion
and in the calculation of the kinetic energy. It may be solved by the
method of Sir G. Stokes, who treated a particular case, corresponding in
fact to our first approximation in which (4) reduces to its first term. It
is required to find the motion of an incompressible fluid in two dimensions
within the semicylinder, the normal velocity being zero over the whole of
the curved boundary (r = c, %TT > 0 > — ^TT) and over the flat boundary having
values prescribed by (4). If i|r be the stream-function, satisfying
the conditions are that ty shall be symmetrical with respect to 6 = 0, that it
be constant when r = c from 6 = 0 to 6 = |TT, and that when 6 = £TT,
= -2q, (r/c) + 4g4 (r/c)3
or ^/c = -q,(r/c)* + q4(r/c)*-q6(r/c)s + ............. (6)
At the edge, where r = c,
^/c = -q, + q4-qe-..., ........................ (7)
and this value must obtain also over the curved boundary.
The conditions may be satisfied* by assuming
ijr/c = q2 (r/c)2 cos 26 + qt (r/c)4 cos 4^ + ...
0, ..................... (8)
in which n = 0, 1, 2, &c. This form satisfies Laplace's equation and the
condition of symmetry since cosines of 6 alone occur. When B = ^TT, it
reduces to (6). It remains only to secure the reduction to (7) whenr = c,
and this can be effected by Fourier's method. It is required that from
0 = 0 to Q = \TT
^Am+l cos (2n + 1) 6 = - q2 (1 + cos 26) + q4 (1 - cos 40) - ....... (9)
It will be convenient to write
^+1 = q^ll + qiA^1 + ..., .................. (10)
so that
ZA^ cos (2n +1)0 = (-!)»- cos 2s0 ............. (11)
In (11) s may have the values 1, 2, 3, &c.
* Lamb's Hydrodynamics, § 72.
410 ON THE CALCULATION OF THE FREQUENCY OF [249
The values of the constants in (11) are to be found as usual. Since
2 I 'cos (2n + 1) 0 . cos (2m + 1) 6 dd
Jo
vanishes when m and n are different, and when m and n coincide has the
value £TT, and since
2 /"**{(_ l)o - cos 2s0] cos (2» + 1) 0 d6
we get
J<2«>_/ 1V+n
in which s = 1, 2, 3, &c., n = 0, 1, 2, &c.
The value of i/r in (8) is now completely determined when <?2, &c. are
known. The velocity-potential </> is deducible by merely writing sines, in
place of cosines, of the multiples of 6.
We have now to calculate the kinetic energy T of the motion thus
expressed, supposing for brevity that the density is unity. We have in
general
where dn is drawn normally outwards and the integration extends over the
whole contour. In the present case, however, d<j>/dn vanishes over the
circular boundary, so that the integration may be limited to the plane part.
Of this the two halves contribute equally. Now when 6 = \ir,
(14)
(15)
Thus
-..., ......... (16)
where Am+l is given by (10) and (12); it is of course a quadratic function
The summation with respect to n is easily effected in particular cases
by decomposition into partial fractions according to the general formula
(2n+2s+I)(2n + 2s' + I) 2(s-s')
1899] VIBRATION OF A SYSTEM IN ITS GRAVEST MODE. 411
If s' = — s, we have
1
(2n + 2s + 1) (2ra - 2s + 1)
4s \2n - 2s + 1 2n + 2s + ij
If s' = s, (17) fails, but we have by a known formula
/ ~ 8 32 52 (2s - I)2 '
Thus for the term in <j22, we have in (16)
in which by (18) 2 (2n + 3)-1 (2n - 1)'1 = 0,
by (17) 2 (2ra + 3)"1 (2n + I)"1 = £2 (2w + I)-1 - J2 (2n + 3)"1
11 \ 1/1 1
and by (19) 2(2n +3)~2= |wa-l.
The complete term (20) in q? is accordingly
*?<*-*+> <21>
The first approximation to p2 is therefore from (5), (21)
-..(22)
or p = T1690 (g/c)*, (23)
which is Prof. Lamb's result*.
For the second approximation we require also the terms in (16) which
involve q? and qzqt, and they are calculated as before. The term in <j42 is
2
_
TT V9 8 / '
The term in q2 qt is made up of two parts. Its complete value is
64c2 . /9_,
-9^** ............................... (25)
* Hydrodynamics, § 238.
412 VIBRATION OF A SYSTEM IN ITS GRAVEST MODE. [249
Thus
+™' ...... (26>
which with (5) gives materials for the second approximation. In proceeding
to this we may drop the symbols c and g, which can at any moment be
restored by consideration of dimensions. Also the factor 8 may be omitted
from the expressions for T and V. On this understanding we have by
comparison with (1),
44 B~\. 04
<-!-£- *--£• *-£-,.
or on introduction of the value of TT,
L = -2439204, M = - -2829420, N = -3463696.
The coefficients of the quadratic (3) are thence found to be
LN-M*= -00443040, AC-&= -0304762,
2MB ^LC-NA=- -0284860 ;
whence on restoration of the factor (0/c)*,
^ = 1-1644 (#/c)*, p2 = 2-2525 fo/c)*, ............ (2V)
the first of which constitutes the second approximation to the value of p
in cos pt, corresponding to the gravest mode of vibration. The small differ-
ence between (23) and (27) shows the success of the method and indicates
that (27) is but very little in excess of the truth.
If the result were of special importance it would be quite practicable
to take another step in the approximation, determining p* as the lowest root
of a cubic equation.
A question naturally suggests itself as to the significance of the value
of' p2 in (27). The general theory of constraints* shows that it may be
regarded as a first, but probably a rather rough, approximation to the
frequency of the second lowest mode of the complete system. Just as for
the gravest mode of all, the second lowest roots of the series of determinants
(of the 2nd, 3rd, and following orders) form successive approximations to
the true value, each value being lower and truer than its predecessor. The
second approximation would be the middle root of the cubic above mentioned.
But for this purpose it is doubtful whether the method is practical.
* Theory of Sound, 2nd ed. § 92 a.
250.
THE THEORY OF ANOMALOUS DISPERSION.
[Philosophical Magazine, XLVIII. pp. 151, 152, 1899.]
I HAVE lately discovered that Maxwell, earlier than Sellmeier or any
other writer, had considered this question. His results are given in the
Mathematical Tripos Examination for 1869 (see Cambridge Calendar for
that year). In the paper for Jan. 21, l|h— 4h, Question IX. is :—
" Show from dynamical principles that if the elasticity of a medium be
such that a tangential displacement 77 (in the direction of y) of one surface
of a stratum of thickness a calls into action a force of restitution equal
to Ei) / a per unit of area, then the equation of propagation of such displace-
ments is
"Suppose that every part of this medium is connected with an atom
of other matter by an attractive force varying as distance, and that there
is also a force of resistance between the medium and the atoms varying
as their relative velocity, the atoms being independent of each other ; show
that the equations of propagation of waves in this compound medium are
where p and a are the quantities of the medium and of the atoms respectively
in unit of volume, y is the displacement of the medium, and 77 + f that
of the atoms, <rp2£ is the attraction, and <rRd£/dt is the resistance to the
relative motion per unit of volume.
414 THE THEORY OF ANOMALOUS DISPERSION. [250
" If one term of the value of 77 be Ge~xl1 cos n (t — xjv), show that
1 1 _ p + <r av? p1 — n2
& + Vn*~~E~* W (pi-tf
2 <rtf En
" If <r be very small, one of the values of tf will be less than E/p, and
if R be very small v will diminish as n increases, except when n is nearly
equal to p, and in the last case I will have its lowest values. Assuming
these results, interpret them in the language of the undulatory theory of
light."
If we suppose that R = 0,
L = £+ <L P*
v2 E E p*-n?'
and
v2 p p*-n2'
if v0 be the velocity corresponding to a- = 0.
251.
INVESTIGATIONS IN CAPILLARITY :— THE SIZE OF DROPS.—
THE LIBERATION OF GAS FROM SUPERSATURATED SOLU-
TIONS. — COLLIDING JETS. — THE TENSION OF CONTA-
MINATED WATER-SURFACES.— A CURIOUS OBSERVATION.
[Philosophical Magazine, XLVIII. pp. 321—337, 1899.]
The Size of Drops.
THE relation between the diameter of a tube and the weight of the drop
which it delivers appears to have been first investigated by Tate*, whose
experiments led him to the conclusion that " other things being the same,
the weight of a drop of liquid is proportional to the diameter of the tube
in which it is formed." Sufficient time must of course be allowed for the
formation of the drops ; otherwise no simple results can be expected. In
Tate's experiments the period was never less than 40 seconds.
The magnitude of a drop delivered from a tube, even when the formation
up to the phase of instability is infinitely slow, cannot be calculated a priori.
The weight is sometimes equated to the product of the capillary tension (T)
and the circumference of the tube (27ra), but with little justification. Even
if the tension at the circumference of the tube acted vertically, and the whole
of the liquid below this level passed into the drop, the calculation would still
be vitiated by the assumption that the internal pressure at the level in
question is atmospheric. It would be necessary to consider the curvatures
of the fluid surface at the edge of attachment. If the surface could be
treated as a cylindrical prolongation of the tube (radius a), the pressure would
be T/a, and the resulting force acting downwards upon the drop would
amount to one-half (jraT) of the direct upward pull of the tension along the
circumference. At this rate the drop would be but one-half of that above
* Phil. Mag. Vol. xxvn. p. 176 (1864).
416 INVESTIGATIONS IN CAPILLARITY. [251
reckoned. But the truth is that a complete solution of the statical problem
for all forms up to that at which instability sets in, would not suffice for the
present purpose, The detachment of the drop is a dynamical effect, and
it is influenced by collateral circumstances. For example, the bore of the
tube is no longer a matter of indifference, even though the attachment of
the drop occurs entirely at the outer edge. It will appear presently that
when the external diameter exceeds a certain value, the weight of a drop
of water is sensibly different in the two extreme cases of a very small and of
a very large bore.
But although a complete solution of the dynamical problem is im-
practicable, much interesting information may be obtained from the principle
of dynamical similarity. The argument has already been applied by Dupre
(Theorie Mecanique de la Chaleur, Paris, 1869, p. 328), but his presentation
of it is rather obscure. We will assume that when, as in most cases, viscosity
may be neglected, the mass (M) of a drop depends only upon the density (cr),
the capillary tension (T), the acceleration of gravity (g), and the linear
dimension of the tube (a). In order to justify this assumption, the form-
ation of the drop must be sufficiently slow, and certain restrictions must be
imposed upon the shape of the tube. For example, in the case of water
delivered from a glass tube, which is cut off square and held vertically, a will
be the external radius; and it will be necessary to suppose that the ratio
of the internal radius to a is constant, the cases of a ratio infinitely small, or
infinitely near unity, being included. But if the fluid be mercury, the flat
end of the tube remains unwetted, and the formation of the drop depends
upon the internal diameter only.
The " dimensions " of the quantities on which M depends are : —
o- = (Mass)1 (Length)"3,
T = (Force)1 (Length)-1 = (Mass)1 (Time)-2,
g = Acceleration = (Length)1 (Time)"2,
of which M, a mass, is to be expressed as a function. If we assume
M oc Tx . gv . <TZ . au,
we have, considering in turn length, time, and mass,
y - 32 + u = 0, 2# + 2y = 0,
so that y=. — x, z = 1 — x, u = 3 —
Ta ( T \x~l
Accordingly M « — ( — .
1899] INVESTIGATIONS IN CAPILLARITY. 417
Since as is undetermined, all that we can conclude is that M is of the -form
9
where F denotes an arbitrary function.
Dynamical similarity requires that T/ga-a? be constant ; or, if g be sup-
posed to be so, that a2 varies as T/<r. If this condition be satisfied, the mass
(or weight) of the drop is proportional to T and to a.
If Tate's law be true, that cwteris paribus M varies as a, it follows from
(1) that F is constant. For all fluids and for all similar tubes similarly
wetted, the weight of a drop would then be proportional not only to the
diameter of the tube but also to the superficial tension, and it would be
independent of the density.
In order to examine how far Tate's law can be relied upon, I have thought
it desirable, with the assistance of Mr Gordon, to institute fresh experiments
with water, in which necessary precautions were observed, especially against
the presence of grease. Attention has been given principally to the two
extreme cases, (i) when the wall of the tube is thin, so that the external
and internal diameters of the tube are nearly equal; (ii) when the bore
is small in comparison with the external diameter. The event showed that
up to an external diameter of one centimetre or more, the size of the bore
is of little consequence, but that for larger diameters the weight of the drop
in (ii) is sensibly less than in (i). It scarcely needs to be pointed out that
in (i) the diameter can only be increased up to a certain limit, after which
the tube would not remain full. In (ii) the diameter can be increased to any
extent, but the drop falling from it reaches a limit. The experiments of Tate
extended also to case (ii), but his results are, I believe, erroneous. For
a diameter of one-half an inch (1'27 cm.) he found for the two cases drops in
the ratio of T56 : 2'84.
In my experiments the thin-walled tubes were of glass, the ends being
ground to a plane, and carefully levelled. Ten drops, following one another
at intervals of about 50 seconds, were usually weighed together. As to the
interval, sufficient time must be allowed for the normal formation of the drop,
but the fact that evaporation is usually in progress forbids too great a pro-
longation. The accuracy attained was not so great as had been hoped for.
Successive collections, made without disturbance, gave indeed closely accord-
ant weights (often to one-thousandth part), but repetitions after cleaning
and remounting indicated discrepancies amounting to one-half per cent., or
even to one per cent. The cause of these minor variations has not been fully
traced; but the results recorded, being the mean of several experiments,
must be free from serious error. Attention may be called to tubes 11 and 12
of nearly the same (external) diameter. Of these 11 was plugged so as to
R. iv. 27
418
INVESTIGATIONS IN CAPILLARITY.
[251
It will be seen
leave only a small bore, the end being carefully ground flat,
that the difference in the weights of the drops was but small.
Again, No. 10 was of barometer- tubing, having a comparatively small
bore, which accounts for the slightly diminished weight of the drop. The
other tubes were thin-walled. In all cases care was taken that the cylindrical
part of the tube, though clean, should remain unwetted, a condition which
precluded the use of diameters much less than those recorded.
Glass
Metal
1
•088 -0375
15
•400
•1446
2
•134
•0526
16
•450
•1662
3
•191
•0712
17
•500
•1882
4
•200
•0755
18
•530
•2023
5
•256
•0923
19
•550
•2130
'
6
•354
•1151
20
•559
•2167
7
•383
•1362
21
•580
•2256
8
•406
•1461 ; 22
•597
•2295
9
•459
•1703
23
•621
•2389
10
•465
•1698
24
•640
•2454
11
•521
•1969
25
•680
•2510
12
•523
•2023
26
•730
•2531
13
•566
•2210
27
•800
•2509
14
•584
•2339
Fig. 1.
The numbers in the second column are the external diameters measured
in inches (one inch = 2'54 cm.), while the third column gives the weight
in grams of a single drop, corrected for temperature to 15° C., upon the
supposition (corresponding to Tate's law) that the weight is proportional to
surface-tension.
The entries under the heading " Metal " relate to experiments in which
the glass tubes were replaced by metal disks, bored cen-
trally and turned true in the lathe. The water was supplied
from above through a metal tube soldered to the back
(upper) face of the disk (Fig. 1). At the time of use only
the lower face was wetted.
A plot of both sets of numbers is shown in the Figure (2).
The two curves practically coincide up to diameters of
about '4 inch, after which that corresponding to the disks
falls below. The lower curve shows some irregularities,
especially in the region of diameters equal to "6 inch.
These appear to be genuine ; they may originate in a sort of reflexion from
1899]
INVESTIGATIONS IN CAPILLARITY.
419
the circumference of the disk of the disturbance caused by the breaking away
of the drop. It is possible that at this stage the phenomenon is sensibly
influenced by fluid viscosity.
Fig. 2.
That the size of the bore should be of secondary importance is easily
understood. Up to the phase of instability, the phenomenon is merely
a statical one, and the element of the size of the bore does not enter. It is
only the rapid motion which occurs during the separation of the drop that
could be influenced. When the diameter is moderate, the most rapid motions
occur at a level considerably below the tube, and the obstruction presented
by the flat face of a thick- walled tube is unimportant.
The observations give materials for the determination of the function F
in (1). In the following table, applicable to thin-walled tubes, the first
column gives values of Tjgcr^, and the second column those of gM/Ta, all
the quantities concerned being in c.G.S. measure, or other consistent system.
Tlgoa?
gM/Ta
2-58
4-13
tie
3-97
•708
3-80
•441
3-73
•277
3-78
•220
3-90
•169
4-06
27—2
420 INVESTIGATIONS IN CAPILLARITY. [251
From this the weight of a drop of any liquid of which the density and the
surface-tension are known can be calculated. For many purposes it may
suffice to treat F as constant, say 3'8. The formula for the weight of a drop
is then simply
.................................... (2)
in which 3'8 replaces the 2?r of the faulty theory alluded to earlier.
The Liberation of Gas from Supersaturated Solutions.
The formation of bubbles upon the sides of a vessel containing " soda-
water" or a gas-free liquid heated above its boiling-point, is a subject
upon which there has been much difference of opinion. In one view, ably
advocated by Gernez, the nucleus is invariably gaseous. That a small volume
of gas, visible or invisible, provided that its dimensions exceed molecular
distances, must act in this way is certain, and the activity of porous solids is
thus naturally and easily explained. But Gernez goes much further, and
holds that the activity of glass or metal rods, immersed in the liquid without
precaution, is of the same nature, and to be attributed to the film of air
which all bodies acquire when left for some time in contact with the atmo-
sphere. If a body is rendered inactive by prolonged standing in cold water ;
by treatment with alcohol, ether, &c., "qui dissolvent les gaz de 1'air, plus
abondamment que 1'eau * " ; or by heating in a flame ; it is because by such
processes the film of air is removed. One cannot but sympathise with
Tomlinson-f* in his repugnance to such an explanation ; but the position main-
tained by the latter, that activity is due to contamination with grease, is also
not without its difficulties.
The question whether contact with air suffices to restore the activity of
a piece of glass or metal that has been rendered inactive by heat or otherwise,
appears to be amenable to experiment, and should not remain an open one.
In 1892 I had a number of glass tubes prepared of about 1 cm. diameter
for experiments in this direction. After a thorough heating in the blowpipe-
flame, the ends of the tubes were hermetically sealed. At intervals since
that date some of the tubes have been opened and compared with others
which had undergone no preparation. Short lengths of rubber provided with
pinch-cocks are fitted to the upper ends, by means of which aerated water
is easily drawn in from a shallow vessel. Three tubes remaining over from
the batch above mentioned were tried a few weeks ago, and establish the
conclusion that seven years contact with air fails to restore activity. A similar
experiment may be made with iron wires. If these be heated and sealed up
in glass tubes, they remain inactive, but exposure to the air of the laboratory
for a day or two restores activity.
* Annales de VEcole Normak, p. 319, 1875.
t Phil. May. Vol. xux. p. 305 (1875).
1899] INVESTIGATIONS IN CAPILLARITY. 421
In opposition to the contention that grease is the primary cause of
activity, Gernez brings forward a striking experiment from which it appears
that a drop of olive-oil itself liberates no gas when introduced with pre-
caution. " Quant au r61e que jouent les corps gras, il est facile de s'en rendre
compte: lorsqu'on frotte un corps quelconque entre les doigts legerement
graisses, on produit a sa surface une se*rie d'eminences line'aires se"pare"es par
les sillons qui correspondent aux lignes de 1'epiderme ; les cavity's forment un
reseau de conduits qui contiennent de 1'air, sont difficilement mouilles par
1'eau et, par consequent, constituent au sein du liquide une atmosphere
eminemment favorable au de*gagement des gaz*."
It seems to me that Tomlinson was substantially correct in attributing
the activity of a non-porous surface to imperfect adhesion. We have to
consider in detail the course of events when a surface, e.g. of glass, is intro-
duced into the liquid. If the surface be clean, it is wetted by the water
advancing over it, whether there be a film -of air condensed upon it or not,
and no gas is liberated from the liquid. But if the surface be greasy, even
in a very slight degree, the behaviour is different. We know that a drop
of water is reluctant to spread over a glass that is not scrupulously clean.
If a large quantity of water be employed, some sort of spreading follows
under the influence of gravity, but there is no proper adhesion, at least for
a time, as appears at once on pouring the water off again. The precise
character of the transition from glass to water when there is grease between
is not well understood. It may be that there is something which can fairly
be called a film of air. If so, its existence is a consequence of the presence
of the grease. On the other hand, it appears at least equally probable that
air is not concerned, and that the activity of the surface is directly due to the
thin film of grease, whose properties, as in the case of greased water surfaces,
are materially different from those of a thick layer.
On this principle, too, it is easier to understand the retention of a visible
bubble when formed — a retention which often lasts for a long time. So
soon as the gas is entirely surrounded by liquid of thickness exceeding the
capillary limit, the bubble is bound to rise. It is difficult to see how the
hypothetical film of air explains the failure of the liquid to penetrate between
the bubble and the solid.
Colliding Jets.
In various papers (Proc. Roy. Soc. Feb. 1879, May 1879, June 1882)
[Vol. I. pp. 372, 377 ; Vol. n. p. 103] I have examined the behaviour of
colliding drops and jets. Experiments with drops are very simply carried
out by the observation of nearly vertical fountains, rising say to two feet
from nozzles ^ inch in diameter. The scattering of the drops, when the
* Loc. cit. p. 346.
422 INVESTIGATIONS IN CAPILLARITY. [251
water is clean and not acted upon by electricity, shows that collision is
followed by rebound. If the water is milky, or soapy with unclarified soap,
or if the jet, though clean, is under the influence of feeble electricity, the
apparent coherence and the heaviness of the patter made by the falling water
are evidence that rebound no longer ensues, but that collision results in
amalgamation. Eye observation, or photography, with the instantaneous
illumination of electric sparks renders the course of events perfectly clear.
[1902. The annexed illustrations are from instantaneous photographs of the
same fountain with and without electrical influence.]
The form of the experiment in which are employed jets, issuing at
moderate velocities and meeting at high obliquities, is the more instructive ;
but it is liable to be troublesome in consequence of the tendency of the jets
to unite spontaneously. It is important to avoid dust both in the water and
in the atmosphere where the collision occurs. An electromotive force of one
volt suffices to determine union ; but so long as the jets rebound there is
complete electrical insulation between them.
As to the manner in which electricity acts, two views were suggested. It
was thought probable that union was the result of actual discharge across the
thin layer of intervening insulation; but it was also pointed out that the
result might be due to the augmented pressure to be expected from the
electrical charges upon the opposed surfaces. From observations upon the
colours of thin plates exhibited at the region of contact, which he found
to be undisturbed by such electrical forces as would not produce union,
Mr Newall* concluded that the second of the above-mentioned explanations
must be discarded.
On the other hand, as has been pointed out by Kaiser f, the progress of
knowledge concerning electrical discharge has rendered the first explanation
more difficult of acceptance. It would appear that some hundreds of volts
are needed in order to start a spark, and that mere diminution of the interval
to be crossed would not compensate for want of electromotive force.
A more attentive examination of the conditions of the experiment may
perhaps remove some of the difficulties which seem to stand in the way of
the second explanation. As the liquid masses approach one another, the
intervening air has to be squeezed out. In the earlier stages of approxima-
tion the obstacle thus arising may not be important ; but when the thickness
of the layer of air is reduced to the point at which the colours of thin plates
are visible, the approximation must be sensibly resisted by the viscosity of
the air which still remains to be got rid of. No change in the capillary
* Phil. Mag. Vol. xx. p. 33 (1885).
t Wied. Ann. LIII. p. 667 (1894). Kaiser's own experiments were made upon the modification
of the phenomenon observed by Boys, where the contact takes place between two soap-films.
1S99] INVESTIGATIONS IN CAPILLARITY. 423
In natural condition.
424 INVESTIGATIONS IN CAPILLARITY. [251
conditions can arise until the interval is reduced to a small fraction of a wave-
length of light; but such a reduction, unless extremely local, is strongly
opposed by the remaining air. It is of course true that this opposition is
temporary. The question is whether the air can be anywhere squeezed out
during the short time over which the collision extends.
It would seem that the electrical forces act with peculiar advantage.
If we suppose that upon the whole the air cannot be removed, so that the
mean distance between the opposed surfaces remains constant, the electric
attractions tend to produce an instability whereby the smaller intervals are
diminished while the larger are increased. Extremely local contacts of the
liquids, while opposed by capillary tension which tends to keep the surfaces
flat, are thus favoured by the electrical forces, which moreover at the small
distances in question act with exaggerated power.
It is probably by promoting local approximations in opposition to capillary
forces that dust, finding its way to the surfaces, brings about union.
A question remains as to the mode of action of milk or soapy turbidity.
The observation, formerly recorded, that it is possible for soap to be in excess
may here have significance. It would seem that the surfaces, coming into
collision within a fraction of a second of their birth, would still be subject
to further contamination from the interior. A particle of soap rising acci-
dentally to the surface would spread itself with rapidity. Now such an
outward movement of the liquid is just what is required to hasten the
removal of the intervening air. It is obvious that this effect would fail if
the contamination of the surface had proceeded too far previously to the
collision.
In order to illustrate the importance of the part played by the intervening
gas, I thought that it would be interesting to compare the behaviour of the
jets when situated in atmospheres of different gases. It seemed that gases
more freely soluble in water than the atmospheric gases would be more easily
got rid of in the later stages of the collision, and that thus union might more
readily be brought about. This expectation has been confirmed in trials
made on several different occasions. It was found sufficient to allow a pretty
strong stream of the gas under examination to play upon the jets at and
above the place of collision. Jets of air, of oxygen, and of coal-gas were
found to be without effect. On the other hand, carbonic acid, nitrous oxide,
sulphurous anhydride, and steam at once caused union. Only in the case
of hydrogen was there an ambiguity. On some occasions the hydrogen
appeared to be without effect, but on others (when perhaps the pressure of
collision was higher) union uniformly followed. Care was taken to verify
that air blown through the same tube as had supplied the hydrogen was
inactive, so that the effect of the hydrogen could not be attributed to dust.
1899] INVESTIGATIONS IN CAPILLARITY. 425
The action of hydrogen cannot be explained by its solubility. Hydrogen is,
however, much less viscous than other gases, and to this we may plausibly
attribute its activity in promoting union. A layer of hydrogen may be
effectively squeezed out in a time that would be insufficient in the case of
air and oxygen.
The Tension of Contaminated Water-Surfaces.
In my experiments upon the superficial viscosity of water (Proc. Roy. Soc.
June 1890) [Vol. ill. p. 375] I had occasion to notice that the last traces
of residual contamination had very little influence upon the surface-tension,
but that they became apparent when compressed in front of the vibrating
needle of Plateau's apparatus. Subsequently I showed (Phil. Mag. Vol.
xxxni. p. 470, 1892) [Vol. in. p. 572] that according to Laplace's theory
of Capillarity, in which matter is regarded as continuous, the effect of a thin
surface-film in diminishing the tension of pure water should be as the
square of the thickness of the film.
The tension of slightly contaminated surfaces was made the subject
of special experiments by Miss Pockels (Nature, Vol. XLIII. p. 437, 1891), who
concluded that a water-surface can " exist in two sharply contrasted conditions;
the normal condition, in which the displacement of the partition [altering
the density of the contamination] makes no impression upon the tension,
and the anomalous condition, in which every increase or decrease alters
the tension." It is only since I have myself made experiments upon the
same lines that I have appreciated the full significance of Miss Pockels'
statement. The conclusion that, judged by surface-tension, the effect of
contamination comes on suddenly, seems to be of considerable importance,
and I propose to illustrate it further by actual curves embodying results
recently obtained.
The water is contained in a trough modelled after that of Miss Pockels.
It is of tin-plate, 70 cm. long, 10 cm. broad, and 2 cm. deep, and it is filled
nearly to the brim. The partitions, by which the oil is confined, are made
of strips of glass resting upon the edge of the trough in such a manner that
their lower surfaces are wetted while the upper surfaces remain dry. The
strips may be 1£ cm. wide, and for convenience of handling their length
should exceed considerably the width of the trough. I have found advantage
in cementing (with hard cement) slight webs of glass to the lower faces. The
length of these is a rough fit with the width of the trough, enabling them to
serve as guides preventing motion of the strips parallel to their length.
In order to observe the surface-tension Miss Pockels used a small disk
(6 mm. in diameter) in contact with the surface, measuring the force necessary
to detach it. In my own experiments I have employed the method of
426 INVESTIGATIONS IN CAPILLARITY. [251
Wilhelmy, which appears to be better adapted to the purpose. A thin
blade is mounted in a balance, its plane being vertical and its lower horizontal
edge dipping under the surface of the liquid. If absolute measures are
7-equired, the edge of the blade should lie at the general level of the surface
when the pointer of the balance stands at zero. If m be the mass in the
other pan needed to compensate the effect of the liquid, I the length of
the blade, the surface-tension (T) may be deduced from the equation
2lT=m(/ (1)
When only differences of tension are concerned, the precise level of the strip
is of no consequence. As regards material, glass is to be preferred and it
should be thin in order not unduly to diminish the sensitiveness of the
balance by the displacement of water. I have used a small frame carrying
three parallel blades, the total length being 27 cm., while the thickness may
be considered nearly negligible. Before use the glass is cleaned with strong
sulphuric acid, and the angle of contact with the water when the balance is
raised appears to be zero. The total value of m for a clean surface may then
be calculated from (1), taking T at 74. We find m = 4'1 gms. The balance
could be read without difficulty to '01 gm., giving abundant accuracy.
The position of the barrier, giving the length of the surface to which the
grease is confined, is measured by a millimetre-scale, but is subject to a
correction needed in order to take account of the additional surface operative
when the suspended strip is raised. This amounts to about 3 cm., and is
to be added to the measured length. In a set of experiments where the
grease is successfully confined, the density is proportional to the reciprocal
of the above corrected length. It sometimes happens that continuity is
lost by the passage of grease across the barrier. This is of course most
likely to happen when the tensions on the two sides differ considerably,
and the danger may be mitigated by the use of a second barrier, so manipu-
lated that the densities are nearly the same on the two sides of the principal
barrier.
In commencing a set of observations the first step is to secure the
cleanness of the surface. To this end the surface is scraped, if the expression
may be allowed, along the whole length by one of the movable partitions,
and, if thought necessary, the accumulated grease at the far end may be
removed with strips of paper. The operation should be repeated two or
three times with intermediate insertion of the balanced strips until it is
certain that no grease remains, competent to affect the tension even when
concentrated. The weights now necessary to bring the pointer to zero give
the standard with which the contaminated surfaces are to be compared.
If it be desired to begin with small contaminations, it is best to contract
the area, say to about one-half the maximum, and then to apply the grease
1899]
INVESTIGATIONS IN CAPILLAEITY.
427
under examination with a previously ignited platinum wire until a small
effect, such as '02 gm., is observed at the balance. If the surface be now
extended to the maximum, the attenuated grease will have lost its power,
and the original reading for clean surfaces will be recovered. The barrier
may now be advanced, readings being taken at intervals as the grease is
concentrated. It is often more convenient to make the final adjustment by
moving the barrier rather than by correcting the weights.
An example will make manifest at once the character of the results
obtained. On May 15, the weight for the clean surface being 1-65 gm.,
the water was greased with castor-oil. With the barrier at 63 cm. this
grease had no effect. The corrected length is 66, and the reciprocal of this,
viz. 152, represents (for this series of observations) the density of the oil.
With the barrier at 40, viz. at density 233, there was no change of the order
of '005 gm. At 36 cm., or density 256, the oil had just begun to show itself
distinctly, the weight being then T64. At density 278 the weight became
1'62. From this point onwards increase of density tells rapidly. At 308 the
weight was I' 55, and at 334 the weight was 1-40. A plot of these results
is given in Fig. 3, and brings out more vividly than any description the
striking character of the law discovered by Miss Pockels.
The effect of concentration beyond 571, giving *70 gm., could not be
examined in the same series. It was necessary to add more oil, and then
of course the reciprocals of the corrected lengths represent the densities
on a different scale from before. Corresponding to 63 cm., of which the
reciprocal is 159, the weight was now T20 gm., falling to TOO at 175, '80
at 204, -70 at 233, '60 at 351, '55 at 488, and finally '52 at 625. These
values are plotted in Fig. 4, and they show that from a certain density
onwards the tension falls very slowly. This curve may be continued
Figs. 3-6.
0 133456
-/erf*
428 INVESTIGATIONS IN CAPILLARITY. [251
backwards by means of the results of Fig. 3, for of course the densities
corresponding to any particular weight, e.g. 1'20 gm., are really the same in
the two series.
It is of interest to inquire what point on these curves corresponds to the
deadening of the movements of small particles of camphor deposited upon the
surface. On a former occasion I have shown (Phil. Mag. Vol. xxxm. p. 366,
1892) [Vol. in. p. 565] that whatever may be the character of the grease the
cessation of the movements indicates that the tension falls short of a particular
value. In the present method of experimenting there is no difficulty in deter-
mining what for brevity may be called the camphor-point. Two precautions
should, however, be observed. It is desirable not to try the camphor until near
the close of a set of experiments, and then to avoid too great a quantity. It
would seem that the addition of camphor may sometimes lower the tension
below the point due to the grease. The second precaution required is the
raising of the balanced strip ; otherwise when a weight is taken the density
of the grease is altered. In several trials with castor and other oils the
camphor-point 'was found to correspond with a drop of tension from that
of clean water amounting to '9 gm. The points thus fixed are marked in
Figs. (3) and (4) with the letter G.
At this stage a certain discrepancy from former results should be remarked
upon. Working by the method of ripples I had concluded that the camphor-
point corresponded to a tension '72 of that of pure water, i.e. to a drop of 28
per cent. But the '9 gm. is only 22 per cent, of the calculated weight for
pure water, i.e. 4*1 gms. At this rate the 72 per cent, would become 78 per
cent., and the difference seems larger than can well be explained as an
alteration of standard in judging when the fragments are nearly dead.
One of the most striking conclusions to be drawn from an inspection
of the curves is the slowness of the fall of tension which sets in soon after
passing the camphor-point. On a rough view it would seem as if a second
limit were being approached. But this idea is scarcely confirmed by actual
further additions of oil, for the tension continues to fall slightly after each
addition, even when large quantities are already present. But there is one
peculiarity in the behaviour of the oil which suggests that the failure to
reach a limit may be due to want of homogeneity. As is well known, the
disk into which a drop deposited upon an already oiled surface at first spreads,
soon breaks up, and the superfluous oil collects itself into little lenses. After
this stage is reached it would be natural to suppose that the affinity of the
surface for oil was fully satisfied, and that no further alteration in tension
could occur. And in fact the balance usually indicated the absence of
immediate effect. But if the surface were expanded so as to spread the
added oil more effectively and then contracted again, a fall in tension was
almost always observed. It would seem as if the surface still retained an
1899]
INVESTIGATIONS IN CAPILLARITY.
affinity for some minor ingredient capable of being extracted, though satiated
as regards the principal ingredient.
The comparison of the present with former results throws an interesting
light upon molecular magnitudes. It has been shown (Proc. Roy. Soc. March
1890) [Vol. in. p. 347] that the thickness of the film of olive-oil, calculated
as if continuous, which corresponds to the camphor-point, is about 2'0 /*/*,*;
while from the present curves it follows that the point at which the tension
begins to fall is about half as much, or TO /A/Z. Now this is only a moderate
multiple of the supposed diameter of a gaseous molecule, and perhaps scarcely
exceeds at all the diameter to be attributed to a molecule of oil. It is obvious
therefore that the present phenomena lie entirely outside the scope of a theory
such as Laplace's, in which matter is regarded as continuous, and that an
explanation requires a direct consideration of molecules.
If we begin by supposing the number of molecules of oil upon a water
surface to be small enough, not only will every molecule be able to approach
the water as closely as it desires, but any repulsion between molecules will
have exhausted itself. Under these conditions there is nothing to oppose
the contraction of the surface — the tension is the same as that of pure water.
Castor Oil.
May 15.
Fig. (3)
Castor Oil.
May 15.
Fig. (4)
Olive Oil.
May 3.
Fig. (5)
Cod Liver Oil.
May 11.
Fig. (6)
Density
Weight
(in grams)
Density
Weight
(in grams)
, Density
Weight
(in grams)
Density
Weight
(in grams)
0
1-65
0 obs.
1-65
0
8-45
0
8-28
152
1-65
98 calc. 1-65
159
8-45
77
8-27
213
1-65
108 „ 1-64
324
8-30
113
8-25
233
1-65
117 „ 1-62
350
8-20
125
8-20
256
1-64
122 „ 1-60
376
8-10
137
8-10
278
1-62
130 „ 1-55
405
8-00
147
8-00
290
1-60
136 „ 1-50
430
7-90
154
7-90
308
1-55
141 „
1-40
461
7'80
171
7-70
323
334
1-50
1-40
148 calc.
159 obs.
1-30
1-20
483
518
7'70
7-60
213
303
7'60
7-50
351
1-30
175 „
1-00
392
7-45
377
1-20
204 „
•80
465
7-40
408
1-10
233 „
•70
526
7-35
435
1-00
351 „
•60
625
7-30
472
•90
488 „
•55
510
•80
625 obs.
•52
571
•70
430 INVESTIGATIONS IN CAPILLARITY. [251
The next question for consideration is — at what point will an opposition
to contraction arise ? The answer must depend upon the forces supposed
to be operative between the molecules of oil. If they behave like the smooth
rigid spheres of gaseous theory, no forces will be called into play until they
are closely packed. According to this view the tension would remain constant
up to the point where a double layer commences to form. It would then
suddenly change, to remain constant at the new value until the second layer
is complete. The actual course of the curve of tension deviates somewhat
widely from the above description, but perhaps not more than could be
explained by heterogeneity of the oil, whereby some molecules would mount
more easily than others, or by reference to the molecular motions which
cannot be entirely ignored. If we accept this view as substantially true,
we conclude that the first drop in tension corresponds to a complete layer one
molecule thick, and that the diameter of a molecule of oil is about 1*0 ///*.
An attractive force between molecules extending to a distance of many
diameters, such as is postulated in Laplace's theory, would not apparently
interfere with the above reasoning. An essentially different result would
seem to require a repulsive force between the molecules, resisting concentra-
tion long before the first layer is complete. In this case the tension would
begin to fall as soon as the density is sufficient to bring the repulsion into
play. On the whole this view appears less probable than the former, the
more as it involves a molecular diameter much exceeding TO/z/u,.
EXPLANATION OF FIGURES.
In the Figures (and in the tables) there is no relation between the scales of
the abscissae representing the densities in the various cases. As regards the
ordinates, representing weights or tensions, the scale is the same in all the cases,
but the zero point is arbitrary. It may be supposed to be situated on the line
of zero densities at a point 4*1 below the starting-point of the curve.
A Curious Observation.
[1902. The present paragraph was accidentally omitted in the original
publication. In experimenting upon a shallow layer of mercury contained in
a glass vessel with a flat bottom, it was noticed that a piece of iron gauze
pressed under the mercury upon the bottom of the vessel unexpectedly re-
mained down. There was no sticky substance present to which the effect
could be referred, and on inspection from below it was seen that the mercury
was out of contact with the bottom at places where the gauze was closest.
The phenomenon was thus plainly of a capillary nature, the mercury refusing
to fill up the narrowest chinks, even though the alternative was a vacuum.
The experiment may be repeated in a simpler form by substituting for
the gauze a piece of plate glass a few cms. square. If the bottom of the
vessel be also of plate glass, the expulsion of the mercury may be observed
from the whole of the contiguous areas.]
252.
THE MUTUAL INDUCTION OF COAXIAL HELICES.
[British Association Report, pp. 241, 242, 1899.]
PROFESSOR J. V. JONES* has shown that the coefficient of mutual induction
etween a circle and a coaxial helix is the same as between the circle
and a uniform circular cylindrical current-sheet of the same radial and axial
dimensions as the helix, if the currents per unit length in helix and sheet be
the same. This conclusion is arrived at by comparison of the integrals
resulting from an application of Neumann's formula; and it may be of
interest to show that it can be deduced directly from the general theory
of lines of force.
In the first place, it may be well to remark that the circuit of the helix
must be supposed to be completed, and that the result will depend upon the
manner in which the completion is arranged. In the general case the
return to the starting-point might be by a second helix lying upon the
same cylinder; but for practical purposes it will suffice to treat of helices
including an integral number of revolutions, so that the initial and final
points lie upon the same generating line. The return will then naturally
be effected along this straight line.
Let us now suppose that the helix, consisting of one revolution or of any
number of complete revolutions, is situated in a field of magnetic force
symmetrical with respect to the axis of the helix. In considering the number
of lines of force included in the complete circuit, it is convenient to follow in
imagination a radius-vector drawn perpendicularly to the axis from any point
of the circuit. The number of lines cut by this radius, as the complete
circuit is described, is the number required, and it is at once evident that the
part of the circuit corresponding to the straight return contributes nothing
to the total f. As regards any part of the helix corresponding to a rotation
* Proc. Roy. Soc. Vol. LXIII. (1897), p. 192.
t This would be true so long as the return lies anywhere in the meridianal plane. In the
general case, where the number of convolutions is incomplete, the return may be made along
a path composed of the extreme radii vectores and of the part of the axis intercepted between
them.
432 THE MUTUAL INDUCTION OF COAXIAL HELICES. [252
of the radius through an angle dO, it is equally evident that in the limit the
number of lines cut through is the same as in describing an equal angle
of the circular section of the cylinder at the place in question, whence
Professor Jones's result follows immediately. Every circular section is sampled,
as it were, by the helix, and contributes proportionally to the result, since at
every point the advance of the vector parallel to the axis is in strict pro-
portion to the rotation. It is remarkable that the case of the helix (with
straight return) is simpler than that of a system of true circles in parallel
planes at intervals equal to the pitch of the helix.
The replacement of the helix by a uniform current-sheet shows that the
force operative upon it in the direction of the axis (dH jdx) depends only upon
the values of M appropriate to the two terminal circles.
If the field is itself due to a current flowing in a helix, the condition of
symmetry about the axis is only approximately satisfied. The question
whether both helices may be replaced by the corresponding current-sheets
is to be answered in the negative, as may be seen from consideration of the
case where there are two helices of the same pitch on cylinders of nearly
equal diameters. In one relative position of the cylinders the paths are in
close proximity throughout, and the value of M will be large ; but this state
of things may be greatly altered by a relative rotation through two right
angles.
But although in strictness the helices cannot be replaced by current-
sheets, the complication thence arising can be eliminated in experimental
applications by a relative rotation. For instance, if the helix to which
the field is supposed to be due be rotated, the mean field is strictly sym-
metrical, and accordingly the mean M is the same as if the other helix were
replaced by a current-sheet. A further application of Professor Jones's
theorem now proves that the first helix may also be so replaced. Under
such conditions as would arise in practice, the mean of two positions distant
180°, or at any rate of four distant 90°, would suffice to eliminate any dif-
ference between the helices and the corresponding current-sheets, if indeed
such difference were sensible at all.
The same process of averaging suffices to justify the neglect of spirality
when the observation relates to the mutual attraction of two helices as
employed in current determinations.
253.
THE LAW OF PARTITION OF KINETIC ENERGY.
[Philosophical Magazine, XLIX. pp. 98 — 118, 1900.]
THE law of equal partition, enunciated first by Waterston for the case
of point molecules of varying mass, and the associated Boltzmann-Maxwell
doctrine respecting steady distributions have been the subject of much
difference of opinion. Indeed, it would hardly be too much to say that no
two writers are fully agreed. The discussion has turned mainly upon
Maxwell's paper of 1879*, to which objections •[• have been taken by Lord
Kelvin and Prof. Bryan, and in a minor degree by Prof. Boltzmann and
myself. Lord Kelvin's objections are the most fundamental. He writes^ :
" But, conceding Maxwell's fundamental assumption, I do not see in the
mathematical workings of his paper any proof of his conclusion ' that the
average kinetic energy corresponding to any one of the variables is the same
for every one of the variables of the system.' Indeed, as a general pro-
position its meaning is not explained, and it seems to me inexplicable. The
reduction of the kinetic energy to a sum of squares leaves the several
parts of the whole with no correspondence to any defined or definable set
of independent variables."
In a short note § written soon afterwards I pointed out some considera-
tions which appeared to me to justify Maxwell's argument, and I suggested
the substitution of Hamilton's principal function for the one employed by
Maxwell ||. The views that I then expressed still commend themselves to
* Collected Scientific Papers, Vol. n. p. 713.
t I am speaking here of objections to the dynamical and statistical reasoning of the paper.
Difficulties in the way of reconciling the results with a kinetic theory of matter are another
question.
J Proc. Roy. Soc. Vol. L. p. 85 (1891).
§ Phil. Mag. April 1892, p. 356. [Vol. m. p. 554.]
|| See also Dr Watson's Kinetic Theory of Gases, 2nd edit. 1893.
R. iv. 28
434 THE LAW OF PAKTITION OF KINETIC ENERGY. [253
me ; and I think that it may be worth while to develop them a little further,
and to illustrate Maxwell's argument by applying it to a particular case
where the simplicity of the circumstances and the familiarity of the notation
may help to fix our ideas.
But in the mean time it may be well to consider Lord Kelvin's " Decisive
Test-case disproving the Maxwell-Boltzmann Doctrine regarding Distribution
of Kinetic Energy*," which appeared shortly after the publication of my
note. The following is the substance of the argument : —
"Let the system consist of three bodies, A, B, C, all movable only in one
straight line, KHL :
" B being a simple vibrator controlled by a spring so stiff that when,
at any time, it has very nearly the whole energy of the system, its extreme
excursions on each side of its position of equilibrium are small :
" G and A, equal masses :
" C, unacted upon by force except when it strikes L, a fixed barrier,
and when it strikes or is struck by B :
"A, unacted on by force except when it strikes or is struck by B, and
when it is at less than a certain distance, HK, from a fixed repellent barrier,
E, repelling with a force, F, varying according to any law, or constant, when
A is between K and H, but becoming infinitely great when (if at any time)
A reaches K, and goes infinitesimally beyond it.
"Suppose now A, B, C to be all moving to and fro. The collisions
between B and the equal bodies A and C on its two sides must equalize,
and keep equal, the average kinetic energy of A, immediately before and
after these collisions, to the average kinetic energy of G. Hence, when
the times of A being in the space between H and K are included in the
average, the average of the sum of the potential and kinetic energies of A
is equal to the average kinetic energy of C. But the potential energy of
A at every point in the space HK is positive, because, according to our
supposition, the velocity of A is diminished during every time of its motion
from H towards K, and increased to the same value again during motion
from K to H. Hence, the average kinetic energy of A is less than the
average kinetic energy of Gl"
The apparent disproof of the law of partition of energy in this simple
problem seems to have shaken the faith even of such experts as Dr Watson
and Mr Burburyf*. M. Poincare, however, considering a special case of Lord
* Phil. Mag. May 1892, p. 466.
t Nature, Vol. XLVI. p. 100 (1892).
1900] THE LAW OF PARTITION OF KINETIC ENERGY. 435
Kelvin's problem*, arrives at a conclusion in harmony with Maxwell's law.
Prof. Bryan^f* considers that the test-case " shows the impossibility of drawing
general conclusions as to the distribution of energy in a single system from
the possible law of permanent distribution in a large number of systems."
It is indeed true that Maxwell's theorem relates in the first instance to
a large number of systems; but, as I shall show more fully later, the ex-
tension to the time-average for a single system requires only the application
of Maxwell's assumption that all phases, i.e. all states, defined both in respect
to configuration and velocity, which are consistent with the energy condition
lie on the same path, i.e. are attained by the system in its free motion sooner
or later. This fundamental assumption, though certainly untrue in special
cases, would appear to apply in Lord Kelvin's problem ; and, if so, Maxwell's
argument requires the equality of kinetic energies for A and C in the time-
averages of a single system.
In view of this contradiction we may infer that there must be a weak
place in one or other argument ; and I think I can show that Lord Kelvin's
conclusion above that the average of the sum of the potential and kinetic
energies of A is equal to the average kinetic energy of C, is not generally
true. In order to see this let us suppose the repulsive force F to be limited
to a very thin stratum at H, so that A after penetrating this stratum is
subject to no further force until it reaches the barrier K ; and let us compare
two cases, the whole energy being the same in both.
In case (i) F is so powerful that with whatever velocity (within the
possible limits) A can approach, it is reflected at H, which then behaves
like a fixed barrier. In case (ii) F is still powerful enough to produce this
result, except when A approaches it with a kinetic energy nearly equal
to the whole energy of the system. A then penetrates beyond H, moving
slowly from H to K and back again from K to H, thus remaining for a
relatively long time beyond H. Lord Kelvin's statement requires that
the average total energy of A should be the same in the two cases ; but this
it cannot be. For during the occasional penetrations beyond H in case
(ii) A has nearly the whole energy of the system ; and its enjoyment of
this is prolonged by the penetration. Hence in case (ii) A has a higher
average total energy than in case (i); and a margin is provided which
may allow the average kinetic energies to be equal. I believe that the
consideration here advanced goes to the root of the matter, and shows
why it is that the possession of potential energy may involve no deduction
from the full share of kinetic energy.
Lord Kelvin's " decisive test-case " is entirely covered by Maxwell's
reasoning — a reasoning in my view substantially correct. It would be
* Revue generate des Sciences, July 1894.
t "Report on Thermodynamics," Part II. S 26. Brit. Ass. Rep. 1894.
28—2
436 THE LAW OF PARTITION OF KINETIC ENERGY. [253
possible, therefore, to take this case as a typical example in illustration
of the general argument ; but I prefer for this purpose, as somewhat simpler,
another test-case, also proposed by Lord Kelvin. This is simply that of
a particle moving in two dimensions; and it may be symbolized by the
motion of the ball upon a billiard-table. If there is to be potential energy,
the table may be supposed to be out of level. The reconsideration of this
problem may perhaps be thought superfluous, seeing that it has been ably
treated already by Prof. Boltzmann*. But his method, though (I believe)
quite satisfactory, is somewhat special. My object is rather to follow closely
the steps of the general theory. If objections are taken to the argument
of the particular case, they should be easy to specify. If, on the other hand,
the argument of the particular case is admitted, the issue is much narrowed.
I shall have occasion myself to make some comments relating to one point
in the general theory not raised by the particular case.
In the general theory the coordinates f of the system at time t are denoted
by qlf q2, ... qn, and the momenta by pl} p2} ... pn. At an earlier time 1f the
coordinates and momenta of the same motion are represented by correspond-
ing letters accented, and the first step is the establishment of the theorem
usually, if somewhat enigmatically, expressed
dq\ dq'2 . . . dq'ndp\dp2 . . . dp'n = dq^qz . . . dq^dftdp, . . . dpn (1)
In the present case ql} q2 are the ordinary Cartesian coordinates (x, y) of
the particle ; and if we identify the mass with unity, pl} p2 are simply the
corresponding velocity-components (u, v) ; so that (1) becomes
dx' dy du' dv' = dx dy du dv (2)
For the sake of completeness I will now establish (2) de novo.
In a possible motion the particle passes from the phase (x, y', u, v')
at time t' to the phase (x, y, u, v) at time t. In the following discussion
t' and t are absolutely fixed times, but the other quantities are regarded
as susceptible of variation. These variations are of course not independent.
The whole motion is determined if either the four accented, or the four
unaccented, symbols be given. Either set may therefore be regarded as
definite functions of the other set. Or again, the four coordinates x ', y ', x, y
may be regarded as independent variables, of which u', v', u, v are then
functions.
The relations which we require are readily obtained by means of Hamilton's
principal function S, where
S=f\T-V)dt (3)
* Phil. Mag. Vol. xxxv. p. 156 (1893).
t Generalized coordinates appear to have been first applied to these problems by Boltzmann.
1900] THE LAW OF PARTITION OF KINETIC ENERGY. 437
In this V denotes the potential energy in any position, and T is the kinetic
energy, so that
S may here be regarded as a function of the initial and final coordinates ;
and we proceed to form the expression for BS in terms of Sx, By', Bx, By.
By (3)
B8= (ST-SV)dt, ........................... (5)
J t1
and
so that
= £&» + #% - t(xBx + yBy)dt;
SS = IxBx + y ByY - I \x Bx + yBy + 8 V) dt.
By the general equation of dynamics the term under the integral sign
vanishes throughout, and thus finally
BS=uSx + vSy-u'Bx'-v'By (6)
In the general theory the corresponding equation is
BS^ZpBq-Zp'Bq' (7)
Equation (6) is equivalent to
u=-dS/dx', u = dS/dx,
.(8)
v = dS/dy. '
It is important to appreciate clearly the meaning of these equations.
S is in general a function of x, y, x, y' ; and (e.g.) the second equation
signifies that u is equal to the rate at which S varies with x, when y, x', y'
are kept constant, and so in the other cases.
We have now to consider, not merely a single particle, but an immense
number of similar particles, moving independently of one another under the
same law (V), and distributed at time t over all possible phases (x, y, u, v).
The most general expression for the law of distribution is
f(x, y, u, v) dxdydudv, (9)
signifying that the number of particles to be found at time t within a
prescribed range of phase is to be obtained by integrating (9) over the range
* As is not unusual in the integral calculus, we employ the same symbols x, &c. to denote the
current and the final values of the variables. If desired, the final values may be temporarily
distinguished as x", &c.
438 THE LAW OF PARTITION OF KINETIC ENERGY. [253
in question. But such a distribution would in general be unsteady. If it
obtained at time t, it would be departed from at time t', and vice versa,
owing to the natural motions of the particles. The question before us is
to ascertain what distributions are steady, i.e. are maintained unaltered
notwithstanding the motions.
It will be seen that it is the spontaneous passage of a particle from one
phase to another that limits the generality of the function /. If there be
no possibility of passage, say, from the phase (#', y', u', v) to the phase
(x, y, u, v), or, as it may be expressed, if these phases do not lie upon the
same path, then there is no relation imposed upon the corresponding values
of/. An example, given by Prof. Bryan (1. c. § 17), well illustrates this point.
Suppose that F=0, so that every particle pursues a straight course with
uniform velocity. The phases (x', y, u', v) and (x, y, u, v) can lie upon the
same path only if u = u, v' = v. Accordingly / remains arbitrary so far
as regards u and v. For instance, a distribution
f(u,v)dxdydudv (10)
is permanent whatever may be the form of f, understood to be independent
of x and y. In this case the distribution is uniform in space, but uniformity
is not indispensable. Suppose, for example, that all the particles move
parallel to x, so that f vanishes unless v = 0. The general form (9) now
reduces to
f(x, y, u) dxdydu; (11)
and permanency requires that the distribution be uniform along any line
for which y is constant. Accordingly, f must be independent of x, so
that permanent distributions are of the form
f(y, u) dxdydu, (12)
in which / is an arbitrary function of y and u. If either y or u be varied, we
are dealing with a different path (in the sense here involved), and there is no
connexion between the corresponding values of f. But if while y and u
remain constant, x be varied, the value of f must remain unchanged, for the
different values of x relate to the same path.
Before taking up the general question in two dimensions, it may be well
to consider the relatively simple case of motion in one dimension, which,
however, is not so simple but that it will introduce us to some of the points
of difficulty. The particles are supposed to move independently upon one
straight line, and the phase of any one of them is determined by the co-
ordinate x and the velocity u. At time t' the phase of a particle will be
denoted by (x', u'), and at time t the phase of the same particle will be (x, u),
where u will in general differ from u', since we no longer suppose that V is
constant, but rather that it is variable in a known manner, i.e. is a known
1900] THE LAW OF PARTITION OF KINETIC ENERGY. 439
function of x. The number of particles which at time t lie within the limits
of phase represented by dxdu is f(x, u) dxdu, and the question is whether
this distribution is steady, and in particular whether it was the same at
time t'. In order to find the distribution at time t1, we regard x, u as known
functions of x', u, and transform the multiple differential. The result of this
transformation is best seen by comparison with intermediate transformations
in which dxdu and dxdu are compared with dxdx'. We have
(13)
du'
dx'du'=dxdx' x -r- ............................ (14)
In du/dx of (13) x is to be kept constant, and in du /dx of (14) x' is to
be kept constant. If we disregard algebraic sign, both are by (8) equal
to d*S/dxdx', and are therefore equal to one another. Hence we may
write
dxdu = dx'du ; .............................. (15)
and the transformation is expressed by
f(x, u) dx du =/ (x', u') dx'du , .................. (16)
where/j (x', u) is the result of substituting for x, u inf(x, u) their values in
terms of x, u'. The right-hand member of (16) expresses the distribution
at time t' corresponding to the distribution at time t expressed by the left-
hand member, as determined by the laws of motion between the two phases.
If the distribution is to be steady, / (x', u') must be identical with / (#', u');
in other words f(x, u) must be such a function of (x, u) that it remains
unchanged when (x, u) refers to various phases of the motion of the same
particle. Now, if E denote the total energy, so that
E=$u*+V, .............................. (17)
then E remains constant during the motion ; and thus, if for the moment
we suppose f expressed in terms of E and x, we see that x cannot enter, or
that /is a function of E only. The only permanent distributions accordingly
are those included under the form
j(E}dxdu, .............................. (18)
where E is given by (17), and /is an arbitrary function.
It is especially to be noticed that the limitation to the form (18) holds
only for phases lying upon the same path. If two phases have different
energies, they do not lie upon the same path, but in this case the independence
of the distributions in the two phases is already guaranteed by the form
of (18). The question is whether all phases of given energy lie upon the
440
THE LAW OF PARTITION OF KINETIC ENERGY.
[253
same path. It is easy to invent cases for which the answer will be in the
negative. Suppose, for example, that there are two centres of force 0, 0'
on the line of motion which attract with a force at first proportional to
distance but vanishing when the distance exceeds a certain value less than
the interval 00'. A particle may then vibrate with the same (small) energy
either round 0 or round 0' ; but the phases of the two motions do not lie
upon the same path. Consequently / is not limited by the condition of
steadiness to be the same in the two groups of phases. In all cases steadiness
is ensured by the form (18); and if all phases of equal energy lie upon the
same path, this form is necessary as well as sufficient.
All the essential difficulties of the theory appear to be raised by the
particular case just discussed, and the reader to whom the subject is new
is recommended to give it his careful attention.
In the more general problem of motion in two dimensions the discussion
follows a parallel course. In order to find the distribution at time If cor-
responding to (9) at time t, we have to transform the multiple differential,
regarding x, y,u,v as known functions of x', y', u', v'. Here again we take
the initial and final coordinates x, y, x'. y as an intermediate set of variables.
Thus
dad dy' du' dv' = dx'dy'dxdy x
dxdydudv = dxdydx'dy' x
dx
_
dy
du
du
dv'
dx
dtf
dy
dv
dx'
dy
.(19)
.(20)
In the determinants of (19), (20) the motion is regarded as a function of
x, y, x', y', and the three quantities which do not appear in the denominator
of any differential coefficient are to be considered constant. This was also
the understanding in equations (8), from which we infer that the two deter-
minants are equal, being each equivalent to
dxdx' ' dxdy'
d*S d*S
.(21)
dx'dy' dydy
Hence we may write
dxdydudv = dx'dy'du'dv, (22)
an equation analogous to (15). By the same reasoning as was employed
1900] THE LAW OF PARTITION OF KINETIC ENERGY. 441
for motion in one dimension it follows that, if the distribution is to be steady,
f(x, y, u, v) in (9) must remain constant for all phases which lie upon the
same path. A distribution represented by
f(E)dxdydudv, .............................. (23)
where
* + V, ........................... (24)
will satisfy the conditions of steadiness whatever be the form of/; but this
form is only necessary under the restriction known as Maxwell's assumption
or postulate, viz. that all phases of equal energy lie upon the same path.
It is easy to give examples in which Maxwell's assumption is violated,
and in which accordingly steady distributions are not limited to (23). Thus,
if no force act parallel to y, so that V reduces to a function of x only, the
component velocity v remains constant for each particle, and no phases
for which v differs lie upon the same path. A distribution
f(E,v)dxdydudv ........................... (25)
is then steady, whatever function /may be of E and v.
That under the distribution (23) the kinetic energy is equally divided
between the component velocities u and v is evident from symmetry. It
is to be observed that the law of equal partition applies not merely upon the
whole, but for every element of area dxdy, and for every value of the total
energy, and at every moment of time. When x and y are prescribed as well
as E, the value of the resultant velocity itself is determined by (24).
Another feature worthy of attention is the spacial distribution ; and it
happens that this is peculiar in the present problem. To investigate it
we must integrate (23) with respect to u and v, x and y being constant.
Since x and y are constant, V is constant ; so that, if we suppose E to lie
within narrow limits E and E + dE, the resultant velocity U will lie between
limits given by
UdU=dE ............................... (26)
If we transform from u, v to U, 6, where
u=Ucos0, v=Usin0, ..................... (27)
dudv becomes UdUdO', so that on integration with respect to 6 we have,
with use of (26),
2TrF(E)dE .dxdy ............................ (28)
The spacial distribution is therefore uniform.
In order to show the special character of the last result, it may be well
to refer briefly to the corresponding problem in three dimensions, where the
442 THE LAW OF PARTITION OF KINETIC ENERGY. [253
coordinates of a particle are x, y, z and the component velocities are u, v, w.
The steady distribution corresponding to (23) is
/(E)dxdydzdudvdw, ........................ (29)
in which
. ............... (30)
Here equation (26) still holds good, and the transformation of dudvdw is,
as is well known, 4nrU2dU. Accordingly (29) becomes
4>TrF(E)dE.(2E-2V^da;dy, .................. (31)
no longer uniform in space, since V is a function of x, y.
In (31) the density of distribution decreases as V increases. For the
corresponding problem in one dimension (18) gives
F(E)dE.(2E-2V)-ldx, ..................... (32)
so that in this case the density increases with increasing V.
The uniform distribution of the two-dimensional problem is thus peculiar.
Although an immediate consequence of Maxwell's equation (41), see (41)
below, I failed to remark it in the note before referred to, where I wrote
as if a uniform distribution in the billiard-table example required that V = 0.
In order to guard against a misunderstanding it may be well to say that
the uniform distribution does not necessarily extend over the whole plane.
Wherever (E — V) falls below zero there is of course no distribution.
We have thus investigated for a particle in two dimensions the law of
steady distribution, and the equal partition of energy which is its necessary
consequence. And we see that " the only assumption necessary to the direct
proof is that the system, if left to itself in its actual state of motion, will,
sooner or later, pass through every phase which is consistent with the
equation of energy" (Maxwell). It will be observed that so far nothing
whatever has been said as to time-averages for a single particle. The law of
equal partition, as hitherto stated, relates to a large number of particles and
to a single moment of time.
The extension to time-averages, the aspect under which Lord Kelvin has
always considered the problem, is important, the more that some authors
appear to doubt the possibility of such extension. Thus Prof. Bryan (Report,
§ 11, 1894), speaking of Maxwell's assumption, writes : — " To discover, if
possible, a general class of dynamical systems satisfying the assumption
would form an interesting subject for future investigation. It is, however,
doubtful how far Maxwell's law would be applicable to the time-averages
of the energies in any such system. We shall see, in what follows, that the
law of permanent distribution of a very large number of systems is in many
cases not unique. Where there is more than one possible distribution it
1900] THE LAW OF PARTITION OF KINETIC ENERGY. 443
would be difficult to draw any inference with regard to the average distri-
bution (taken with respect to the time) for one system."
The extension to time-averages appears to me to require nothing more
than Maxwell's assumption, without which the law of distribution itself
is only an artificial arrangement, sufficient indeed but not necessary for
steadiness. We shall still speak of the particle moving in two dimensions,
though the argument is general. It has been shown that at any moment
the w-energy and the v-energy of the group of particles is the same ; and
it is evident that the equality subsists if we integrate over any period of
time. But if this period be sufficiently prolonged, and if Maxwell's assumption
be applicable, it makes no difference whether we contemplate the whole group
of particles or limit ourselves to a single member of it. It follows that
for a single particle the time-averages of u2 and v2 are equal, provided the
averages be taken over a sufficient length of time.
On the other hand, if in any case Maxwell's assumption be untrue, not
only is the special distribution unnecessary for steadiness, but even if it
be artificially arranged, the law of equal time-averages does not follow as
a consequence.
Having now considered the special problem at full — I hope it may not
be thought at undue — length, I pass on to some remarks on the general
investigation. This proceeds upon precisely parallel lines, and the additional
difficulties are merely those entailed by the use of generalized coordinates.
Thus (1) follows from (7) by substantially the same process (given in my
former note) that (22) follows from (6). Again, if E denote the total energy
of a system, the distribution
f(E)dq1...dqndp1 ...dpn, (33)
where f is an arbitrary function, satisfies the condition of permanency ; and,
if Maxwell's assumption be applicable, it is the only form of distribution that
can be permanent.
As I hinted before, some of the difficulties that have been felt upon this
subject may be met by a fuller recognition of the invariantic character of
the expressions. This point has been ably developed by Prof. Bryan, who
has given (loc. cit. § 14) a formal verification that (33) is unaltered by a change
of coordinates. If we follow attentively the process by which (1) is established,
we see that in (3) there is no assumption that the system of coordinates is
the same at times t' and t, and that accordingly we are not tied to one system
in (33). Indeed, so far as I can see, there would be no meaning in the
assertion that the system of generalized coordinates employed for two different
configurations was the same*.
* It would be like saying that two points lie upon the same curve, when the character of the
curve is not denned.
444 THE LAW OF PARTITION OF KINETIC ENERGY. [253
We come now to the deduction from (33) of Maxwell's law of partition
of energy. On this Prof. Bryan (loc. cit. § 20) remarks: — "Objections have
been raised to this step in Maxwell's work by myself (' Report on Thermo-
dynamics/ Part I. § 44) on the ground that the kinetic energy cannot in
general be expressed as the sum of squares of generalized momenta corre-
sponding to generalized coordinates of the system, and by Lord Kelvin
(Nature, Aug. 13, 1891) on the ground that the conclusion to which it leads
has no intelligible meaning. Boltzmann (Phil. Mag. March 1893) has put the
investigation into a slightly modified form which meets the first objection,
and which imposes a certain restriction upon the generality of the result.
Under this limitation the result is perfectly intelligible, and the second
objection is therefore also met." At this point I find myself in disagreement
with all the above quoted authorities, and in the position of maintaining
the correctness of Maxwell's original deduction.
Prof. Boltzmann considers that " Maxwell committed an error in assuming
that by choosing suitable coordinates the expression for the vis viva could
always be made to contain only the squares of the momenta." This is
precisely the objection which I supposed myself to have already answered
in 1892. I wrote, " It seems to be overlooked that Maxwell is limiting his
attention to systems in a given configuration, and that no dynamics is founded
upon the reduced expression for T. The reduction can be effected in an
infinite number of ways. We may imagine the configuration in question
rendered one of stable equilibrium by the introduction of suitable forces
proportional to displacements. The principal modes of isochronous vibration
thus resulting will serve the required purpose."
It is possible, therefore, so to choose the coordinates that for a given
configuration (and for configurations differing infinitely little therefrom) the
kinetic energy T,.which is always a quadratic function of the velocities, shall
reduce to a sum of squares with, if we please, given coefficients. Thus in
the given configuration
T = W + b&+...+W; ..................... (34)
and, since in general p = dT/dq,
so that T = to2 + ib>22+---+to2 ...................... (35)
Whether the coordinates required to effect a similar reduction for other
configurations are the same is a question with which we are not concerned.
The mean value of pr2 for all the systems in the given configuration is,
according to (33),
* Confer Bryan, loc. cit.
1900] THE LAW OF PARTITION OF KINETIC ENERGY. 445
The limits for each variable may be supposed to be ±00; but the large
values do not really enter if we suppose F(E) to be finite for moderate,
perhaps for nearly definite, values of E only.
It is now evident that the mean value is the same for all the momenta p ;
and accordingly that for each the mean value of £p2 is 1/n of the mean
value of T. This result holds good for every moment of time, for every
configuration, for every value of E, and for every system of resolution (of
which there are an infinite number) which allows T to be expressed in the
form (35).
In the case where the " system " consists of a single particle, (35) is justified
by any system of rectangular coordinates ; and although we are not bound to
use the same system for different positions of the particle, it would conduce
to simplicity to do so. If the system be a rigid body, we may measure the
velocities of the centre of inertia parallel to three fixed rectangular axes,
while the remaining momenta refer to rotations about the principal axes
of the body. If Maxwell's assumption hold good, a permanent distribution
is such that in one, or in any number of positions, the mean energy of each
rotation and of each translation is the same. And under the same restriction
a similar assertion may be made respecting the time-averages for a single
rigid body.
There is much difficulty in judging of the applicability of Maxwell's
assumption. As Maxwell himself showed.it is easy to find cases of exception;
but in most of these the conditions strike one as rather special. It must
be observed, however, that if we take it quite literally, the assumption is
of a severely restrictive character; for it asserts that the system, starting
from any phase, will traverse every other phase (consistent with the energy
condition) before returning to the initial phase. As soon as the initial
phase is recovered, a cycle is established, and no new phases can be reached,
however long the motion may continue.
We return now to the question of the distribution of momenta among
the systems which occupy a given configuration, still supposing the coordinates
so chosen as to reduce T to a sum of squares (35). It will be convenient
to fix our attention upon systems for which E lies within narrow limits,
E and E + dE. Since E is given, there is a relation between pl}p2, ... pn,
and we may suppose pn expressed in terms of E and the remaining momenta.
By (35)
since the configuration is given, and thus (33) becomes
f(E)dE.dq1...dqn.pn-1dpl...dpn-l ................ (37)
446 THE LAW OF PARTITION OF KINETIC ENERGY. [253
For the present purpose the latter factors alone concern us, so that what
we have to consider is
in which T, being equal to E — V, is given. For the moment we may suppose
that 2T is unity.
The whole number of systems is to be found by integrating (38), the
integral being so taken as to give the variables all values consistent with the
condition that p? + p2* + . . . + p*n-! is not greater than unity. Now
(1 -pflP-idp! (39)
J J V {J- — Pi' — ••- —p~n-i< *- \^'*— 2J-' -I
and
(40)
in which F (|) = V71"- Thus the whole number of systems is
or on restoration of 2T, equal to 2E — 2V,
y^{2#-2Fp- ......................... (41)
To this we shall return later; but for the present what we require to
ascertain is the distribution of one of the momenta, say plt irrespectively
of the values of the remaining momenta. By (39), (40) the number of
systems for which pl lies between pi and p± + dp^ in comparison with the
whole number of systems is
_ £ u»-i dPl
1 '
rg)r(t»-»
This is substantially Maxwell's investigation, and (42) corresponds with his
equation (51). As was to be expected, the law of distribution is the same
for all the momenta. From the manner of its formation, we note that the
integral of (42), taken between the limits pt = + \f (2T), is equal to unity.
Maxwell next proceeds to the consideration of the special form assumed
by (42), when the number n of degrees of freedom is extremely great*. This
part of the work seems to be very important ; but it has been much neglected,
probably because the result was not correctly stated.
* The particular cases where n = 2, or n=B, are also worthy of notice.
1900] THE LAW OF PARTITION OF KINETIC ENERGY. 447
Dropping the suffix as unnecessary, we have to consider the form of
I-" -'
when n is very great, the mean value of p* becoming at the same time small
in comparison with 2T. If we write
........................... (43)
we have
Limit l - =e-*'/4jr = <rP*'2pl ................ (44)
The limit of the fraction containing the F functions may be obtained
by the formula
r (m + 1) = e-mmm V (2m7r) ;
and the limiting form of (42) becomes
dp
P ................... (4<
-*** dp
It may be observed that the integral of (45) between the limits + oo is
unity, and that this fact might have been used to determine the numerical
factor.
Maxwell's result is given in terms of a quantity k, analogous to K, and
defined by
W = k .................................. (46)
It is
•;.;; ,., ; Tifez''** ............................ (47)
The corresponding form from (45) is
In like manner if we inquire what proportion of the whole number of
systems have momenta lying within the limits denoted by dpidp^ ... dpr,
where r is a number very small relatively to n, we get
... dpr
'
or, if we prefer it,
-(*»+:'4*>IU* .dpr
These results follow from the general expression (38), in the same way as
.does (45), by stopping the multiple integration at an earlier stage. The
448 THE LAW OF PARTITION OF KINETIC ENERGY. [253
remaining variables range over values which may be considered in each case
to be unlimited. If the integration between ± oo be carried out completely,
we recover the value unity.
The interest of the case where n is very great lies of course in the
application to a gas supposed to consist of an immense number of similar
molecules*, or of several sets of similar molecules; and the question arises
whether (45) can be applied to deduce the Maxwellian law of distribution
of velocities among the molecules of a single system at a given instant of
time. A caution may usefully be interposed here as to the sense in which
the Maxwellian distribution is to be understood. It would be absurd to
attempt to prove that the distribution in a single system is necessarily such
and such, for we have already assumed that every phase, including every
distribution of velocities, is attainable, and indeed attained if sufficient time
be allowed. The most that can be proved is that the distribution will
approximate to a particular law for the greater part of the time, and that
if sensible deviations occur they will be transitory.
In applying (45) to a gas it will be convenient to suppose in the first
instance that all the molecules are similar. Each molecule has several
degrees of freedom, but we may fix our attention upon one of them, say
the ^-velocity of the centre of inertia, usually denoted by u. In (45) the
whole system is supposed to occupy a given configuration ; and the expression
gives us the distribution of velocity at a given time for a single molecule
among all the systems. The distribution of velocity is the same for every
other molecule, and thus the expression applies to the statistics of all the
molecules of all the systems. Does it also apply to the statistics of all the
molecules of a single system ? In order to make this inference we must
assume that the statistics are the same (at the same time) for all the systems,
or, what comes to the same thing (if Maxwell's assumption be allowed), that
they are the same for the same system at the various times when it passes
through a given configuration.
Thus far the argument relates only to a single configuration. If the
configuration be changed, there will be in general a change of potential
energy and a corresponding change in the kinetic energy to be distributed
amongst the degrees of freedom. But in the case of a gas, of which the
statistics are assumed to be regular, the potential energy remains approxi-
mately constant when exclusion is made of exceptional conditions. The same
law of distribution of velocity then applies to every configuration, that is,
it may be asserted without reference to the question of configuration. We
thus arrive at the Maxwellian law of velocities in a single gas, as well as the
* The terms "gas" and "molecule" are introduced for the sake of brevity. The question is
still purely dynamical.
1900] THE LAW OF PARTITION OF KINETIC ENERGY. 449
relation between the velocities in a mixture of molecules of different kinds
first laid down by Waterston.
The assumptions which we have made as to the practical regularity of
statistics are those upon which the usual theory of ideal gases is founded;
but the results are far more general. Nothing whatever has been said as
to the character of the forces with which the molecules act upon one another,
or are acted upon by external agencies. Although for distinctness a gas has
been spoken of, the results apply equally to a medium constituted as a liquid
or a solid is supposed to be. A kinetic theory of matter, as usually under-
stood, appears to require that in equilibrium the whole kinetic energy shall
be equally shared among all the degrees of freedom, and within each degree
of freedom be distributed according to the same law. It is included in this
statement that temperature is a matter of kinetic energy only, e. g. that when
a vertical column of gas is in equilibrium, the mean velocity of a molecule
is the same at the top as at the bottom of the column.
Reverting to (37), (41), in order to consider the distribution of the
systems as dependent upon the coordinates independently of the velocities,
we have, omitting unnecessary factors,
{E-V}^dq,dq,...dqn ......................... (51)
If n= 2, e.g. in the case already considered of a single particle moving in two
dimensions, or of two particles moving in one dimension, or again whatever
n may be, provided V vanish, the first factor disappears, so that the distri-
bution is uniform with respect to the coordinates ql ... qn. If n > 2 and V
be finite, the distribution is such as to favour those configurations for which
V is least.
" When the number of variables is very great, and when the potential
energy of the specified configuration is very small compared with the total
energy of the system, we may obtain a useful approximation to the value of
{E — V}^~1 in an exponential form ; for if we write (as before) E = nK,
(52)
nearly, provided n is very great and V is small compared with E. The
expression is no longer approximate when V is nearly as great as E, and it
does not vanish, as it ought to do, when V = E" (Maxwell.)
In the case of gas composed of molecules whose mutual influence is
limited to a small distance and which are not subject to external forces,
the distribution expressed by (51) is uniform in space except near the
boundary. For if q1 denote the ^-coordinate of a particular molecule, and
if we effect the integration with respect to all the coordinates of other
molecules as well as the other coordinates of the particular molecule, we
must arrive at a result independent of x, provided x relate to a point well
R. iv. 29
450 THE LAW OF PARTITION OF KINETIC ENERGY. [253
in the interior. That is to say in the various systems contemplated the
particular molecule is uniformly distributed with respect to x. The same
is true of y and z, and thus the whole spacial distribution is uniform. If
the single system constituting the gas has uniform statistics, it will follow
that the distribution in it of molecules similar to the particular molecule
is uniform.
The uniformity of the distribution is disturbed if an external force acts.
In illustration of this we may consider the case of gravity. From (52) the
distribution with respect to the coordinates of the particular molecule
will be
and the same formula gives the density of molecules similar to the particular
molecule in a single system.
The main purpose of this paper is now accomplished ; but I will take
the opportunity to make a few remarks upon some general aspects of a
kinetic theory of matter. Many writers appear to commit themselves to
absolute statements, but Kelvin* and Boltzmann and Maxwell fully recognize
that conclusions can never be more than probable. The second law of thermo-
dynamics itself is in this predicament. Indeed it might seem at first sight
as if the case were even worse than this. Mr Culverwell has emphasized
a difficulty, which must have been pretty generally felt, arising out of the
reversibility of a d3rnamical system. If during one motion of a system energy
is dissipated, restoration must occur when the motion is reversed. How then
is one process more probable than the other ? Prof. Boltzmann has replied
to this objection, upon the whole I think satisfactorily, in a very interesting
letterf. The available (internal) energy of a system tends to zero, or rather
to a small value, only because the conditions, or phases as we have called
them, corresponding to small values are more probable, i.e. more numerous.
If there is considerable available energy at any moment, it is because the
condition is then exceptional and peculiar. After a short interval of time
the condition may become more peculiar still, and the available energy may
increase, but this is improbable. The probability is that the available energy
will, if not at once, at any rate after a short interval, decrease owing to the
substitution of a more nearly normal state of things.
There is, however, another side to this question, which perhaps has been
too much neglected. Small values of the available energy are indeed more
* Witness the following remarkable passage : — " It is a strange but nevertheless a true
conception of the old well-known law of the conduction of heat to say that it is very improbable
that in the course of 1000 years one-half the bar of iron shall of itself become warmer by a degree
than the other half; and that the probability of this happening before 1,000,000 years pass
is 1000 times as great as that it will happen in the course of 1000 years, and that it certainly will
happen in the course of some very long time."— (Nature, Vol. ix. p. 443, 1874.)
t Nature, Vol. LI. p. 413 (1895).
1900] THE LAW OF PARTITION OF KINETIC ENERGY. 451
probable than large ones, but there is a degree of smallness below which it is
improbable that the value will lie. If at any time the value lies extremely
low, it is an increase and not a decrease which is probable. Maxwell showed
long ago how a being capable of dealing with individual molecules would
be in a position to circumvent the second law. It is important to notice
that for this end it is not necessary to deal with individual molecules. It
would suffice to take advantage of local reversals of the second law, which
will involve, not very rarely, a considerable number of neighbouring molecules.
Similar considerations apply to other departures from a normal state of
things, such, for example, as unequal mixing of two kinds of molecules,
or such a departure from the Waterston relation (of equal mean kinetic
energies) as has been investigated by Maxwell and by Tait and Burbury.
The difficulties connected with the application of the law of equal parti-
tion of energy to actual gases have long been felt. In the case of argon
and helium and mercury vapour the ratio of specific heats (1'67) limits the
degrees of freedom of each molecule to the three required for translatory
motion. The value (1'4) applicable to the principal diatomic gases gives
room for the three kinds of translation and for two kinds of rotation. Nothing
is left for rotation round the line joining the atoms, nor for relative motion of
the atoms in this line. Even if we regard the atoms as mere points, whose
rotation means nothing, there must still exist energy of the last-mentioned
kind, and its amount (according to the law) should not be inferior.
We are here brought face to face with a fundamental difficulty, relating
not to the theory of gases merely, but rather to general dynamics. In most
questions of dynamics a condition whose violation involves a large amount of
potential energy may be treated as a constraint. It is on this principle that
solids are regarded as rigid, strings as inextensible, and so on. And it is upon
the recognition of such constraints that Lagrange's method is founded. But
the law of equal partition disregards potential energy. However great may
be the energy required to alter the distance of the two atoms in a diatomic
molecule, practical rigidity is never secured, and the kinetic energy of the
relative motion in the line of junction is the same as if the tie were of
the feeblest. The two atoms, however related, remain two atoms, and the
degrees of freedom remain six in number.
What would appear to be wanted is some escape from the destructive
simplicity of the general conclusion relating to partition of kinetic energy,
whereby the energy of motions involving larger amounts of potential energy
should be allowed to be diminished in consequence. If the argument, as
above set forth after Maxwell, be valid, such escape must involve a repudiation
of Maxwell's fundamental postulate as practically applicable to systems with
an immense number of degrees of freedom.
29-2
254.
ON THE VISCOSITY OF ARGON AS AFFECTED BY
TEMPERATURE.
[Proceedings of the Royal Society, LXVI. pp. 68—74, 1900.]
ACCORDING to the kinetic theory, as developed by Maxwell, the viscosity
of a gas is independent of its density, whatever may be the character of the
encounters taking place between the molecules. In the typical case of a
gas subject to a uniform shearing motion, we may suppose that of the three
component velocities v and w vanish, while u is a linear function of y, indepen-
dent of x and z. If p be the viscosity, the force transmitted tangentially
across unit of area perpendicular to y is measured by pdu/dy. This repre-
sents the relative momentum, parallel to x, which in unit of time crosses the
area in one direction, the area being supposed to move with the velocity
of the fluid at the place in question. We may suppose, for the sake of
simplicity, and without real loss of generality, that u is zero at the plane.
The momentum, which may now be reckoned absolutely, does not vanish,
as in the case of a gas at rest throughout, because the molecules come from
a greater or less distance, where (e.g.) the value of u is positive. The distance
from which (upon the average) the molecules may be supposed to have come
depends upon circumstances. If, for example, the molecules, retaining their
number and velocity, interfere less with each other's motion, the distance
in question will be increased. The same effect will be produced, without
a change of quality, by a simple reduction in the number of molecules, i.e.,
in the density of the gas, and it is not difficult to recognize that the distance
from which the molecules may be supposed to have come is inversely as the
density. On this account the passage of tangential momentum per molecule
is inversely as the density, and since the number of molecules crossing is
directly as the density, the two effects compensate, and upon the whole
the tangential force and therefore the viscosity remain unaltered by a change
of density.
1900] ON THE VISCOSITY OF ARGON AS AFFECTED BY TEMPERATURE. 453
On the other hand, the manner in which this viscosity varies with
temperature depends upon the nature of the encounters. If the molecules
behaved like Boscovich points, which exercise no force upon one another until
the distance falls to a certain value, and which then repel one another
infinitely (erroneously called the theory of elastic spheres), then, as Maxwell
proved, the viscosity would be proportional to the square root of the absolute
temperature. Or again, if the law of repulsion were as the inverse fifth
power of the distance, viscosity would be as the absolute temperature.
In the more general case where the repulsive force varies as r~n, the
dependence of p upon temperature may also be given. If v be the velocity
of mean square, proportional to the square root of the temperature, p, varies
n+%
as vn~l, a formula which includes the cases (n = 5, n= oo ) already specified.
If we assume the law already discussed — that p is independent of density —
this conclusion may be arrived at very simply by the method of " dimensions."
In order to see this we note that the only quantities (besides the density)
on which //, can depend are m the mass of a particle, v the velocity of mean
square, and k the repulsive force at unit distance. The dimensions of these
quantities are as follows : —
p = (mass)1 (length)"1 (time)"1,
TO = (mass)1,
v = (length)1 (time)"1,
k = (mass)1 (length)"*1 (time)-2.
Thus, if we assume
pKmv.vy.k' .............................. (1)
we have l=x + z, — l=y + (n + l)z, — l= — y — 2z,
n+l n+B 2
whence * = ^> 2/ = — r = — r
n+l n+8 2
Accordingly f* = a.mn-1 .vn~l .k~n-\ ........................ (2)
where a is a purely numerical coefficient. For a given kind of molecule,
w and k are constant. Thus
n+8 n+3
............................... (3)
The case of sudden impacts (n = oo ) gives, as already remarked,
Hence k disappears, and the consideration of dimensions shows that fj, oc d~2,
where d is the diameter of the particles.
454 ON THE VISCOSITY OF ARGON AS AFFECTED BY TEMPERATURE. [254
The best experiments on air show that, so far as a formula of this kind
can represent the facts, p oc 00"77. It may be observed that n = 8 corre-
sponds to /* oc 00"79.
When we remember that the principal gases, such as oxygen, hydrogen,
and nitrogen, are regarded as diatomic, we may be inclined to attribute
the want of simplicity in the law connecting viscosity and temperature to
the complication introduced by the want of symmetry in the molecules and
consequent diversities of presentation in an encounter. It was with this idea
that I thought it would be interesting to examine the influence of tempera-
ture upon the viscosity of argon, which in the matter of specific heat behaves
as if composed of single atoms. From the fact that no appreciable part of
the total energy is rotatory, we may infer that the forces called into play
during one encounter are of a symmetrical character. It seemed, therefore,
more likely that a simple relation between viscosity and temperature would
obtain in the case of argon than in the case of the " diatomic " gases.
The best experimental arrangement for examining this question is probably
that of Holman*, in which the same constant stream of gas passes in suc-
cession through two capillaries at different temperatures, the pressures being
determined before the first and after the second passage, as well as between
the two. But to a gas like argon, available in small quantities only, the
application of this method is difficult. And it seemed unnecessary to insist
upon the use of constant pressures, seeing that it was not proposed to in-
vestigate experimentally the dependence of transpiration upon pressure.
The theoretical formula for the volume of gas transpired, analogous
to that first given by Stokes for an incompressible fluid, was developed by
O. E. Meyerf. Although not quite rigorous, it probably suffices for the
purpose in hand. If pli Vl denote the pressure and volume of the gas as
it enters the capillary, p^, F2 as it leaves the capillary, we have
In this equation t denotes the time of transpiration, R the radius of the
tube, I its length, and jj, the viscosity measured in the usual way.
In order to understand the application of the formula for our present
purpose, it will be simplest to consider first the passage of equal volumes
of different gases through the capillary, the initial pressures, and the constant
temperature being the same. In an apparatus, such as that about to be
described, the pressures change as the gas flows, but if the pressures are
definite functions of the amount of gas which at any moment has passed the
* Phil. Mag. Vol. m. p. 81 (1877).
t Fogg. Ann. Vol. cxxvii. p. 269 (1866).
1900] ON THE VISCOSITY OF ARGON AS AFFECTED BY TEMPERATURE. 455
capillary, this variation does not interfere with the proportionality between
t and fj,. For example, if the viscosity be doubled, the flow takes place
precisely as before, except that the scale of time is doubled. It will take
twice as long as before to pass the same quantity of gas.
Although different gases have been employed in the present experiments,
there has been no attempt to compare their viscosities, and indeed such
a comparison would be difficult to carry out by this method. The question
has been, how is the viscosity of a given gas affected by a change of tempera-
ture ? In one set of experiments the capillary is at the temperature of the
room; in a closely following set the capillary is bathed in saturated steam
at a temperature that can be calculated from the height of the barometer.
If the temperature were changed throughout the whole apparatus from
one absolute temperature 0 to another absolute temperature 6' ' , we could
make immediate application of (4) ; the viscosities (/A, /*') at the two tempera-
tures would be directly as the times of transpiration (t, t'). The matter is
not quite so simple when, as in these experiments, the change of temperature
takes place only in the capillary. A rise of temperature in the capillary now
acts in two ways. Not only does it change the viscosity, but it increases the
volume of gas which has to pass. The ratio of volumes is 6', 0; and thus
subject to a small correction for the effect of temperature upon the dimensions
of the capillary. It is assumed that the temperature of the reservoirs is the
same in both transpirations.
The apparatus is shown in the figure. The gas flows to and fro between
the bulbs A and B, the flow from A to B only being timed. It is confined by
mercury, which can pass through U connexions of blown glass from A to C
and from B to D. The bulbs B, C, D are supported upon their seats with
a little plaster of Paris. The capillary is nearly 5 feet (150 cm.) in length
and is connected with the bulbs by gas tubing of moderate diameter, all
joints being blown. E represents the jacket through which steam can be
passed ; its length exceeds that of the capillary by a few inches.
In order to charge the apparatus, the first step is the exhaustion. This
is effected through the tap, F, with the aid of a Topler pump, and it is
necessary to make a corresponding exhaustion in C and D, or the mercury
would be drawn over. To this end the rubber terminal H is temporarily
connected with 0, while / leads to a common air-pump. When the exhaustion
is complete, the gas to be tried is admitted gradually at F, the atmosphere
being allowed again to exert its pressure in G and D. When the charge is
sufficient, F is turned off, after which G remains open to the atmosphere, and
H is connected to a manometer.
456 ON THE VISCOSITY OF ARGON AS AFFECTED BY TEMPERATURE. [254
When a measurement is commenced, the first step is to read the tempera-
tures of the bulbs and of the capillary ; / is then connected to a force pump,
and pressure is applied until so much of the gas is driven over that the
mercury below A and in B assumes the positions shown in the diagram.
/ is then suddenly released so that the atmospheric pressure asserts itself
in D, and the gas begins to flow back into B. The bulb / allows the flow
a short time in which to establish itself before the time measurement begins
as the mercury passes the connexion passage K. When the mercury reaches
L, the time measurement is closed.
One of the points to be kept in view in designing the apparatus is to
secure long enough time of transpiration without unduly lowering the driving
pressure. At the beginning of the measured transpiration the pressure
in A was about 30 cm. of mercury above atmosphere, and that in B about
2 cm. below atmosphere. At the end the pressure in A was 20 cm., and
in B 3 cm., both above atmosphere. Accordingly the driving pressure fell
from 32 to 17 cm.
1900] ON THE VISCOSITY OF ARGON AS AFFECTED BY TEMPERATURE. 457
Three, or, in the case of hydrogen, five, observations of the time were
usually taken, and the agreement was such as to indicate that the mean
would be correct to perhaps one-tenth of a second. The time for air at the
temperature of the room was about ninety seconds, and for hydrogen forty-four
seconds, but these numbers are not strictly comparable.
When the low temperature observations were finished, the gas was lighted
under a small boiler placed upon a shelf above the apparatus, and steam was
passed through the jacket. It was necessary to see that there was enough
heat to maintain a steady issue of steam, yet not so much as to risk a sensible
back pressure in the jacket. The time of transpiration for air was now about
139 seconds. Care was always taken to maintain the temperature of the
bulbs at the same point as in the first observations.
There are one or two matters as to which an apparatus on these lines
is necessarily somewhat imperfect. In the high temperature measurements
the whole of the gas in the capillary is assumed to be at the temperature
of boiling water, and all that is not in the capillary to be at the temperature
of the room, assumptions not strictly compatible. The compromise adopted
was to enclose in the jacket the whole of the capillary and about 2 inches
at each end of the approaches, and seems sufficient to exclude sensible error
when we remember the rapidity with which heat is conducted in small spaces.
A second weak point is the assumption that the instantaneous pressures are
represented by the heights of the moving mercury columns. If the connecting
U -tubes are too narrow, the resistance to the flow of mercury enters into
the question in much the same way as the flow of gas in the capillary. In
order to obtain a check upon this source of error the apparatus has been
varied. In an earlier form the connecting U -tubes were comparatively
narrow; but the result for the ratio of viscosities of hot and cold air was
substantially the same as that subsequently obtained with the improved
apparatus, in which these tubes were much widened. Even if there be a
sensible residual error arising from this cause, it can hardly affect the com-
parison of temperature -coefficients of gases whose viscosity is nearly the same.
I will now give an example in detail from the observations of December 21
with purified argon. The times of transpiration at the temperature of the
room (15° C.) were in seconds
104f , 104£, 104|. Mean, 104'67.
When the capillaries were bathed in steam, the corresponding times were
167$, 167i, 167f. Mean, 167-58.
The barometer reading (corrected) being 767'4 mm., we deduce as the
temperature of the jacket 100'27° C. Thus 6 = 287'5, 0' = 372'8. The re-
duction was effected by assuming
458 ON THE VISCOSITY OF ARGON AS AFFECTED BY TEMPERATURE. [254
With the above values we get
#=1-812.
As appears from (5), the integral part of x relates merely to the expansion
of the gas by temperature. If we take
we get n = 0*812.
This number is, however, subject to a small correction for the expansion
of the glass of the capillary. As appears from (4), the ratio //, /* as used
above requires to be altered in the same ratio as that in which the glass
expands by volume. The value of n must accordingly be increased by O'OIO,
making
n = 0-822.
The following table embodies the results obtained in a somewhat extended
series of observations. The numbers given are the values of n in (7), corrected
for the expansion of the glass.
Air (dry) 0754
Oxygen 0782
Hydrogen 0'681
Argon (impure) .... 0'801
Argon (best) 0'815
In the last trials, the argon was probably within 1 or 2 per cent, of
absolute purity. The nitrogen lines could no longer be seen, and scarcely any
further contraction could be effected on sparking with oxygen or hydrogen.
It will be seen that the temperature change of viscosity in argon does
not differ very greatly from the corresponding change in air and oxygen.
At any rate the simpler conditions under which we may suppose the collisions
to occur, do not lead to values of n such as 0'5, or I'O, discussed by theoretical
writers.
I may recall that, on a former occasion*, I found the viscosity of argon
to be 1-21 relatively to that of air, both being observed at the temperature
of the room.
[1902. See further, Vol. iv. p. 481.]
* Roy. Soc. Proc. January, 1896. [Vol. iv. p. 222.]
255.
ON THE PASSAGE .OF ARGON THROUGH THIN FILMS OF
INDIARUBBER.
[Philosophical Magazine, XLix. pp. 220, 221, 1900.]
SOON after the discovery of Argon it was thought desirable to compare
the percolation of the gas through indiarubber with that of nitrogen, and
Sir W. Roberts- Austen kindly gave me some advice upon the subject. The
proposal was simply to allow atmospheric air to percolate through the rubber
film into a vacuum, after the manner of Graham, and then to determine the
proportion of argon. It will be remembered that Graham found that the
percentage of oxygen was raised in this manner from the 21 of the atmo-
sphere to about 40. At the time the experiment fell through, but during
the last year I have carried it out with the assistance of Mr Gordon.
The rubber balloon was first charged with dry boxwood sawdust. This
rather troublesome operation was facilitated by so mounting the balloon that
with the aid of an air-pump the external pressure could be reduced. When
sufficiently distended the balloon was connected with a large Tb'pler pump,
into the vacuous head of which the diffused gases could collect. At intervals
they were drawn off in the usual way.
The diffusion was not conducted under ideal conditions. In order to
make the most of the time, the apparatus was left at work during the night,
so that by the morning the internal pressure had risen to perhaps three
inches of mercury. The proportion of oxygen in the gas collected was deter-
mined from time to time. It varied from 34 per cent, when the vacuum
was bad to about 39 per cent, when the vacuum was good. On an average
it was estimated that the proportion of oxygen would be about 37 per cent,
of the whole. The total quantity of diffused gas reckoned at atmospheric
pressure was about 300 c.c. per twenty-four hours.
460 PASSAGE OF ARGON THROUGH THIN FILMS OF INDIARUBBER. [255
On removal from the pump the gas was introduced into an inverted flask
standing over alkali, and with addition of oxygen as required was treated
with the electrical discharge from a transformer in connexion with the public
supply of alternating current. In this way the nitrogen was gradually
oxidized and absorbed. Towards the close of operations the gas was trans-
ferred to a smaller vessel, where it was further sparked until no further
contraction occurred, and the lines of nitrogen had disappeared from the
spectrum. The excess of oxygen was then removed by phosphorus.
It remains only to record the final figures. The residue, free of oxygen
and nitrogen, from 3205 c.c. of diffused gas was 39 c.c. The most instructive
way of stating the result is perhaps to reckon the argon as a percentage, not
of the whole, but of the nitrogen and argon only. Of the 3205 c.c. total,
2020 c.c. would be nitrogen and argon, and of this the 39 c.c. argon would
be 1*93 per cent. Since, according to Kellas (Proc. Roy. Soc. Vol. LIX. p. 67,
1895), 100 c.c. of mixed atmospheric nitrogen and argon contains T19 per cent,
of argon, we see that in the diffused gas the proportion of argon is about half
as great again as in the atmosphere. Argon then passes the indiarubber film
more readily than nitrogen, but not in such a degree as to render the diffusion
process a useful one for the concentration of argon from the atmosphere.
256.
ON THE WEIGHT OF HYDROGEN DESICCATED BY
LIQUID AIR.
[Proceedings of the Royal Society, LXVI. p. 344, 1900.]
IN recent experiments by myself and by others upon the density of
hydrogen, the gas has always been dried by means of phosphoric anhydride ;
and a doubt may remain whether on the one hand the removal of aqueous
vapour is sufficiently complete, and on the other whether some new impurity
may not be introduced. I thought that it would be interesting to weigh
hydrogen dried in an entirely different manner, and this I have recently been
able to effect with the aid of liquid air, acting as a cooling agent, supplied
by the kindness of Professor Dewar from the Royal Institution. The opera-
tions of filling and weighing were carried out in the country as hitherto.
I ought, perhaps, to explain that the object was not so much to make a new
determination of the highest possible accuracy, as to test whether any serious
error could be involved in the use of phosphoric anhydride, such as might
explain the departure of the ratio of densities of oxygen and hydrogen from
that of 16 : 1. I may say at once that the result was negative.
Each supply consisted of about 6 litres of the liquid, contained in two
large vacuum-jacketed vessels of Professor Dewar's design, and it sufficed
for two fillings with hydrogen at an interval of two days. The intermediate
day was devoted to a weighing of the globe empty. There were four fillings
in all, but one proved to be abortive owing to a discrepancy in the weights
when the globe was empty, before and after the filling. The gas was exposed
to the action of the liquid air during its passage in a slow stream of about
half a litre per hour through a tube of thin glass.
I have said that the result was negative. In point of fact the actual
weights found were ^ to -^ milligrams heavier than in the case of hydrogen
dried by phosphoric anhydride. But I doubt whether the small excess is of
any significance. It seems improbable that it could have been due to residual
vapour, and it is perhaps not outside the error of experiment, considering
that the apparatus was not in the best condition.
257.
THE MECHANICAL PRINCIPLES OF FLIGHT.
[Manchester Memoirs, XLIV. pp. 1 — 26, 1900.]
THE subject under discussion includes both natural and artificial flight.
Although we are familiar with the flight of birds, there are many interesting
questions which arise in connexion with natural flight, and some of them are
yet very obscure.
In still air a bird, being heavier than the fluid displaced, cannot maintain
his level for more than a short time without working his wings. In this
matter the vicarious principle holds good. If the bird is not to fall, some-
thing must fall instead of him, and this can only be air. The maintenance
of the bird thus implies the perpetual formation of a downward current of
air, and involves therefore performance of work. Later we shall consider
more particularly how this work is applied; but a preliminary difficulty
remains to be discussed. It is well known that large birds, such as vultures
and pelicans, are often observed to maintain their level for considerable
periods of time, without flapping or visibly working their wings. On a
smaller scale, and in more special situations, sea-gulls in these latitudes
perform similar feats. This question of the soaring or sailing flight of birds
has given rise to much difference of opinion. Few of the naturalists, to
whom we owe the observations, are familiar with mechanical principles, and
thus statements are often put forward which amount to mechanical impossi-
bilities. The arm-chair theorist at home, on the other hand, may be too
willing to discredit reports of actual observations, especially when they are
made in other parts of the world. On both sides it seems to be admitted
that there is no sailing flight in the absence of wind ; but observers, un-
trained in dynamics and misled by the analogy of the kite, are apt to suppose
that the existence of wind at once removes the difficulty. The doctrine of
relative motion shows however that, so long as there is no connexion with
the ground, a uniform horizontal wind is for this purpose the same thing as
absolutely still air.
1900] THE MECHANICAL PRINCIPLES OF FLIGHT. 463
In a short paper upon this subject (Nature, xxvu. p. 534, 1883 [Vol. n.
p. 194]) I pointed out that, " Whenever a bird pursues his course for some
time without working his wings, we must conclude either (1) that the course
is not horizontal, (2) that the wind is not horizontal, or (3) that the wind is
not uniform. It is probable that the truth is usually represented by (1) or
(2), but the question I wish to raise is whether the cause suggested by (3)
may not sometimes come into operation." Case (1) is that of a rook gliding
downwards from a tree in still air with motionless wings. We shall presently
consider upon what conditions depend the time and distance of travel possible
with a given descent. Case (2) is closely related to case (1). If the air have
an upward velocity equal to that at which the rook falls through it in a
vertical direction, the vertical motion is compensated, and the course of the
rook relatively to the ground becomes horizontal. It is not necessary, of
course, that the whole motion of the air be upwards ; a horizontal motion of
the air is simply superposed. A bird gliding into a wind having a small
upward component may thus maintain relatively to the ground an absolutely
fixed position, or he may advance over the ground to windward at a fixed
level.
There can be no doubt that the vertical component of wind plays a large
part, not merely in the flight of birds, but in general atmospheric phenomena.
Living at the bottom of the atmospheric ocean, where the wind is necessarily
parallel to the ground, we are liable to overlook the importance of vertical
motions. This is the more remarkable when we consider that wind is due to
atmospheric expansion and condensation, so that the primary movements are
vertical and not horizontal. Thus the inhabitants of an oceanic island are
specially interested in the so-called land and sea breezes, but the primary
phenomenon is the rise and fall of air over the island as it is heated by the
sun during the day and cooled by radiation at night.
A recent American observer (Huffaker, Smithsonian Report for 1897) has
recorded many examples of vultures soaring under circumstances which sug-
gested that they take advantage of the upward currents which rise locally
from the ground when it is strongly heated by the sun. On dull days and
in light winds the vultures were not seen to soar. There is no doubt that
under the influence of a strong sun the layers of air near the ground approach
an unstable condition, and that comparatively slight causes may determine
local upward currents. Mr Huffaker suggests that in some cases the birds
themselves, by flying round, may determine the upward current. Some of
his observations certainly point in this direction ; but it must be remembered
that the immediate effect of flight will be a downward and not an upward
current.
The more obvious examples of upward motion occur when an otherwise
horizontal wind meets an obstruction. Some years ago I visited the north
464 THE MECHANICAL PRINCIPLES OF FLIGHT. [257
side of Madeira, where cliffs, nearly 2,000 feet high, rise perpendicularly
from the sea. Being on the top of the cliff, we had difficulty in finding
a sheltered spot until we noticed that close to the edge there was almost com-
plete calm. Lying upon the ground and moving only one's arms, it was
possible to hold a handkerchief by the corner so that a little behind the
plane of the cliff it hung downwards as in still air, and a little in front of the
cliff was carried upwards in the vertically rising stream. A ball of crumpled
paper thrown outwards was carried up high over our heads. Of course gulls
and other birds found no difficulty in rising up the face of the cliff without
working their wings. During a recent visit to India, I frequently watched
the effect of similar upward currents deflected by rocky fortresses which rise
from the plains. Kites could be seen to maintain themselves for minutes
together without a single flap of the wings. When this occurred, the birds
were sailing to and fro over the windward side of the rock.
We now turn to the consideration of case (3).
"In a uniform wind the available energy at the disposal of the bird
depends upon his velocity relatively to the air about him. With only a
moderate waste this energy can at any moment be applied to gain elevation,
the gain of elevation being proportional to the loss of relative velocity
squared. It will be convenient for the moment to ignore the waste referred
to, and to suppose that the whole energy available remains constant, so that
however the bird may ascend or descend, the relative velocity is that due to
a fall from a certain level to the actual position, the certain level being of
course that to which the bird might just rise by the complete sacrifice of
relative velocity."
In illustration of case (3) I instanced a wind blowing everywhere hori-
zontally but with a velocity increasing upwards, taking for the sake of
simplicity the imaginary case of a wind uniform above and below a certain
plane where the velocity changes. Since a uniform motion has no effect, we
may suppose without further loss of generality, that the velocities of the
wind above and below the plane are + u and — u. Let us consider how a
bird, sailing somewhat above the plane of separation and endowed with an
initial relative velocity v, might take advantage of the position in which he
finds himself.
The first step is, if necessary, to turn round until the relative motion is
down wind (in the upper stratum) and then to drop through the plane of
separation. In falling down to the level of the plane there is a gain of
relative velocity, but this of no significance for the present purpose, as it is
purchased by the loss of elevation ; but in passing through the plane there
is a really effective gain. In entering the lower stratum the actual velocity
is indeed unaltered, but the velocity relatively to the surrounding air (moving
in the opposite direction) is increased.
1900] THE MECHANICAL PRINCIPLES OF FLIGHT. 465
If h denote the height above the plane of separation to which the initial
relative velocity v is due, we have v2 = *2gh. Here v is the velocity, relatively
to the air in the upper stratum, with which the bird crosses the plane.
After crossing, the velocity, now reckoned relatively to the air in the lower
stratum, becomes v + Zu, and the new value of h is given by
2gh' = (v + 2w)2,
so that 2g (h' — h) = 4<uv + 4u2 = 4<u (u + v).
Here (h' — h) is the gain of potential elevation and, if u is given, it increases
as v increases.
At this stage the bird is moving against the direction of the wind in the
lower stratum. He next turns round — it is supposed without loss of relative
velocity — until his direction is reversed so as to be with the wind of the
lower stratum and contrary to the wind of the upper stratum. A passage
upwards through the plane now secures another gain of relative velocity,
or of potential elevation, of nearly the same value as before. The process
may be repeated. At every passage through the plane (whether in the
upwards or in the downwards direction) there is a gain of potential elevation,
and if this gain outweighs the losses all the while in progress, the bird may
maintain or improve his position without doing a stroke of work.
It may be of interest to consider a numerical example.
Suppose that
v = 30 miles per hour = 1'34 x 10s cm. per second,
and that h' - h = 10 feet = 305 cm. ;
then in C.G.s. measure
(v + 2uJ> = v* + 2g (Ji - h) = T80 x 10B + -60 x 10" = 2'40 x 106,
and w + 2w = 1-55 x 10s;
so that 2w = '21 x 103 cm./sec. = 47 miles/ hour.
In this case a freshening of the wind amounting to 4'7 miles per hour
is equivalent to a gain of 10 feet of potential elevation.
In order to take advantage of the gradual increase of wind with elevation
usually to be met with, a bird may describe circles in an inclined plane,
always descending when moving to leeward and ascending when moving to
windward. Whether the differences of velocity available at considerable
elevations in the atmosphere are sufficient to allow a bird to maintain his
position without working his wings appears to be doubtful. Near the level
of the ground or sea these differences are greater, and probably suffice to
explain much of the sailing flight of albatrosses and other sea-birds.
R. iv. 30
466 THE MECHANICAL PRINCIPLES OF FLIGHT. [257
Another way in which a bird may draw upon the internal energy of the
wind has been specially discussed by Dr Langley (Smithsonian Contributions,
1893), who calls attention to the fact that the well-known gustiness of the
wind, at any rate near the earth's surface, is underestimated in the usual
meteorological records. The differences of horizontal velocity involved in
what are commonly called gusts of wind imply in general vertical motions
also, but near the ground these latter may, perhaps, be left out of account.
The advantage which a bird may take of the variations in the speed of the
wind is explicable upon the principles already applied, the inertia of the
bird playing in some sort the part of the string of a kite.
If u denote the speed of the wind at any moment, and v the speed of the
bird in the opposite direction, both e.g., reckoned relatively to the ground,
the available energy is measured by ^ (v + uf. Suppose now that the wind
freshens, u becoming u + du, while v remains constant. The increment of
available energy is
k(v + u + duf - £ 0 + uf = (v + u) du ;
rt
or in time t, I (v + u)du (1)
.'o
The speed of the wind being supposed to be periodic, and the integration
being taken over a sufficiently long period of time, we have
and thus the mechanical advantage may be reckoned as
vdu (2)
In order that this may have a finite value, v must vary ; the principle
being that to get the most advantage v must be great when du is positive,
that is when the wind is freshening, and smaller when the wind is failing.
The higher velocity required to meet the freshening wind is to be obtained
by a previous fall to a lower level.
As an example, let us suppose that u and v are periodic, so that
« = M0 + ui sin pt, v = va + vl cos (pt + e) ;
then I vdu = pu^v^ I cos pt . cos (pt + e) dt,
and, when t is great, Ivdu = ^pttt1vlcose ............................ (3)
The mechanical advantage obtained in time t is greatest when e vanishes,
i.e., when du and v are in the same phase. This mechanical advantage is to
1900] THE MECHANICAL PRINCIPLES OF FLIGHT. 467
be set against the frictional and other losses neglected in our original suppo-
sition. Were there no such losses, the value of v, or of the elevation, might
continually increase.
This example shows that it is quite possible for a bird moving in a very
natural manner against a strong and variable wind to maintain himself and
to advance over the ground without working his wings. Observations of this
kind are recorded by Mr Huffaker. It will be understood, of course, that
a bird, not being interested in simplifying the calculation, will take any
advantage that offers itself of the internal energy of the wind and of upward
currents in order to attain his objects.
In the preceding discussions we have assumed, for the sake of simplicity,
that a bird or a flying machine is able to glide in still air without loss of
energy. It is needless to say that the truth of such an assumption can,
at best, be only approximate. Apart from frictional losses, the maintenance
of a given level implies the continual formation of a downward aerial current,
and consequent expenditure of energy. We have next to consider the mag-
nitude of these losses, taking the case of a plane moving at a uniform speed.
And, in the first instance, we shall neglect the frictional forces, assuming that
the reaction of the air upon the plane is truly normal.
Before we can advance a step in the desired direction we must know how
the normal pressure upon an aeroplane is related to the size and shape of the
plane, to the velocity of the motion, and above all to the angle between the
plane and the direction of motion. According to an erroneous theory, to
some extent sanctioned by Newton, the mean pressure would depend only
upon the area of the plane and the resolved part of the velocity in a direction
perpendicular to the plane. If V be the velocity, a the angle between V and
the plane, p the density of the air (or other fluid concerned), the pressure p
would be given by
(4)
That this formula is quite erroneous, especially when a is small, has long
been known*. At small angles the pressure is more nearly proportional to
sin a than to sin2 a and, as was strongly emphasized by Wenham in an early
and important paper on aerial locomotion f, the question of shape and pre-
sentation is by no means indifferent. In the case of an elongated shape
moving with given velocity V, and at a given small inclination a, the pressure
is much greater when the long dimension of the plane is perpendicular than
when it is (nearly) parallel to V.
* A further discussion will be found in Phil. Mag. Vol. n. p. 430, 1876 ; Scientific Papers,
Vol. i. p. 287; and in Nature, Vol. XLV. p. 108, 1892. [Vol. in. p. 491.]
t Report of Aeronautical Society, 1866, p. 10.
30—2
468 THE MECHANICAL PRINCIPLES OF FLIGHT. [257
According to a theoretical formula developed on the basis of Kirchhoff's
analysis (Phil. Mag. loc. cit.) we should have for the mean pressure, instead
of (4),
Trsma
v 4 + TT sm a H
This applies strictly to motion in two dimensions, or practically to the case
of a very elongated blade, whose length is perpendicular to V.
At perpendicular incidence (a = 90°) the difference between (4) and (5) is
not important ; but when a is small, the value of p in (5) may be enormously
greater than the corresponding value from (4).
As regards numerical values, if we use C.G.S. measure, so that V is
measured in centimetres per second, we have in the case of air under standard
conditions p = '00128, and p, at perpendicular incidence, measured in dynes
per square centimetre, is according to (4),
^ = •000641^ (6)
This does not differ greatly from the data given in engineering tables. To
compare with Langley's more recent experiments, we may express V in
metres per second and p in grams weight per square centimetre. Thus
p' = -0065F'2; (7)
while the mean of Langley's numbers gives
p' = -0087F'2, (8)
about 30 per cent, greater. The difference is accounted for, at any rate
partly, by the suction, which experiment shows to exist at the back of the
plate.
As regards the law of obliquity, the early experiments of Vince (1798)
sufficed to show that the effect was more nearly as sin a than as sin2 a. In
recent times this subject has been very thoroughly investigated by Langley,
who has examined not only the influence of obliquity, but also of the shape
and presentation of the plane. His results for the case to which (5) relates
indicate an even greater relative effect at small angles, probably referable to
the back suction. A laboratory experiment to demonstrate the reality of
this suction was described in one of the papers already referred to (Nature,
loc. cit.).
Experiments upon the law of obliquity, as executed for the case of air, by
Dines* and Langley f, involve cumbrous and costly whirling machines, and
if made in the open are greatly embarrassed by wind. An apparatus capable
* Proc. Roy. Soc. June, 1890.
t Smithsonian Contributions to Knowledge, 1891.
1900] THE MECHANICAL PRINCIPLES OF FLIGHT. 469
of working in the laboratory, or as a lecture illustration, has long been a
desideratum. With the aid of Mr Gordon I have recently constructed one
which, while very simple and inexpensive, performs sufficiently well. It may
be regarded as a kind of adjustable windmill. An axis of hard steel, finely
pointed at the ends, is carried by agate cups. From a central boss six spokes
of round steel project symmetrically, carrying at their ends six similar vanes
of tin-plate. The vanes are provided with projecting sockets of brass tubing,
which fit the spokes somewhat tightly, but yet allow the vanes to be rotated
when desired. The vanes are 4 inches long and !•£ inches wide, the distance
of their inner ends from the axis being about 3*7 inches. The whole appa-
ratus is as light as may be (about 120 gm.) consistently with the necessary
rigidity.
If the vanes are all inclined at the same angle, the apparatus works like
an ordinary windmill, and may be set into rapid rotation by a motion through
the air parallel to the axis. This motion may take place either in a horizontal
or in a vertical direction. If means were provided for estimating the couple
needed to prevent rotation, we should obtain the efficiency of the vanes at
the given obliquity and speed. Observations at the same speed and at other
obliquities would then give the means of determining the law of obliquity.
Such a procedure would be analogous to that adopted in former ex-
periments with whirling machines. The essential feature of the present
method consists in setting some of the vanes to compensate others inclined
at different angles. The balance of effects is independent of the speed of the
wind, so long as it is uniform over the whole section in operation. To guard
against errors that might arise from a deficient fulfilment of this condition,
I have preferred so to arrange that opposite vanes were inclined always at
the same angle. For example, two pairs of opposite vanes might be set so
that their planes make an angle of 6° with the axis. The remaining pair
of opposite vanes would then be set at a greater angle, and this would be
varied until no tendency remained to turn in either direction. The exact
point of balance could be inferred either from the absence of observable
effect, or by interpolation from equal slight effects in opposite directions.
As has been suggested, the motion itself may be either horizontal or
vertical. Fair results may be obtained indoors at a walking speed, and my
first idea was to determine balances by holding the wheel overhead while
travelling in a dog-cart at 10 or 12 miles per hour. But when the axis is
horizontal, much time is lost owing to the necessity of readjusting the centre
of gravity after almost every shifting of the vanes. With a nearly vertical
motion the position of the centre of gravity is of less consequence, and it
was found that very good results could be arrived at by somewhat rapidly
lowering the apparatus while held in the hands with axis vertical. It is
possible that part of the delicacy obtained in this way is due to a partial
470
THE MECHANICAL PRINCIPLES OF FLIGHT.
[257
annulment of gravity during the downward acceleration and consequent
diminution of frictional effect at the bearings.
Some of the observations presently to be discussed were made in this
way, but in most of them the arrangement was rather different. The wheel
was removed from its bearings and suspended by a fine wire, whose torsion
was insufficient to check the rotation seriously. The wire was pulled up
vertically by a cord running over a pulley overhead. Although this arrange-
ment offered some advantages, they were largely neutralised by disturbances
due to draughts; and it is probable that equally good balances might be
obtained by the simpler method.
According to an old and long discredited law, the normal pressure upon
a vane moving through the air at given speed would be proportional to the
square of the sine of the angle (a) between the plane of the vane and the
direction of motion. The resolved part of this in the direction of rotation
would be sin2 a cos a, which expression would represent the efficiency of the
vanes of our mill as dependent upon the angle of setting. When a is small,
the second factor is of little importance. A very simple experiment will
now decide whether the law of sin2 a is, or is not, an approximation to the
truth. We find, in fact, that four vanes set at 6° markedly overpower two
vanes set at 9°, whereas according to the law of sin2 a the reverse should
happen. In order to balance the four vanes at 6°, the two vanes need to
be at about 14|°.
By observations of this kind materials are collected for a complete plotting
out of the curve of efficiency. The efficiency necessarily vanishes when
a = 0, and also on account of the resolving factor, when a = 90°. In order to
balance four vanes set at 5°, we may set the remaining two vanes either at
10£° or at about 58°. The efficiency reaches a maximum in the neighbour-
hood of 27°. The results are shown in A (Fig. 1), or in the second column
of the accompanying table. The scale of the ordinates is, of course, arbitrary.
The efficiency for 5° is assumed to be 10.
a
Rotatory E fficiency
Normal Pressure
a
Rotatory Efficiency
Normal Pressure
0
o-o
0
40
27-0
88
5
10-0
25
50
23-5
91
10
19-0
48
60
19-0
94
15
24-8
64
70
13-2
96
20 28-0
75
80
6-9
99
25 29-2
80
90
o-o
100
30
29-0
84
:
In order to deduce the normal pressure, the results for the rotatory
efficiency must be divided by cos a, and accuracy is necessarily lost in the
1900]
THE MECHANICAL PRINCIPLES OF FLIGHT.
471
case of the larger angles. The numbers thus arrived at are plotted in
curve B, and are given in the third column of the table, reduced so as to
make the maximum (at 90°) equal to 100. As regards the relative pressures
at the smaller angles, the results appear to be at least as accurate as those
obtained on a larger scale with the whirling machine ; but the reference to
the pressure operative at 90° is probably less accurate. The principal con-
clusion that at small angles the pressure is proportional to sin a, and by no
means to sin2 a, is abundantly established.
472 THE MECHANICAL PRINCIPLES OF FLIGHT. [257
In applying these results, the first problem which suggests itself for
solution is that of the gliding motion of an aeroplane. It was first success-
fully treated by Penaud*, and it may be taken under slightly different forms.
We may begin by supposing the motion to be strictly horizontal, the velocity
being V and the inclination of the plane to the horizon being a. Under these
circumstances a propelling force F is required, which we suppose to act
horizontally. The mean pressure upon the plane we will denote by /cV'2sma,
the assumption of proportionality to sin a being amply sufficient for the case
of small angles, with which alone we are practically concerned. If S be the
area of the plane, the whole normal force is /c$F2sina. In view of the
smallness of a, we may equate this to the weight (W) supported. Thus
W = /eSF2sina, (9)
also F=KSV*am*a (10)
If F be independent of V, as approximately in the method of rocket
propulsion, these equations show at once that there is no limit to the weight
that may be supported by a given F. It is only necessary to make a small
enough, and to take V large enough to satisfy (9).
In other methods of propulsion we should have to do rather with the
rate (H) at which energy is expended than with the force F itself. The
relation is
H = FV, (11)
so that in place of (10) H= tcSV3 sin3 a (12)
Or, again, since in many cases the power that might be expended is pro-
portional to the weight lifted, we may conveniently write
H= WU. (13)
From these equations we derive
W* W
V- - - —
~KSH~KSU' '•
and it is possible so to determine F and a that, with a given U and a given
S, any weight W can be supported. As W increases, F must be greater and
a smaller. The same is true, in an enhanced degree, if it be H that is given
in place of U.
According to what has been shown (6), (7), (8), Fig. 1, we have in C.G.s.
measure
K sin 5° = -25 x -00085,
so that « = '0024 (16)
* Societe Philomathique de Paris, 1876; Report of Aeronautical Society, 1876. See also
W. Froude, Glasgow Proceedings, VoL xvm. p. 65, 1891.
1900] THE MECHANICAL PRINCIPLES OF FLIGHT. 473
In the case of a very elongated plane the value of K would be a little
higher. We must remember that V is reckoned in centimetres per second,
S in square centimetres, and the normal force in dynes.
The conclusion that a weight, however great, may be supported with a
given S and a given U, or even a given H, is unpractical for more than one
reason. There must be a limit below which a cannot be reduced, if only
because of the high degree of instability that such an adjustment must have
to contend with. Another important matter is the tangential force upon the
plane, although some distinguished experimenters have expressed the opinion
that it is negligible. In order to take account of it, we may add to the
right-hand member of (10) a term proportional to V2, but independent of a.
Thus (12) becomes
H = WU=(KSsm2a + n) V3, ..................... (17)
(9) remaining unchanged. Eliminating V, we find
W
We may apply (18) to find for what value of sin a the quantity t/"2 attains a
minimum. By the ordinary rules,
sin-a^^g, (19)
and, of course, this value of sin3 a must be small, if the investigation is to be
applicable. If //- vanish, sin a diminishes without limit. In general the
minimum value of U2 is given by
WV> » * ...(20)
and the corresponding value of F2 by
^2 = OA ..A ..i vu (21)
These equations show that the necessary work depends entirely upon /*, and
that without a knowledge of this element no numerical conclusions can be
arrived at.
It might be supposed that /JL, so far as it depends upon the aeroplane,
would be proportional to S, but this relation is more than doubtful. In any
case of a practical machine there must at any rate be a part of p not
proportional to 8.
It may be well to recall that U represents the velocity at which a weight
equal to W would have to be raised in order to do work equal to that done
by the propelling force F. By (20), caeteris paribus, U varies as /S>~*.
We may now pass to the case of an aeroplane gliding in still air, the path
being slightly inclined downwards. If 0 be the small angle between the
474 THE MECHANICAL PRINCIPLES OF FLIGHT. [257
path and the horizontal, we may regard the component of gravity in this
direction, viz., W sin 6, as the propelling force F. Thus
..................... (22)
so that U=VsmO ............................... (23)
The same equations apply as before, with the understanding that a, being
the inclination of the plane to the direction of motion through the air, is no
longer identical with the inclination of the plane to the horizon. The latter
angle, reckoned positive when the leading edge is downwards, will now be
denoted by (6 - a).
Introducing (23) into (14), (15), we get
W «SF2sm2tf
V* = —~-r—6, sma= --- -Wr— -, ............ (24)
/cS sm 0
from which it appears that whatever may be the values of W and 8, 0 may
still be as small as we please. Thus, if frictional forces can be neglected, a
high speed is all that is required in order to glide without loss of energy.
This is the supposition upon which we discussed the manner in which a bird
may take advantage of the internal work of the wind ; and we see that the
motion of the bird must be of such a character that he always retains a high
velocity relatively to the surrounding air. The advantage that we showed to
be obtainable must be set against losses due to friction and to imperfect
fulfilment of the condition just specified.
When frictional forces are included we may use equation (18), merely
substituting V sin 0 for U. The problem already considered of making U a
minimum is still pertinent, since U denotes the rate of vertical descent. By
(19), (20), (21)
3/A . na U2 16/i, /ocv
-"'«-;& ^'-Ti-ia ................ (25)
so that, 0 and a being small,
a = f0, 0-a = £a = i0 ...................... (26)
This result, due to Penaud, shows that when the rate of vertical descent is
slowest, or when the time of falling a given height is greatest, the slope of
the plane to the horizon is downwards in front, and equal to one-quarter of
the slope of the line of motion. The actual minimum rate of vertical descent
is given by (20). This rate is relative to still air. If there be a wind having
a vertical component of the same amount, the course of the plane -may be
horizontal.
Another slightly different minimum problem is also treated by Pe'naud,
in which it is required to determine how far it is possible to glide while
1900] THE MECHANICAL PRINCIPLES OF FLIGHT. 475
falling through a given vertical height. From (9), (17), (23), we have in
general
s\n0 = S^« + ^KS ......................... (27)
sin a.
When 0 is a minimum by variation of a,
sin a = i sin 0 = VO*/*S) ...................... (28)
In this case the plane bisects the angle between the horizontal and the
direction of motion.
In the flying machines of Penaud, Langley, and Maxim, the propelling
force is obtained by a screw, acting like the screw-propeller of a ship.
A rough theory of this action is easily given and is of interest, not only in
the application to the horizontal propulsion of an aeroplane, but also because
a screw rotating about a vertical axis may be used for direct maintenance.
The latter question may conveniently be considered first.
The screw is supposed to maintain a weight W at a fixed position in still
air. This it does by creating a downward current of velocity v. If $' be the
area of section of the current, equal to that swept through by the screw, the
volume of air acted upon per second is S'v, and the momentum generated
per second is S'v.pv, or S'pv2. Hence
W = S'pv* ............................... (29)
Again, the kinetic energy generated per second is ^S'ptf; so that if U be the
velocity at which W would have to be lifted to do a corresponding amount of
work, we may, neglecting frictioiial losses, equate the above to UW. Thus
(30)
From (29), (30), $v=U,
So far as these equations are concerned, any weight can be maintained by a
limited expenditure of work, but the smaller the power available the larger
must be the section of the stream of air, and consequently of the screw, or
other machinery, by which the air is set in motion. Again from (31)
(32)
so that if S' be given, the whole power required varies as TP*.
To obtain numbers applicable to the case of a man supporting himself in
this way by his own muscular power, we take in c.G.s. measure
W =68000 x 981, U=15, p = sh>
thus finding S' = G'O x 107 sq. cm.
476 THE MECHANICAL PRINCIPLES OF FLIGHT. [257
This represents the cross-section of the descending column of air. If we
equate S' to ^ird2, d will be the diameter of the screw required, and we get
d = 90 metres. It is to be observed that this assumed value of U corresponds
to the power which a man may exercise when working for eight hours a day.
But even if he could do ten times as much for a few minutes, d would still
amount to 9 metres ; and in this estimate nothing has been allowed for the
weight of the mechanism, or for frictional losses. It seems safe to conclude
that a man will never support himself in this manner by his own muscular
power.
A screw works to better advantage when it has a forward motion through
the fluid, for then a larger mass comes under its influence. Let us suppose
that a screw, now rotating about a horizontal axis, is advancing through still
air with horizontal velocity V. Also let v be the actual velocity with which
the column of air leaves it. The volume acted on per second is S'(V + v).
If F be the propulsive force
F=S'p(V + v)v ............................ (33)
Again, the work per second required to generate the kinetic energy of the
column is
%S'p(V+v)v* ............................... (34)
The whole work expended per second (H') is accordingly
) ............. (35)
When V is great compared with v, the right-hand member of (35) reduces to
its first term. We conclude that when a screw advances at a sufficiently
rapid rate, the energy left behind in the fluid is negligible, so that the whole
work done is available for propulsion. The distinction between H ' and H,
as formerly employed, then disappears.
If U denote the rate at which W would have to be lifted in order to do
the work actually performed by the machine, we may now take from (15), as
applicable to the rapid flight of an aeroplane,
(36)
In the case of direct maintenance by a screw rotating about a vertical
axis, (31) gives
It may be interesting to compare the powers required in the two methods,
especially as some high authorities have favoured direct maintenance, without
1900] THE MECHANICAL PRINCIPLES OF FLIGHT. 477
the use of an aeroplane, as the more economical. The ratio of the values of
U in (36), (37) is
or, in the case of air, since K = '0024, p = "0012,
V(2 sin a . S'/S) ............................ (39)
Since a may be made small, and 8 the area of the plane may be a large
multiple of S' the area swept over by the screw, it would appear that the
advantage must lie with the aeroplane, even if the object be mere main-
tenance, and not a rapid transit from place to place.
But although the flying machine of the future will, as it appears to me,
be on the principle of the aeroplane, it cannot be denied that the method
of direct maintenance by a vertically rotating screw offers certain present
advantages. Among the most important of these are a much better ensured
stability, and less danger in alighting owing to the absence of rapid horizontal
motion. The first experiments might well be made with screws driven by
electric motors, the power being supplied from the ground by means of
vertical wires 30 or 40 feet long. In this way the necessary experience
would be easily gained, and most of the doubtful points settled, before a
completely self-contained machine was attempted.
In natural flight revolving mechanism is not, and apparently could not
have been, used. As we all know, a bird flying horizontally through still
air performs the necessary work by flapping his wings. The effect of a
reciprocating motion in modifying the action of an aeroplane was, I believe,
first considered in detail by Professor M. Fitzgerald*. It may be convenient
to give, as naturally connected with the foregoing, an outline of this theory
in a modified form, following Professor Fitzgerald in assimilating the wing
to a simple aeroplane, upon which is imposed (without rotation) a vertical
reciprocating motion.
We denote by u the horizontal velocity of the plane supposed uniform,
by v the vertical velocity at time t, by 6 the inclination of the plane to the
horizon at time t, while S and W denote the area and weight as before. If
we assume the same formula for the pressure as before, although the
application is now to an unsteady motion, and further suppose that v/u
and 0 are always small, we get as in (9) for the whole normal pressure upon
the plane at time t
icS(u? + tf)(e + v/u), ........................ (40)
in which however vz in (w2 + v2) may be omitted.
* Proc. Roy. Soc. Vol. LXIV. p. 420, 1899.
478 THE MECHANICAL PRINCIPLES OF FLIGHT. [257
We now assume that 6 and v are periodic, for example that
0=0a + 01cospt, ........................... (41)
v(u = j3cos(pt + €), ................ ........... (42)
where the periodic time r is related to p according to
At this stage the criticism may present itself that the assumed motion
involves a reaction for which we have made no provision. In practice the
reaction is supplied by the inertia of the body of the bird to which the wings
are attached. The difficulty would be got over by supposing that there are
several planes executing similar movements, but in different phases regularly
disposed. It seems hardly worth while to complicate the present investigation
by introducing a vertical movement of the weight.
By (40) the whole pressure at time t, perpendicular to the plane, is
KSu*{00 + 01cospt + l3cos(pt + €)} ................ (43)
Of this the mean value is to be equated to the weight W supported, so that
W =KSu?e<> ............................... (44)
The horizontal component of the whole pressure at time t is
S.Ku*.{0 + v/u}0, ........................... (45)
and of this the mean value is to be supposed to be zero, in order that the
plane may move with uniform horizontal velocity. Thus
6>o2 + W + I/30J cos e = 0 ...................... (46)
Again, if WU be the (mean) rate of expenditure of work,
*). ...(47)
If we eliminate /3 between (46), (47), we get
**1**'), ............ (48)
from which we see that if 0l be given (as well as S, W, u), U is least when
6 = 0, viz., when the phase of maximum vertical velocity coincides with the
phase of greatest inclination. In this case by use of (44) we have
.(49)
If we regard W, S, u as given, the smallest value of U corresponds to 01 being
large in comparison with 00 which is given by (44)*.
* It must not be forgotten that Ol itself has been assumed to be small.
1900] THE MECHANICAL PRINCIPLES OF FLIGHT. 479
The smallest value is
The work required to be done is here the same function of S, W, and the
horizontal velocity as was found in (14), where V has the meaning here
assigned to u.
We see from (46) that, under the circumstances supposed, 0l + ft is
numerically small in comparison with 00, and a fortiori in comparison with
0j. Accordingly the forward edge of the plane is inclined downwards when
the motion of the plane is downwards.
As regards the pressure, it is by (43) proportional to
in which the second term is relatively small. The pressure acts always upon
the under side of the plane, and the weight is approximately supported in all
phases.
258.
ON THE LAW OF RECIPROCITY IN DIFFUSE REFLEXION.
[Philosophical Magazine, XLIX. pp. 324, 325, 1900.]
IN the current number of the Philosophical Magazine (Vol. XLIX. p. 199)
Dr Wright discusses the question of the amount of light diffusely reflected
from a given area of a matt surface as dependent upon the angle of incidence
(i) and the angle of emission (e). According to Lambert's law the function
of i and e is
cos i cos e ; (1)
and this law, though in the present case without theoretical foundation,
appears approximately to represent the facts. The question may indeed be
raised whether it is possible so to define an ideally matt surface that
Lambert's law may become strictly applicable.
The conclusion drawn by Dr Wright from his experiments with com-
pressed powders upon which I desire to comment is that numbered (4) in his
resume of results, viz. " A law for the intensity of reflected scattered light
cannot be symmetric in reference to i and e." It appears to me that this
statement is in contradiction to a fundamental principle of reciprocity, of
such generality that escape from it is difficult. This principle is discussed
at length in my book on the Theory of Sound, § 109. Its application to the
present question may be thus stated : — Suppose that in any direction (i) and
at any distance r from a small surface (S) reflecting in any manner there be
situated a radiant point (A) of given intensity, and consider the intensity of
the reflected vibrations at any point B situated in direction e and at distance
r' from S. The theorem is to the effect that the intensity is the same as it
would be at A if the radiant point were transferred to B*. The conclusion
follows that whatever may be its character in other respects, the function
of i and e which represents the intensity of the reflected scattered light must
be symmetrical with respect to these quantities.
The actual departures from the reciprocal relation found by Dr Wright
were not very large, and they may possibly be of the nature of experimental
errors. In any case it seems desirable that the theoretical difficulty in
accepting Dr Wright's conclusion should be pointed out.
* I have not thought it necessary to enter into questions connected with polarization, but
a more particular statement could easily be made.
259.
ON THE VISCOSITY OF GASES AS AFFECTED BY
TEMPERATURE.
[Proceedings of the Royal Society, LXVII. pp. 137 — 139, 1900.]
A FOEMER paper* describes the apparatus by which I examined the in-
fluence of temperature upon the viscosity of argon and other gases. I have
recently had the opportunity of testing, in the same way, an interesting
sample of gas prepared by Professor Dewar, being the residue, uncondensed
by liquid hydrogen, from a large quantity collected at the Bath springs. As
was to be expected -f-, it consists mainly of helium, as is evidenced by its
spectrum when rendered luminous in a vacuum tube. A line, not visible
from another helium tube, approximately in the position of D5 (Neon) is also
apparent]:.
The result of the comparison of viscosities at about 100° C. and at the
temperature of the room was to show that the temperature effect was the
same as for hydrogen.
* Roy. Soc. Proc. Vol. LXVI. (1900), p. 68. [Vol. iv. p. 452.]
t Roy. Soc. Proc. Vol. LIX. (1896), p. 207 ; Vol. LX. (1896), p. 56. [Vol. iv. p. 225.]
£ I speak doubtfully, because to my eye the interval from D1 to D3 (helium) appeared about
equal to that between D3 and the line in question, whereas, according to the measurements of
Eamsay and Travers (Roy. Soc. Proc. Vol. LXIII. (1898), p. 438), the wave-lengths are—
D! 5895-0
Z)2 5889-0
D3 5875-9
D5 5849-6,
so that the above-mentioned intervals would be as 19 -1 : 26 -3. [June 23. — Subsequent observations
with the aid of a scale showed that the intervals above spoken of were as 20 : 21. According to
this the wave-length of the line seen, and supposed to correspond to D5, would be about 5855
on Rowland's scale, where D1 = 5896-2, D2 = 5890-2, Z>3 = 5876-0.] I may record that the
refractivity of the gas now under discussion is 0-132 relatively to air.
R. IV. 31
482 ON THE VISCOSITY OF GASES AS AFFECTED BY TEMPERATURE. [259
In the former paper the results were reduced so as to show to what
power (n) of the absolute temperature the viscosity was proportional.
n
c
Air
0754
111-3
Oxygen
0-782
128-2
Hydrogen )
0-681
72-2
Helium (
Argon . ...
0-815
150-2
Since practically only two points on the temperature curve were ex-
amined, the numbers obtained were of course of no avail to determine
whether or no any power of the temperature was adequate to represent the
complete curve. The question of the dependence of viscosity upon tempe-
rature has been studied by Sutherland*, on the basis of a theoretical
argument which, if not absolutely rigorous, is still entitled to considerable
weight. He deduces from a special form of the kinetic theory as the function
of temperature to which the viscosity is proportional
....(I)
c being some constant proper to the particular gas. The simple law #*,
appropriate to " hard spheres," here appears as the limiting form when 6 is
very great. In this case, the collisions are sensibly uninfluenced by the
molecular forces which may act at distances exceeding that of impact.
When, on the other hand, the temperature and the molecular velocities are
lower, the mutual attraction of molecules which pass near one another in-
creases the number of collisions, much as if the diameter of the spheres was
increased. Sutherland finds a very good agreement between his formula (1)
and the observations of Holman and others upon various gases.
If the law be assumed, my- observations suffice to determine the values
of c. They are shown in the table, and they agree well with the numbers
for air and oxygen calculated by Sutherland from observations of Obermayer.
* Phil. Mag. Vol. xxxvi. (1893), p. 507.
260.
REMARKS UPON THE LAW OF COMPLETE RADIATION.
[Philosophical Magazine, XLIX. pp. 539, 540, 1900.]
BY complete radiation I mean the radiation from an ideally black body,
which according to Stewart* and Kirchhoff is a definite function of the
absolute temperature 0 and the wave-length \. Arguments of (in my opinion^)
considerable weight have been brought forward by Boltzmann and W. Wien
leading to the conclusion that the function is of the form
0*<j>(0\)d\, ................................. (1)
expressive of the energy in that part of the spectrum which lies between
X and A, + d\. A further specialization by determining the form of the
function </> was attempted later J. Wien concludes that the actual law is
(2)
in which Cj and c2 are constants, but viewed from the theoretical side the
result appears to me to be little more than a conjecture. It is, however,
supported upon general thermodynamic grounds by Planck §.
Upon the experimental side, Wien's law (2) has met with important
confirmation. Paschen finds that his observations are well represented,
if he takes
c2 = 14,455,
* Stewart's work appears to be insufficiently recognized upon the Continent. [See Phil. Mag.
i. p. 98, 1901 ; p. 494 below.]
t Phil. Mag. Vol. XLV. p. 522 (1898).
$ Wied. Ann. Vol. LVIII. p. 662 (1896).
§ Wied. Ann. Vol. i. p. 74 (1900).
31—2
484 REMARKS UPON THE LAW OF COMPLETE RADIATION. [260
6 being measured in centigrade degrees and \ in thousandths of a millimetre
(//,). Nevertheless, the law seems rather difficult of acceptance, especially
the implication that as the temperature is raised, the radiation of given wave-
length approaches a limit. It is true that for visible rays the limit is out
of range. But if we take \ = 60 /JL, as (according to the remarkable researches
of Rubens) for the rays selected by reflexion at surfaces of Sylvin, we see that
for temperatures over 1000° (absolute) there would be but little further
increase of radiation.
The question is one to be settled by experiment ; but in the meantime
I venture to suggest a modification of (2), which appears to me more probable
d priori. Speculation upon this subject is hampered by the difficulties
which attend the Boltzmann-Maxwell doctrine of the partition of energy.
According to this doctrine every mode of vibration should be alike favoured ;
and although for some reason not yet explained the doctrine fails in general,
it seems possible that it may apply to the graver modes. Let us consider
in illustration the case of a stretched string vibrating transversely. According
to the Boltzmann-Maxwell law the energy should be equally divided among
all the modes, whose frequencies are as 1, 2, 3, .... Hence if k be the
reciprocal of X, representing the frequency, the energy between the limits
k and k + dk is (when k is large enough) represented by dk simply.
When we pass from one dimension to three dimensions, and consider
for example the vibrations of a cubical mass of air, we have (Theory of Sound,
§ 267) as the equation for k2,
where p, q, r are integers representing the number of subdivisions in the
three directions. If we regard p, q, r as the coordinates of points forming
a cubic array, k is the distance of any point from the origin. Accordingly
the number of points for which k lies between k and k + dk, proportional
to the volume of the corresponding spherical shell, may be represented
by k*dk, and this expresses the distribution of energy according to the
Boltzmann-Maxwell law, so far as regards the wave-length or frequency.
If we apply this result to radiation, we shall have, since the energy in each
mode is proportional to 0,
0k*dk, .................................... (3)
or, if we prefer it,
e\~*d\ .................................. (4)
It may be regarded as some confirmation of the suitability of (4) that it is of
the prescribed form (1).
The suggestion is that (4) rather than, as according to (2),
\-*d\ .................................... (5)
1900] REMARKS UPON THE LAW OF COMPLETE RADIATION. 485
may be the proper form when X0 is great*. If we introduce the exponential
factor, the complete expression will be
c10\-4e-c«/A»dX (6)
If, as is probably to be preferred, we make k the independent variable,
(6) becomes
Cl0k2e-c*edk (7)
Whether (6) represents the facts of observation as well as (2) I am not in
a position to say. It is to be hoped that the question may soon receive
an answer at the hands of the distinguished experimenters who have been
occupied with this subject.
* [1902. This is what I intended to emphasize. Very shortly afterwards the anticipation
above expressed was confirmed by the important researches of Rubens and Kurlbaum (Drude
Ann. iv. p. 649, 1901), who operated with exceptionally long waves. The formula of Planck,
given about the same time, seems best to meet the observations. According to this modification
of Wien's formula, e~c^9 in (2) is replaced by l-^(eC2/Xfl-l). When X0 is great, this becomes
X0/c2, and the complete expression reduces to (4).]
261.
ON APPROXIMATELY SIMPLE WAVES.
[Philosophical Magazine, L. pp. 135—139, 1900.]
THE _first_question that arises is as to the character of absolutely simple
waves; and here "it may be well to emphasize that a simple vibration implies
infinite continuance, and does not admit of variations of phase or amplitude,
To suppose, as is sometimes done in optical speculations, that a train of simple
waves may begin at a given epoch, continue for a certain time involving it
may be a large number of periods, and ultimately cease, is a contradiction
in terms*." A like contradiction is involved if we speak of unpolarized light
as homogeneous, really homogeneous light being necessarily polarized.
This much being understood, approximately simple waves might be
defined as waves which for a considerable succession deviate but little from
a simple train. Under this definition large changes of amplitude and fre-
quency would not be excluded, provided only that they entered slowly enough.
More frequently further limitation would be imposed, and approximately
simple waves would be understood to mean waves which for a considerable
succession can be approximately identified with a simple train of given
frequency, if not of given amplitude. But the phase-f of the simple train
approximately representing the given waves would vary from place to place,
slowly indeed but to any extent.
Thus if we take, as analytically expressing the dependence of the displace-
ment upon time,
^, ........................... (1)
* Theory of Sound, 2nd ed. § 65 a, 1894.
+ What is here called for brevity the phase is more properly the deviation of phase from that
of an absolutely simple train of waves.
1900] ON APPROXIMATELY SIMPLE WAVES. 487
where H and K are slowly varying functions of t, the frequency may be
regarded as constant, while the amplitude \/(H2 + K2) and the phase
tan"1 (K/H) vary slowly but without limit. It scarcely needs to be pointed
out that a slow uniform progression of phase is equivalent to a small change
of frequency.
( / In one important class of cases the phase remains constant and then,
since a constant addition to t need not be regarded, (1) is sufficiently repre-
sented by
H cospt (2)
simply. If the changes of amplitude are periodic, we may write
H = H0 + Hl cos qt + H,' sin qt + H2 cos 2qt + H2' sin 2qt + . . . , . . .(3)
in which q is supposed to be small. The vibration (2) is then always
equivalent to a combination of simple vibrations of frequencies represented by
p, p + q, p — q, p + 2<£, p — *2q, &c.
Under this head may be mentioned the case of ordinary beats, so familiar
in Acoustics. Here
H=H1cosqt, .............................. (4)
and Hcospt=%Hlcos(p + q)t + ^Hlcos(p-q)t ............. (5)
It may be observed that although the phase is regarded as constant, the
change of sign in the amplitude has the same effect as an alteration qf_ _
phase of 180°.
Another important example is that of intermittent vibrations. If we put
H='2(l + cosqt\ .............................. (6)
the amplitude is always of one sign, and
H cos pt = 2 cos pt + cos (p + q) t + cos (p — q) t .......... (7)
Three simple vibrations are here required to represent the effect.
Again (Theory of Sound, § 65 a), if
(8)
we have
H cospt = | cospt + cos (p + q) t + cos (p — q) t
+ \cos(p + 2q)t + $cos(p-'2q)t ....... (9)
If K also be variable and periodic in the same period as H, so that .
K = Ko + Kl cos qt + KI sin qt + K* cos 2qt + K2' sin 2qt + . . . , . . .(10)
we have the most general periodicity expressed when we substitute these
488 ON APPROXIMATELY SIMPLE WAVES. [261
values in (1); and the general conclusion as to the periods of the simple
vibrations required to represent the effect remains undisturbed.
If K and H vary together in such a manner that the amplitude \/(H* + K2)
remains constant, the sole variation is one of phase. My object at present
is to call attention to this class of cases, so far as I know hitherto neglected,
unless an example (Phil. Mag. xxxiv. p. 409, 1892 [Vol. iv. p. 16]) in
which an otherwise constant amplitude is periodically and suddenly reversed
be considered an exception.
If we take
H = cos (a sin qt), K = sin (a sin qt), ............... (11)
H and K are of the required periodicity, and the condition of a constant
amplitude is satisfied. In fact (1) becomes
cos (pt — a. sin qt) ........................ ____ (12)
Now, since
eiacose = JQ (a) + 2^ («) cos 6 + 2»V, (a) cos 2^ + ...
+ 2iV«(a)cosn0+...,
we have
eiasmqt = J0 + 2tV1 sin qt + 2JZ cos 2qt + 2iJs sin 3qt
+ 2/4 cos 4^+..., ............ (13)
and thus
cos (a sin qt) = J0 (a) + 2/2 (a) cos 2qt + 2/4 (a) cos 4>qt + . . . , . . .(14)
sin (a sin qt) = 2Jt (a) sin qt + 2/3 (a) sin 3gtf + ... , ............ (15)
where J0, J1} &c. denote (as usual) the Bessel's functions of the various
orders. In the notation of (3). and
H, = #3 = ... =,0, #/ = #/ = #,'=...=<),
H0 = J0(a\ H2 = 2J2(a), fT4=2J4(a), &c.,
K0 = K, = K2= ... = 0, Kl = K^ = ... = 0,
JST1' = 2J1(a), K3'=2J3(«), K5' = 2J5(a), &c.
Accordingly (12), expressed as a combination of simple waves, is
«/0 (a) cos pt + Jz (a) {cos (p- 2g) t + cos (p + 2q) t]
+ /j (a) {cos (p-q)t- cos (p + q) t}
+ J3(a){cos(p-3q)t-cos(p+3q)t} + ............. (16)
1900]
ON APPROXIMATELY SIMPLE WAVES.
489
n
<M3)
/.(«)
Jn (12)
Jn (18)
Jn (24)
0
1
2
3
4
5
6
7
8
9
10
11
12
- -26005
+ -33906
+ -48609
+ -30906
+ -13203
+ -04303
+ •01 139
+ -00255
+ -00049
+ -00008
+ -00001
+ -15065
- -27668
- -24287
+ •11477
+ •35764
+ -36209
+ -24584
+ -12959
+ -05653
+ -02117
+ -00696
+ -00205
+ '00055
+ -04769
- -22345
- -08493
+ •19514
+ -18250
- -07347
- -24372
-•17025
+ -04510
+ -23038
+ -30048
+ -27041
+ -19528
- -01336
- -18799
- -00753
+ -18632
+ -06964
- -15537
- -15596
+ -05140
+ -19593
+ -12276
-•07317
- -20406
•17624
- -05623
- -15404
+ -04339
+ •16127
- -00308
- -16230
- -06455
+ -13002
+ -14039
- -03643
- '16771
- -10333
+ -07299
13
14
+ -00013
+ -00003
+ •12015
+ -06504
- -03092
+ •13157
+ •17632
+ •11803
15
+ -00001
+ -03161
+ '23559
- -03863
16
+ •01399
+ -26108
- -16631
17
+ '00570
+ •22855 '
- -18312
18
+ -00215
+ •17063
•09311
19
+ -00076
+ •11271
+ -04345
20
+ -00025
+ '06731
+ •16191
21
+ -00008
+ -03686
+ '22640
22
+ •00002
+ -01871
+ -23429
23
24
+ -00001
+ -00886
+ '00395
+ •2031 3
+ -15504
25
+ -00166
+ -10695
26
+ -00066
+ -06778
27
+ -00025
+ -03990
28
+ •00009
+ -02200
29
+ -00003
+ •01143
30
+ -00001
+ -00563
31
+ -00263
32
+ •001 18
33
+ •00050
34
+ -00021
35
+ -00008
36
+ -00003
37
+ -00001
490 ON APPROXIMATELY SIMPLE WAVES. [261
If a, representing the maximum disturbance of phase, be small, we may
write approximately
while 1/3 &c. are of higher powers in a than a2. Thus if we stop at the first
power of a, we are concerned only with the multiples of t represented by
p, p-q, p + q;
while if we include a2 we have
p, p-q, p + q, p-2q, p + 2q.
But when a is not small, the convergence is slow, and a large number
of terms will be required even for a moderately close approximation. The
preceding table, due to Meissel, is condensed from Gray and Mathew's Bessel's
Functions. So far as TT can be identified with 3, the values of a equal to
3, 6, 12, 18, 24 correspond to maximum deviations of phase (in both directions)
equal to \, 1, 2, 3, 4 periods respectively. It appears that, the largest value
of Jn (a) occurs for a value of n somewhat less than a. Indeed, it is at once
evident from (12) that frequencies in the neighbourhood of p ± qa. will be
important elements.
262.
ON A THEOREM ANALOGOUS TO THE VIRIAL THEOREM.
[Philosophical Magazine, L. pp. 210—213, 1900.]
As an example of the generality of the theorem of Clausius, Maxwell*
mentions that " in any framed structure consisting of struts and ties, the sum
of the products of the pressure in each strut into its length, exceeds the
sum of the products of the tension of each tie into its length, by the product
of the weight of the whole structure into the height of its centre of gravity
above the foundations." It will be convenient to sketch first the proof of
the purely statical theorem of which the above is an example, and afterwards
of the corresponding statical applications of the analogue. The proof of the
general dynamical theorem will then easily follow.
If X, Y, Z denote the components, parallel to the axes, of the various
forces which act upon a particle at the point x, y, z, then since the system
is in equilibrium,
If we multiply these equations by x, y, z respectively, and afterwards effect
a summation over all the particles of the system, we obtain a result which
may be written
2£] = 0 ...................... (1)
The utility of the equation depends upon an alteration in the manner
of summation, and in particular upon a separation of the forces R (considered
positive when repellent) which act mutually between two particles along their
* Nature, Vol. x. p. 477, 1874 ; [Maxwell's] Scientific Papers, Vol. n. p. 410.
492 ON A THEOREM ANALOGOUS TO THE VIRIAL THEOREM. [262
line of junction p. If x, y, z and x ', y', z' be the coordinates of the particles,
we have so far as regards the above-mentioned forces,
or with summation over every pair of particles HRp. The complete equation
may now be written
0, ..................... (2)
where in the first summation X, Y, Z represent the components of the
external forces operative at the point x, y, z. In Maxwell's example the only
external forces are the weights of the various parts of the system (supposed
to be concentrated at the junctions of the struts and ties), and the reactions
at the foundations.
The analogous theorem, to which attention is now called, is derived in
a similar manner from the equally evident equation
(3)
We have to extract from the summation on the left the force R mutually
operative between the particles at x, y, z and at x, y', z' ; and we shall limit
ourselves to the case of two dimensions. If X, Y be the components of force
acting upon the latter particle, p the distance between the particles, and
<p the inclination of p to the axis of x, we have
so that if now X, Y represent the total external force acting at x, y,
(3) becomes
2£ = 0, ..................... (4)
where the first summation extends to every particle and the second to every
pair of particles.
If the external force at x, y be P and be inclined at an angle a, we have
X = Pcosa, F=Psina;
so that, if x = r cos 8, y = r sin 6 as usual, (4) may be written
0 ................... (5)
As simple examples of these equations, consider the square framework
with one diagonal represented in Figs. 1 and 2, and take the coordinate axes
parallel to the sides of the square. Since sin 20 = 0 for all four sides of
the square, the only R that occurs is that which acts along the diagonal
where sin 20 = - 1. In Fig. 1 opposed forces P act at the middle points of
the sides, but since in each case 6 + a = 0, the terms containing P disappear.
Hence R = 0.
1900]
ON A THEOREM ANALOGOUS TO THE VIRIAL THEOREM.
493
In Fig. 2, where external forces P act diagonally at the unconnected
corners, sin (6 + a) = — 1, and since p = 2r, R = — P, signifying that the
diagonal piece acts as a tie under tension P. In neither case would the
weight of the members disturb the conclusion.
Fig. 1. Fig. 2.
The forces exercised by the containing vessel upon a liquid confined
under hydrostatic pressure p contribute nothing to the left-hand member
of (4). The normal force acting inwards upon the element of boundary
ds is pds, so that
X = —pdy, Y = pdx,
and accordingly
vanishing when the integration extends over the whole boundary.
Abandoning now the supposition that the particle at x, y is at rest,
we have
d2 (xy) _ dx dy d2y d2x
dt* ~ dt dt X~dt2 y dt2 '
so that if m be the mass of the particle, X, Y the components of force acting
upon it,
o™ dx dy -
or with summation over all the particles of the system,
•(7)
We now take the mean values with respect to time of the various terms
in (7). If the system be such that
does not continually increase, we obtain, as in the case of the virial theorem,
It would seem that this equation has application to the molecular theory
of the viscosity of gases, analogous to that of the virial as applied to hydro-
static pressure.
263.
ON BALFOUR STEWART'S THEORY OF THE CONNEXION
BETWEEN RADIATION AND ABSORPTION.
[Philosophical Magazine, I. pp. 98—100, 1901.]
ON a recent occasion* I remarked that Stewart's work appeared to me
to be insufficiently recognized upon the Continent. One reason for this is
probably the comparative inaccessibility of the Edinburgh Transactions in
which his first paper appeared f. Another may be found in the fact that the
paper itself is not well arranged, and that the principal conclusion is put
forward in the first instance as if it were the result of Stewart's special
experiments. The experiments were indeed of great value ; but this course
gave an opening to Kirchhoff's objection that " this proof [of the law that the
absorption of a plate equals its radiation and that for every description of
heatl] cannot be a strict one, because experiments which have only taught
us concerning more and less, cannot strictly teach us concerning equality^."
I am inclined to think that Stewart would have received more recognition
if he had never experimented at all !
While yielding to no one in admiration for Kirchhoff, I can hardly regard
him as in this matter an impartial critic. In a paper|| which should be
studied by the historical inquirer, Stewart himself protests against some of
Kirchhoff's remarks, and to my judgment makes out his case. In his ex-
cellent Handbuch der tipectroscopie, recently published, Prof. Kayser, with
evident desire to be impartial, gives Stewart much, but not all, of the credit
that I would claim for him. But, so far as I have seen, neither Stewart
himself nor any of his critics favourable or unfavourable have cited the
paragraph upon which he mainly relies. It may be of service to readers who
are unlikely to see the original, if I reproduce it here, exactly as it stood : —
* Phil. Mag. S. 5, Vol. XLIX. p. 539 (1900). [Vol. iv. p. 483.]
t Edin. Trans. Vol. xxn. p. 1, March 1858.
J The italics are Stewart's.
§ Kirchhoff, " On the History of Spectrum Analysis," &c., Phil. Mag. Vol. xxv. p. 258 (1863).
|| Phil. Mag. Vol. xxv. p. 354 (1863).
1901] ON STEWART'S THEORY OF RADIATION AND ABSORPTION. 495
'20. A more rigid demonstration may be given thus: — Let AB, BC be
two contiguous, equal, and similar plates in the interior of a substance of
indefinite extent, kept at a uniform temperature.
The accumulated radiation from the interior im-
pinges on the upper surface of the upper plate ;
let us take that portion of it which falls upon
the particles A, in the direction DA. This ray,
in passing from A to B will have been partly
absorbed by the substance between A and B\
but the radiation of the upper plate being equal
to its absorption (since its temperature remains
the same), the ray will have been just as much recruited by the united
radiation of the particles between A and 5, as it was diminished in intensity
by their absorption. It will therefore reach B with the same intensity as it
had at A. But the quality of the ray at B will also be the same as its
quality at A. For, if it were different, then either a greater or less pro-
portion would be absorbed in its passage from B to C, than was absorbed
of the equally intense ray at A, in its passage between A and B. The
amount of heat absorbed by the particles between B and C would therefore
be different from that absorbed by the particles between A and B. But this
cannot be ; for, on the hypothesis of an equal and independent radiation of
each particle, the radiation of the particles between B and C is equal to that
of the particles between A and B, and their absorption equals their radiation.
Hence the radiation impinging on B, in the direction of DB, must be equal
in quality as well as quantity to that impinging upon A ; and, consequently,
the radiation of the particles between A and B must be equal to their ab-
sorption, as regards quality as well as quantity ; that is, this equality between
the radiation and absorption must hold for every individual description of
heat*."
Surely this goes to the root of the matter, and it presents the argument
in its most natural form. Kirchhoff 's independent investigation of a year
and a half later f is more formal and elaborate, but scarcely more convincing.
No one in England or elsewhere disputes the great obligations under
which Spectrum Analysis lies to Kirchhoff. In a passage quoted by Dr
Kayser (loc. cit. p. 92) from Lord Kelvin — " To Kirchhoff belongs, I believe,
solely the great credit of having first actually sought for and found other
metals than sodium in the sun " — the force of " solely" seems to have been
misunderstood. I have Lord Kelvin's authority for interpreting this to mean
that the entire credit of the discovery mentioned belongs to Kirchhoff, not
that he is entitled to no credit in other directions.
* Edin. Trans. Vol. xxn. p. 13, March 1858.
+ Monatsbericht der Akad. d. Wiss. zu Berlin, Dec. 1859.
264.
SPECTROSCOPIC NOTES CONCERNING THE GASES OF THE
ATMOSPHERE.
[Philosophical Magazine, I. pp. 100—105, 1901.]
On the Visibility of Hydrogen in Air.
MY first experiments upon this question were made in July 1897. The
sparks were taken between platinum points in a small chamber through
which dried air at atmospheric pressure could be led, and the spectrum was
examined with a spectroscope of two prisms. The (7-line could be very
nearly obliterated by careful drying. But if jfa part by volume of hydrogen
were added and the mixture passed afterwards through the phosphoric
anhydride, the increased visibility of C was very marked. At that time
I was occupied with the density of carbonic oxide and interested in the
question as to whether it contained appreciable quantities of hydrogen*.
When carbonic oxide, prepared from prussiate of potash and dried as for
weighing, was passed through the apparatus, the (7-line became nearly in-
visible ; but the test with carbonic oxide was thought to be less delicate than
with air in consequence of the proximity of another bright line in the former
I have lately resumed these experiments, induced thereto principally by
the remarkable results of M. Gautier. This observer, working by chemical
methods, finds that air normally contains about 10to0o of hydrogen in addition
to variable amounts of hydrocarbons. It appeared to me that a spectro-
scopic confirmation would be interesting.
* Proc. Boy. Soc. Vol. LXII. p. 205 (1897). [Vol. iv. p. 348.]
1901] SPECTROSCOPIC NOTES CONCERNING GASES OF THE ATMOSPHERE. 497
Using the old apparatus, in which the tubes conveying the gas and the
electrodes were fitted into a rubber cork, I could not succeed in getting quit
of C from the spectrum of somewhat powerful sparks, however carefully the
air were dried. The coil was excited with five Grove cells and a large leyden-
jar*was connected with the secondary in the usual way. This observation
was of course consistent with the presence of hydrogen in the atmosphere ;
but it was suspicious that the best approach to evanescence was obtained
with a somewhat brisk rather than with a slow current of air, indicating
that the source of the hydrogen was in the apparatus rather than in the
atmosphere. As it seemed desirable to apply heat, I discarded the old
apparatus, substituting for it a simpler one consisting merely of a small
bulbous enlargement of the gas-leading tube. Into this the platinum elec-
trodes were sealed. The gases under examination were stored under slight
pressure and on leaving the reservoirs were partially dried with sulphuric
acid. A three-way tap allowed the easy substitution of one gas for another.
After passing this tap the gas was further dried by phosphoric anhydride on
its way to the sparking-bulb.
The application of heat to the bulb and to the short length of tubing
between the bulb and the phosphoric anhydride led, as was expected, to
a recrudescence of 0. Subsequently there seemed to be an improvement.
Not only was C less conspicuous, but its visibility remained about the same
although the rate of flow were varied. It is difficult to describe in words
the effect upon the eye, but I may say that with the actual spectroscopic
arrangements including a somewhat wide slit the line could be certainly and
steadily seen.
The above was the appearance with a stream of (country) air. When air
to which jr-^ part of hydrogen (by volume) had been added was substituted,
the visibility of C was markedly increased ; and the difference was such that
one could easily believe that the proportion of hydrogen actually operative
had been doubled. This conclusion would be in precise agreement with
M. Gautier, could we assume that the smaller quantity of hydrogen really
accompanied the air. But the facts now to be recorded render this assump-
tion extremely doubtful.
In the first place the visibility of G with ordinary air was not perceptibly
diminished by passage of the air over red-hot cupric oxide included between
the sulphuric acid and the three-way tap. It may be argued that cupric
oxide is not competent in moderate length to remove the last traces of
hydrogen from air, even though the air be passed over it in a slow stream.
I found, however, on a former occasion* that hydrogen purposely introduced
* On an Anomaly encountered in Determinations of the Density of Nitrogen Gas, Proc. Roy.
Soc. Vol. LV. p. 343 (1894). [Vol. iv. p. 107.]
E. iv. 32
498 SPECTKOSCOPIC NOTES CONCERNING [264
into nitrogen could be so far removed in this way that the weight remained
sensibly unaffected, although 10*000 of residual hydrogen might be expected
to manifest itself.
Moreover, when air purposely contaminated as above with ^^5- of hy-
drogen was passed over the copper oxide, the additional hydrogen appeared
to be removed, the visibility of C reducing itself to that corresponding to
untreated air.
Being desirous of testing the matter as far as possible, I have ex-
perimented also with nitrous oxide and with oxygen. In the former case
the general appearance of the spectrum is much the same as with air. The
(7-line was thought to be, if anything, more visible than in the case of air,
but the difference could not be depended upon. Two samples of gas were
tried, one from an iron bottle as supplied commercially, the other prepared
in the laboratory from ammonium nitrate. Oxygen from permanganate of
potash also showed the (7-line more distinctly than air, but this may probably
be attributed to the elimination of a neighbouring nitrogen line. It is
possible, of course, that these gases may have contained traces of hydrogen,
but in that case it is strange that the proportion should be so nearly the
same as in air.
These observations certainly seem to leave a minimum of room for the
hydrogen found by M. Gautier, but I should be unwilling to call his con-
clusion in question on the strength of what are after all but eye estimates.
I have not been able to find a detailed account of M. Gautier's experiments
or of what precautions he took to assure himself that the water collected
could not have had its origin in the glass or copper oxide of his hot tubes*.
The most satisfactory test would be comparison experiments in which oxygen
or nitrous oxide is substituted for air, or, perhaps better still, in which air is
used over and over again.
If, as I should suppose were I to judge from my own experiments alone,
the residual (7-line was not wholly or even principally due to hydrogen in the
air, it would have to be explained by hydrogen evolved from the glass of the
sparking-chamber or from the platinum electrodes. In view of what is known
respecting the behaviour of vacuum-tubes, such an explanation does not
appear improbable.
Experiments upon the visibility of C in vacuum-tubes have shown a much
smaller degree of sensibility. The tube was in connexion with a Topler
pump and was traversed by a stream of air. The passage from high (atmo-
spheric) to low pressure took place at a glass capillary which allowed about
30 c.c. per hour (reckoned at atmospheric pressure) to leak past. When
* [1902. See Annales de Chimie, xxii. Jan. 1901. Further experiments of my own are detailed
in Phil. Mag. in. p. 416, 1902.]
1901]
THE GASES OF THE ATMOSPHERE.
moist air from the room on a damp day (15° C.) was admitted, the hydrogen
(7-line was very bright, nearly obliterating one of the dark bands of nitrogen.
On drying the air with phosphoric anhydride, the (7-line disappeared. Air
mixed with 1 per cent, of hydrogen showed it doubtfully, 1£ per cent, plainly,
2 per cent, perhaps equally with the moist air. The much smaller sensibility
(about 50 times) in these experiments may be partly due to the less favour-
able character of the ground upon which the hydrogen line has to show itself.
Demonstration at Atmospheric Pressure of Argon from very small
quantities of Air.
Success in reducing the necessary amount of air depends a good deal
upon the form of tube employed. That sketched (Fig. 1) allows a minimum
residue to be sparked and examined. In some ex-
periments the tube, standing over weak alkali, was Fi8- *•
charged with 5 c.c. only of air. The first part of
the sparking is with electrodes ending in platinum
points and brought up in U-shaped tubes of which
the bends are filled with mercury. A Ruhmkorff
actuated by two or three Grove cells is employed
at this stage, and the sparks pass just under the
shoulder of the containing tube, oxygen being sup-
plied as required from a small electrolytic gene- / \ Scale f •
rator. When the volume is sufficiently reduced
and most of the nitrogen has disappeared, the
electrodes above spoken of are removed.
In the next stage the sparks are taken between
a sealed-in electrode at the top of the containing
tube, and another sealed into the top of a single
U-tube brought round through the alkali, and
rising (as shown) through the narrow part of the
containing tube. In order to avoid splashing and
consequent risk of fracture from sudden cooling of
the heated glass, it was thought an advantage that
the tubes through part of their length should make a tolerably close fit.
But the most important precaution appears to be the use of very short
sparks and a reduction of the battery to two cells. When it is desired to
observe the spectrum, a small jar must be connected in the usual way.
The spectroscope employed had two prisms, and the sparks were focused
upon a somewhat wide slit by a 2-inch lens. A low-power eyepiece was
favourable.
32—2
500 SPECTROSCOPIC NOTES CONCERNING [264
The group of lines in the argon spectrum first observed by Schuster*
[for figure, see Vol. iv. p. 199] is easily seen. Owing to the warmth F is
very diffuse, sometimes nearly filling up the interval between 4879 and 4847.
On one occasion when the original air taken was only 5 c.c., the group was
almost as distinct as with pure argon. The residual gas, measured cold, was
probably no more than 1 c.c. This was rather an extreme case, and it would
not have been possible to renew the sparking without an addition of oxygen,
to be afterwards removed by careful additions of hydrogen. But the argon
spectrum shows fairly well even when the gas is diluted with two or three
times its volume of oxygen.
I have described this experiment at some length because I think that it
would make a good exercise for students, requiring no special apparatus but
what they should be able to construct for themselves. Although, as I have
said, 5 c.c. of air is ample, a novice would naturally begin with 10 or 15 c.c.
Concentration of Helium from the Atmosphere.
In a footnote [p. 266] to a paper on the Separation of Gases by Diffusionf.
I suggested that the lighter constituent of a mixture might be concentrated
by causing it to diffuse against a stream of an easily absorbable gas, such as
carbonic acid. In Jan. 1899 a good many trials upon these lines were made
with the object of putting in evidence the helium of the atmosphere, and
a certain degree of success was attained. A stream of carbonic acid (pre-
pared from marble and hydrochloric acid and reckoned at 3 litres per hour)
is maintained for say 14 hours through a diffusion-tube open above to the
atmosphere. This tube, placed vertically, is about 40 cm. long and of about
5 cm. diameter. The gases of the atmosphere diffuse downwards into the
tube, but the heavier constituents are held almost entirely at bay by the
stream of carbonic acid. If we draw off continuously a supply from a point
say halfway down the diffusion-tube, we shall obtain carbonic acid with a
small admixture of atmospheric gases in which the lighter ingredients, e.g.
water, hydrogen, and helium, are relatively much concentrated. In my ex-
periments the lateral stream was about 250 c.c. per hour, and was manipulated
with the aid of a Sprengel pump. Between the pump and the diffusion-tube
was interposed a short length of tobacco-pipe through the walls of which the
gas had to pass and which presented the right degree of obstruction. After
passage to the low-pressure side, the bulk of the C02 was absorbed with
alkali, and the residual gases collected over alkali at the foot of the Sprengel
in the usual way.
* Rayleigh and Ramsay, Phil. Trans. 186, p. 224 (1895). [Vol. iv. p. 169.] See also Nature,
Vol. m. p. 163 (1895). [Vol. iv. p. 199.]
t Phil. Mag. Vol. XLII. p. 493 (1896). [Vol. iv. p. 266.]
1901] THE GASES OF THE ATMOSPHERE. 501
The subsequent treatment for removal of nitrogen by the electric dis-
charge was conducted as usual, towards the close in the tube described and
figured above. The final residue on the occasion when D3 was best seen
(under the jar discharge) was about '25 c.c. Argon was also plainly visible
and probably constituted the greater part of the bulk. When the volume
was doubled by addition of oxygen, Dg was seen less well.
Success depended a good deal upon precautions to avoid the presence of
gases, and especially of argon, which had not undergone diffusion. It was
necessary to eliminate the dissolved gases of the dilute hydrochloric acid
with which the C0.2 was prepared, and to keep an atmosphere of C02 in the
supply- vessel. Until these precautions were taken, D3, though frequently
suspected, was not clearly and steadily seen. Even at the best, good measure-
ments could hardly have been taken ; but the line appeared to be in the right
place for helium, as distinguished for example from neon.
265.
ON THE STRESSES IN SOLID BODIES DUE TO UNEQUAL
HEATING, AND ON THE DOUBLE REFRACTION RESULT-
ING THEREFROM*
[Philosophical Magazine, I. pp. 169—178, 1901.]
THE phenomena of light and colour exhibited in the polariscope when
strained glass is interposed between crossed nicols are well known to every
student of optics. The strain may be of a permanent character, as in glass
imperfectly annealed or specially unannealed, or it may be temporary, due
to variations of temperature or to mechanical force applied from without.
One of the best examples under the last head is that of a rectangular bar
subjected to flexure, the plane of the flexure being perpendicular to the
course of the light. The full effect is obtained when the length of the
bar is at 45° to the direction of polarization. The revival of light is a maximum
at the edges, where the material traversed is most stretched or compressed,
while down the middle a dark bar is seen representing the " neutral axis."
It is especially to be noted that the effect is due to the glass being unequally
stretched in the two directions perpendicular to the line of vision. Thus
in the case under discussion no force is operative perpendicular to the length
of the bar. Under a purely hydrostatic pressure the singly refracting
character of the material would not be disturbed.
When a piece of glass, previously in a state of ease, is unequally heated,
double refraction usually ensues. This is due, not directly to the heat,
but to the stresses, different in different directions and at different places,
caused by the unequal expansions of the various parts. The investigation
of these stresses is a problem in Elasticity first attacked, I believe, by
* From the Lorentz Collection of Memoirs.
1901] ON STRESSES DUE TO UNEQUAL HEATING. 503
J. Hopkinson*. It will be convenient to repeat in a somewhat different
notation his formulation of the general theory, and afterwards to apply it
to some special problems to which the optical method of examination is
applicable.
In the usual notation f if P, Q, R, S, T, U be the components of stress ;
u, v, w the displacements at the point x, y, z ; \, /* the elastic constants ;
we have such equations as
'du dv dw\ . du
¥dy+d,) + ^di' (1)
•dw dv^
These hold when the material is at the standard temperature. If we suppose
that the temperature is raised by 6 and that no stresses are applied,
du dv dw n
-T~ = j- = rr- = *0>
dx dy dz
while dw/dy &c. vanish. The stresses that would be needed to produce the
same displacements without change of temperature are
Hence, so far as the principle of superposition holds good, we may write
in general
.......... <3>
with similar equations for Q, R, T, U.
If there be no bodily forces, the equation of equilibrium is
^ + f + f=0, .............................. (5)
dx dy dz
with two similar equations ; or with use of (3) and (4)
= <> ............. (6)
if
(7)
One of the simplest cases that can be considered is that of a plate, bounded
by infinite planes parallel to xy, and so heated that 6 is a function of z only.
* Mess, of Math. Vol. vm. p. 168 (1879). [1902. From a notice by W. Konig (Beiblatter, 1901)
I gather that some of the problems here dealt with had already been treated by Neumann in 1841.]
t See, for example, Love's Theory of Elasticity, Cambridge University Press, 1892.
504 ON THE STRESSES IN SOLID BODIES DUE TO UNEQUAL HEATING, [265
If, further, 6 be symmetrical with respect to the middle surface, the plate
will remain unbent; and if the mean value of 0 be zero, the various plane
sections will remain unextended. Assuming, therefore, that u, v vanish while
w is variable, we get from (3) and (4)
(8)
P=Q = X-70, ................................. (9)
S = T= U=0 ..................................... (10)
In (8) R is assumed to vanish, since no force is supposed to act upon the
faces. From (8), (9)
If the plate be examined in the polariscope by light traversing it in
the direction of y. the double refraction, depending upon the difference
between R and P, of which the former is zero, is represented simply by (11).
Dark bars will be seen at places where 6 = 0. If the direction of the light
be across the plate, i.e. parallel to z, there is no tendency to double refraction,
since everywhere P = Q.
In the above example where every layer parallel to xy remains unextended,
the local alteration of temperature produces its full effect. But in general
the circumstances are such that the plate is able to relieve itself to a
considerable extent. A uniform elevation of temperature, for instance, would
entail no stress. And again, a uniform temperature gradient, such as would
finally establish itself if the two surfaces of the plate were kept at fixed
temperatures, is compensated by bending and entails no stress. In such
cases before calculating the stress by (11) we must throw out the mean value
of 6 so as to make fPdz = 0, and also such a term proportional to the distance
from the middle surface as shall ensure that fPzdz=0. Otherwise the
edges of the plate could not be regarded as free from imposed stress in
the form of a force or couple.
The assumption in (1), (2) that u = v = Q is now replaced by
u = (a + /3z)a;, v = (a + $z)y, .................. (12)
and w = w' - \$ (of + t/2), ........................ (12')
where w' is a function of z only. We find
(13)
U=0 ........................................ (15)
1901] AND ON THE DOUBLE REFRACTION RESULTING THEREFROM. 505
Since R is supposed to vanish, we get
In (16) a and /3 are to be determined by the conditions
or, which comes to the same, we are to reject from 6 such linear terms as
will leave
Q ......................... (17)
Since w' and 6 are independent of x and y, the equations of equilibrium (5)
are satisfied.
It is of interest to trace the influence of time upon the double refraction
of the heated plate when light passes through it edgeways, e.g. parallel to y.
Initially 6 may be supposed to be an arbitrary function of z, while the faces
of the plate, say at 0 and c, are maintained at given temperatures. Ultimately
the distribution of temperature is expressed by a linear function of z, say
H' + Kz ; and, as is known from Fourier's theory, the distribution at time
t may be expressed by
0 = H'+Kz + ^Ane-P''f sin (nirz/c), ............... (18)
where n is an integer and pn, depending also upon the conductivity, is
proportional to n2. After a moderate interval, the terms corresponding
to the higher values of n become unimportant.
In the subsequent calculation it is convenient to take the origin of z
in the middle surface, instead of as in (18) at one of the faces. Thus
0 = H + Kz + A.e-P^ cos — - Aze~^ cos — + ...
-A^sin— + A<tr**aa— - ............. (19)
c c
If 0' represent the value of 0 when reduced by the subtraction of the
proper linear terms as already explained, we find
*
V = Alf" (cos ™ - *) - A,r» (cos ?= + £
\ c TT) \ C BTT
...... (20)
After a moderate time the term in A l usually acquires the preponderance,
and then 0' = 0 when cos (trz/c) = 2/w. When the plate is looked at edge-
ways in the polariscope, dark bars are seen where z = ± '280 c, c being the
whole thickness of the plate.
506 ON THE STRESSES IN SOLID BODIES DUE TO UNEQUAL HEATING, [265
As a particular case of (19), (20) let us suppose that the distribution
of temperature is symmetrical, or that K vanishes as well as the coefficients
of even suffix A^, At, &c. H then represents the temperature at which the
two faces are maintained, and (19) reduces to
-
(21)
c c
If we suppose further that the initial temperature is uniform and equal
to ®, we find by Fourier's methods
A = ^(®-#), AZ = ^(®-H}, 4,= l(e-J50, &c- -(22)
and
2
.., ......... (23)
where also
p3 = 9Pl, p5 = 25Pl, &c ................... (24)
At the middle surface, where z = 0, the right-hand member of (23)
becomes
e-,,(1_?)_je-w(1 + J?.) + ................ (25)
Initially
as was required. If we put e~p^ = T, (25) may be written
...; ......... (26)
and (26) may be tabulated as a function of T, and thence of t. It vanishes
when T=l and when T=0. The maximum value occurs when jT='747.
When T is less than this, which corresponds to an increased value of t, only
the first two or three terms in (26) need be regarded. The above value
of T gives
pj = -292 ;
and if, as for glass, the diffusivity for heat in C.G.s. measure be '004, we get
(27)
Thus if a plate of glass be one centimetre thick, so that c = 1, the light
seen in the polariscope at the centre of the thickness is a maximum about
7£ seconds after heat is applied to the faces.
1901] AND ON THE DOUBLE REFRACTION RESULTING THEREFROM.
507
The following small table will give an idea of the relation between
(26) and T.
T
(26)
T
(26)
o-o
o-oooo
0-6
0-2139
0-1
0-0363
0-7
0-2381
0-2
0-0727
0-8
0-2371
0-3
0-1090
0-9
0-1823
0-4
0-1453
1-0
o-oooo
0-5
0-1809
In his paper above referred to, Hopkinson considered the strains produced
by unequal heating in a spherical mass, under the supposition that the
temperature was everywhere the same at the same distance from the centre.
A similar analysis applies in the two-dimensional problem, which is of greater
interest from the present point of view. We suppose that everything is
symmetrical with respect to an axis, taken as axis of z, and that 6 is a
function of r, equal to *j(a? + y2), only. The displacements in the directions
of z and r will be denoted by w and u ; in the third direction, perpendicular
to z and r, there is supposed to be no displacement.
We may commence with the strictly two-dimensional case where w = 0
throughout. This implies a stress R whose magnitude is given by
in which
represents the dilatation.
The other principal stresses operative radially and tangentially are
(30)
(31)
The equation of equilibrium, analogous to (5), is obtained by considering
the stresses operative upon the polar element of area. It is
508 ON THE STRESSES IN SOLID BODIES DUE TO UNEQUAL HEATING, [265
Substituting from (30), (31), we get
d?u 1 du u _ 7 dO
dr* + ^r dr~r* = \ + fy dr '
so that
du u yd
where a. is an arbitrary constant. Integrating a second time we find
.................. (34)
fr
J o
in which, however, £ must vanish, if the cylinder is complete through r = 0.
From (34)
\Terdr, . ...(35)
a.-r^J Mr-^T.> (36>
and
*-**&&-*£
It is on (P — Q) that the double refraction depends when light traverses
the cylinder in a direction parallel to its axis.
In (35), (36), (37)
- I 0rdr
represents the mean temperature (above the standard) of the solid cylinder
of radius r. It is to be remarked that the double refraction of the ray at
r is independent of the values of 0 beyond r, and also of any boundary-
pressure. If 0 increases (or decreases) continuously from the centre out-
wards, the double refraction never vanishes, and no dark circle is seen in the
polariscope.
In the above solution if the cylinder is terminated by flat faces, we
must imagine suitable forces R, given by (28), to be operative over the faces.
The integral of these forces may be reduced to zero by allowing a suitable
expansion parallel to the axis. Regarding dw/dz as a constant (not necessarily
zero), independent of r and z, we have in place of (28)
The additions to P and Q are \dw/dz, while (P - Q) remains unchanged.
If the cylinder is long relatively to its diameter, the last state of things
may be supposed to remain approximately unchanged, even though the
1901] AND ON THE DOUBLE REFRACTION RESULTING THEREFROM. 509
terminal faces be free from applied force. In the neighbourhood of the ends
there will be local disturbances, requiring a more elaborate analysis for their
calculation, but the simple solution will apply to the greater part of the
length.
The case of a thin plate whose faces are everywhere free from applied
force is more difficult to treat in a rigorous manner, but the following is
probably a sufficient account of the matter. By supposing R = 0 in (38)
we get
(X4^g^-x(**?); (39)
and using this value of diujdz,
Comparing these with (30), (31), we see that the only difference is that
X and 7 of those equations are now replaced by
,
'
Hence, instead of (37), we should have
(42)
and the same general conclusions follow.
In the preceding calculations we have supposed that the solid is free
from stress at a uniform standard temperature when u, v, w vanish. In
the case of unannealed glass, it would require a variable temperature to
relieve the material from stress. To meet this, 6 in the above equations
would have to be reckoned from the variable temperature corresponding
to the state of ease, rather than from a uniform standard temperature.
Some of the questions above considered are easily illustrated experi-
mentally. A slab of glass about 8 cm. square and 1 cm. thick, polished
upon opposite edges, when placed in the polariscope shows but little revival
of light so long as the temperature is uniform. The contact of the hands
with the two faces suffices to cause an almost instantaneous illumination,
rising to a maximum at the middle of the thickness after a few seconds.
Dark bands situated about halfway between the middle and the faces are
a conspicuous feature. After about 30 or 40 seconds the light fades greatly,
a result more rapidly attained if the hands be removed after 10 or 20 seconds'
contact. In the earlier stages of the heating thfe outside layers are the
510 ON STRESSES DUE TO UNEQUAL HEATING. [265
warmer, and being prevented from expanding fully are in a condition of
compression. The inner layers at the same time are in tension, a conclusion
that may be verified by interposition of another piece of glass, of which
the mechanical condition is known, and of which the effect may be either
an augmentation or a diminution of the light.
An examination in the polariscope of the so-called toughened glass, intro-
duced a few years ago, is interesting. It was understood to be prepared
by a sudden cooling in oil while still plastic with heat. When it is examined
through the thickness of the sheet, a great want of uniformity is manifested.
In spite of the shortness of the distance traversed, there is in places con-
siderable revival of light with intermediate irregularly disposed dark bands.
The course of these bands is altered when by fracture any part is relieved
from the constraining influence of neighbouring parts. To make an examina-
tion by light transmitted edgewise, it was necessary to immerse the glass in a
liquid of nearly equal refractivity (benzole with a little bisulphide of carbon)
contained in a small tank. The width, traversed by the light, was about
] cm. In this way, and with the aid of a magnifier, the condition of the
various layers could be well made out. The dark bands of no double refraction
seemed to be nearer to the faces than according to the calculation made
above, but the whole thickness is so small that this observation is scarcely
to be relied upon. The interior was in a state of tension, and the double
refraction was nearly sufficient at the middle to give the yellow or brown
of the first order. By the action of hydrofluoric acid on the lower end of one
of the strips the outermost layers were dissolved away. This caused a drawing
together of the dark bands towards the middle, and though a good deal
remained the light was much reduced.
The cause of the toughening has been sought in a special crystalline
condition due to the sudden cooling. There may be something of this nature ;
but it would seem that most of the peculiarities manifested may be explained
by reference to the known condition of stress. The fracture of glass is usually
due to bending, and the failure occurs at the surface which is under tension.
If, initially, the superficial layers are under strong compression, a degree
of bending may be harmless which otherwise would cause fatal results. It
seems possible also that the superficial compression may be the explanation
of the special hardness observed.
A short length of glass rod in its natural imperfectly annealed condition
may be used to illustrate symmetrical stress. The ends may be ground, and
[then] either polished or provided with cover-glasses cemented with Canada
balsam. In the specimen examined by me the colours varied from the
black of the first order on the axis to the red of the second order near the
surface. The length of the cylinder was T6 cm. and the diameter T8 cm.
266.
ON A NEW MANOMETER, AND ON THE LAW OF THE
PRESSURE OF GASES BETWEEN 1-5 AND O'Ol MILLI-
METRES OF MERCURY.
[Phil. Trans. CXCVIA. pp. 205—223, 1901.]
Received January 15, — Read February 21, 1901.
Introduction.
THE behaviour of air and other gases at low densities is a subject which
presents peculiar difficulties to the experimenter, and highly discrepant
results have been arrived at as to the relations between density and pressure.
While Mendeleef and Siljerstrom have announced considerable deviations
from Boyle's law, Amagat* finds that law verified in the case of air to the
full degree of accuracy that the observations admit of. In principle Amagat's
method is very simple. The reservoir consists mainly of two nearly equal
bulbs, situated one above the other and connected by a comparatively narrow
passage. By the rise of mercury from a mark below the lower bulb to
another on the connecting passage, the volume is altered in a known ratio
which is nearly that of 2 : 1. The corresponding pressures are read with a
specially constructed differential manometer. Of this the lower part which
penetrates the mercury of the cistern is single. Near the top it divides into
a U, widening at the level of the surface of the mercury into tubes of
2 centims. diameter. Higher up again these tubes re-unite and by means of
a three-way tap can be connected either with an air-pump or with the upper
bulb. Suitable taps are provided by which the two branches can be isolated
from one another. During the observations one branch is vacuous and the
other communicates with the enclosed gas, so that the difference of levels
represents the pressure. This difference is measured by a cathetometer.
It is evident that when the pressure is very low the principal difficulty
relates to the measurement of this quantity, and that the errors to be feared
in respect to volume and temperature are of little importance. Amagat,
fully alive to this aspect of the matter, took extraordinary pains with the
manometer and with the cathetometer by which it was read. An insidious
* Ann. de Chimie, Vol. xxvni. p. 480 (1883).
512 ON A NEW MANOMETER, AND ON THE LAW OF THE PRESSURE [266
error may enter from the refraction of the walls of the tubes through which
the mercury surfaces are seen. But after all his precautions Amagat found
that he could not count upon anything less than -^ millim., even in the
means of several readings. It may be well to give his exact words (p. 494): —
" Dans les experiences dont je donnerai plus loin les resultats numeriques,
les determinations sont faites en ge"ne"ral en alternant cinq fois les lectures
sur chaque menisque ; les lectures etaient faites au demi-centieme, et les
divergences dans les series regulieres oscillent ordinairement entre un
centieme et un centieme et demi ; en prenant la moyenne, il ne faut pas
compter sur plus d'un centieme ; et cela, bien entendu, sans tenir compte des
causes d'erreur independantes de la lecture cathetome'trique
Les resultats numeriques consignes aux Tableaux que je vais donner main-
tenant sont eux-memes la moyenne de plusieurs experiences ; car, outre que
les lectures ont et^ faites en general cinq fois en alternant, on est toujours,
apres avoir reduit le volume a moitie, revenu au volume primitif, puis au
volume moitie : chaque experience a done ete faite aux moins deux fois, et
souvent trois et quatre."
The following are the final results for air : —
Pression
initiale
en millims.
pv
p'v'
Pression
initiale
en millims.
pv
p'v'
Pression
initiale
en millims.
pv
p'v'
millims.
millims.
millims.
12-297
0-9986
3-770
1-0019
1-377
1-0042
12-260
1-0020
3-663
0-9999
1-316
1-0137
10-727
0-9992
3-165
1-0015
1-182
1-0030
7'462
1-0013
2-531
1-0013
1-140
1-0075
7-013
1-0015
2-180
1-0015
1-100
0-9999
6-210
1-0021
1-898
1-0050
0-978
1-0160
6-160
1-0025
1-852
0-9986
0-958
1-0100
4-946
1-0010
1-751
[1]-0030
0-860
1-0045
4-275
1-0048
1-457
1-0150
0-295
0-9680
3-841
1-0027
1-414
1-0143
Since, as it would appear, the " initial " pressure is the smaller of a pair,
the lowest pressure concerned is about "3 millim. of mercury, and the error at
this stage is about 3 per cent. It is not quite clear which is which of pv
and p'v. For while it is expressly stated that p is smaller than p, the
value of v'jv is given at 2'076. I think that this is really the value of vfv'.
But any lingering doubt that may be felt upon this point is of no con-
sequence here, inasmuch as Amagat's comment upon the tabular numbers is
" On ne saurait done se prononcer, ni sur les sens ni meme sur 1'existence de
ces ecarts."
1901] OF GASES BETWEEN 1'5 AND O'Ol MILLIMETRES OF MERCUKY. 513
After such elaborate treatment by the greatest authority in these matters,
the question would probably have long remained where Amagat left it, had
not C. Bohr found reason to suspect the behaviour of oxygen at low pressures.
This led to a prolonged and apparently very careful investigation, of which
the conclusion was that at a pressure of 7 millim. of mercury the law con-
necting pressure and volume is subject to a discontinuity.
" 1. Bei einer Temperatur zwischen 11° und 14° C. weicht der Sauerstoff
innerhalb der beobachteten Druckgrenzen von dem Boyle-Mariotte'schen
Gesetze ab. Die Abhangigkeit zwischen Volumen und Druck fiir einen
Werth des letztgenannten, grosser als 070 mm., kann man annahernd durch
die Formel
(p + 0109) v = k
ausdriicken, wahrend die Formel fur Werthe der Drucke, welche kleiner als
070 mm. sind :
(p + 0-070) v = k
ist.
2. Sinkt der Druck unterhalb 070 mm., so erleidet der Sauerstoff eine
Zustandsveranderung ; er kann wieder durch ein Erhb'hen des Druckes bis
liber 070 mm. die urspriingliche Zustandsform ubergefuhrt werden*."
Fig; l.
I O 2O 3O 4O SO 80 70 8O 9O IOO
Fig. 1 is a reproduction of one of Bohr's curves, in which the ordinate
represents pv and the abscissa represents p on such a scale that 1 millim. of
mercury corresponds to the number 20. It will be seen that at the place of
discontinuity a change of pv to no less than -^ of its amount occurs with no
perceptible concomitant change in the value of p. In the neighbourhood of
the discontinuity the pressure is uncertain. Thus (p. 475) " Wenn man bei
einer gewissen Sauerstoffmenge im Rohre a das Quecksilber erst in der Art
einstellt, dass der Druck einen etwas geringeren Werth als 070 millim. hat,
und dann durch Verringern des Volumens den Druck tiber 0*70 millim.
steigert (z.B. bis 0'8 millim.), so zeigt sich, dass dieser Druck nicht constant
bleibt, sondern im Verlaufe von 3 — 5 Stunden bis zu einem Werthe sinkt, der
ungefahr 10 Proc. kleiner ist, als der ursprtingliche."
* Wied. Ann. Vol. xxvn. p. 479 (1886).
R. iv. 33
514 ON A NEW MANOMETER, AND ON THE LAW OF THE PRESSURE [266
So far as I am aware, no attempt to repeat Bohr's difficult and remarkable
experiments has been recorded, but some confirmation of anomalous behaviour
of oxygen in this region of pressure is afforded by the observations of Ramsay
and Baly*. Sutherland! interprets the results as a " Spontaneous Change of
Oxygen into Ozone and a Remarkable Type of Dissociation," and connects
therewith some observations of Crookes relating to radiometer effects in
oxygen gas. On the other hand, chemical tests applied by Professor Threlfall
and Miss Martin J failed to indicate the presence of ozone in suitably expanded
oxygen.
Improved Apparatus for Measuring very small Pressures.
In spite of the interest attaching to the anomaly encountered by Bohr,
I should hardly have ventured to attack the question experimentally myself,
had I not seen my way to what promised to be an improved method of dealing
with very small pressures. In operations connected with the weighing of
gases, extending over a series of years, I have had much experience of a
specially constructed manometric gauge in which an iron rod, provided above
and below with suitable points, is actually applied to the two mercury
surfaces arranged so as to be situated in the same vertical line§. Although
two variable quantities had to be adjusted — the pressure of the gas and the
supply of mercury — no serious difficulty was encountered ; and the delicacy
obtained in the observation of the approximation of a point and its image in
the mercury surface with the assistance of an eye-lens of 25 millims. focus
was very satisfactory. In order to get actual measures of the delicacy, a
hollow glass apparatus in the form of a fork was mounted upon a levelling
table. The stalk below was terminated with a short length of rubber tubing
compressible by a screw. This allowed the supply of mercury to be adjusted.
The mercury surfaces in the U were about 20 millims. in diameter, and were
exposed to the air. They were to be adjusted to coincidence with needle
points, rigidly connected to the glass-work, by suitable use of the compressor
and of the screw of the levelling table. Readings of the latter in successive
and independent settings showed that a degree of accuracy was attainable
much superior to the limit fixed by Amagat for the best work with the
cathetometer. It is unnecessary to record the numbers obtained at this
stage of the work, inasmuch as the final results to be given below prove that
the errors of setting are considerably less than y^1^ millim.
It will now be possible to form a preliminary idea of the proposed mano-
meter. The readings of the levelling screw, obtained as above, may be
* Phil. Mag. Vol. xxxvm. p. 301 (1894).
+ Phil. Mag. Vol. xmi. p. 201 (1897).
J Proc. Roy. Soc. of New South Wale*, 1897.
§ " On the Densities of the Principal Gases," Proc. Roy. Soc. Vol. LIH. p. 134, 1893. [Vol. iv.
p. 39.1
1901] OF GASES BETWEEN 1'5 AND O'Ol MILLIMETRES OF MERCURY. 515
regarded as corresponding to the zero of pressure, or rather of pressure
difference. If the pressures operative upon the mercury surfaces be slightly
different, the setting is disturbed ; and the change of reading at the screw
required to re-establish the adjustment represents the difference of pressures.
In order to interpret the result absolutely it is only necessary to know further
the pitch of the levelling screw, the leverage with which it acts, and the
distance between the points to which the mercury surfaces are set. If the
space over one mercury surface be
vacuous, the change of reading at
the levelling screw represents the
absolute pressure in the space over
the other mercury surface.
The difficulty, which will at once
present itself to the mind of the
reader, in the use of a manometer on
this plan, is the necessity for a flex-
ible connexion between the instru-
ment and the rest of the apparatus,
such as the air-pump and the vessel
in which the pressure is required to
be known. With the aid of short
lengths of rubber tubing this re-
quirement could be easily met, but
the class of work for which such a
manometer is wanted would usually
preclude the use of rubber. In my
apparatus the requisite flexibility is
obtained by the insertion of con-
siderable lengths (3 metres) of glass
tubing between the manometer and
the parts which cannot turn with
it. Although the adjustment was
made by the screw of a levelling
table as described, the actual readings
were taken by the mirror method,
the supports of the mirror being
connected as directly as possible
with the pointswhose angular motion
is to be registered. In this way we
become independent of the rigidity
of the glass-work, and are permitted to use wood freely in the levelling
table and in its supports. It frequently happened that an adjustment left
correct was found to be out after an interval. The screw had not been
33—2
516 ON A NEW MANOMETER, AND ON THE LAW OF THE PRESSURE [266
moved, but the mirror-reading was altered. On resetting by use of the
screw, the original mirror-reading was recovered within the limits of
error.
Fig. 3.
The essential parts of the manometer, as finally employed, are shown
(Fig. 2) in elevation and plan, and the general scheme of the mounting is
indicated in Fig. 3. At A is the stalk of the glass fork, of such length that
the mercury in the hose below is always at a pressure above atmosphere ;
B, B are bulbs of about 25 millims. diameter, at the centres of which are
situated the points. These are of glass *, which need not be opaque ; and
they must be carefully finished upon a stone. A considerable degree of
sharpness is desirable, but similarity is more important than the extreme of
sharpness. In the actual apparatus complete similarity was not attained, and
in the first trials the difference was rather embarrassing. However, after a
little practice the eye becomes educated to set the mercury to each point in
* At first iron needle points were tried.
1901] OF GASES BETWEEN T5 AND O'Ol MILLIMETRES OF MERCURY. 517
a constant manner, and this is all that is really required. The same con-
sideration shows that minute outstanding capillary differences should not
lead to error. It may be remarked that the mercury is always on the rise at
the time of adjustment, and in fact it was found best to make it a rule not
to allow the points to be drowned at any time when it could be avoided.
After such a drowning it was usually (perhaps always) found that the
mercury surface was disturbed by the proximity of the points without actual
contact, an effect attributed to electrification.
The presentation of the point to the mercury, or rather of the point to
its image as seen by reflection in the mercury, was examined with the aid of
two similar eye-lenses (not shown) of 22 millims. focus. The illumination,
from a small gas flame suitably reflected by mirrors, was from behind, and it
and the lenses were so arranged that both points could be seen without a
motion of the head. Precautions were required to prevent the radiation
from the gas flame and from the observer from producing disturbance,
especially by unequal heating of the two limbs of the U. The U itself was
well bandaged up, and between it and the observer were interposed sheets
of copper and of insulating material so as to ensure that at all events
there should be no want of symmetry in any heating that might take place.
The adjustment itself is a double one, requiring both the use of the
levelling screw J and an accurate feed of mercury. The hose terminates
as usual in a small mercury reservoir D. This facilitates the preliminary
arrangements, but in use the reservoir is cut off by a screw clamp E just
below it. The rough adjustment of the supply of mercury is effected by
a large wooden compressor F. The fine adjustment required for the actual
setting is a more delicate matter. The first attempts were by fine screw
compressors acting upon the pendent part of the hose, but the tremors thence
arising were found very disturbing. A remedy was eventually applied by
operating upon the part of the hose which lies flat upon the floor or rather
on the bottom of a mercury tray. The compressor is shown at G, Fig. 3 ;
the screw being provided with a long handle H to bring it within con-
venient reach. The advantage accruing from this small device would scarcely
be credited.
The glass-work is attached by cement to a board, which hangs down-
wards in face of the observer and is itself fixed rigidly to the levelling stand
K. This is supported at two points /, which define the axis of rotation, and
by a finely adjustable screw J, within reach of the observer. The whole
stands in a very steady position upon the floor of an underground cellar in
my country house.
The arrangements for the connexion of the mirror must now be de-
scribed. The glass stems, whose lower extremities form the "points," are
prolonged upwards by substantial tubing, and terminate above in three
518 ON A NEW MANOMETER, AND ON THE LAW OF THE PRESSURE
slightly rounded ends, L, L, suitable for the support of the mirror plat-
form N. The two supports necessary on the left are obtained by a
symmetrical branching of the tube on that side. The platform is of worked
glass, so that a slight displacement of the contacts has no effect on the
slope of the mirror. The latter is of worked glass silvered in front. Suitable
stops are provided to guide the mirror platform into the right position and to
prevent accidents, but these exercise no constraint.
The axis // about which the apparatus rotates is horizontal and parallel
to the face of the mirror, so that the sine of the angle 6 of rotation from the
zero position represents the difference of levels of the mercury surfaces.
The axis // lies approximately in the mirror surface and at about the middle
of the height of the operative part. The rotation of the mirror is observed
in the usual way by means of a telescope and vertical millimetre scale. The
aperture of the object-glass is 30 millims., and the distance from the mirror
3150 millims. The readings can be taken to about '1 millim.
In many kinds of observation the zero can only be verified at intervals,
as it requires the pressures over the mercury to be equalised. On the whole
the zero was tolerably constant to within two or three-tenths of a millimetre
of the scale. A delicate level was attached to the telescope to give warning
of any displacement of the stand (all of metal) or of the ground.
The differences of pressure to be evaluated are not quite in simple propor-
tion to the scale reading from zero. The latter varies as tan 20, while the
former varies as sin 6. The correcting factor is therefore
f lm~20 = X - 1*9* approximately.
If the zero reading (in millimetres) be a, and the current reading as, D the
distance between telescope and mirror,
a x — a . . .
8 = —nj) approximately;
so that the correcting factor is
The actual correction to be applied to (x — a) is thus
In practice (x — a) rarely exceeded 350, for which the correction would
be — 1'6. When (x — a) falls below 120, the correction is insensible.
The next question is the reduction to absolute measure. What (cor-
rected) scale-reading corresponds to 1 millim. actual difference of mercury
levels ? The distance between the points is 27'3 millims., so that 1 millim.
1901] OF GASES BETWEEN 1'5 AND O'Ol MILLIMETRES OF MERCURY. 519
mercury corresponds to 231 millims. of the telescope scale. The highest
pressure that could be dealt with is about 1^ millims. of mercury.
The above reckoning proceeds upon the supposition that the distance
between the points can be regarded as invariable. Certain small discrepancies
manifested at the higher slopes of the apparatus induced me to examine the
question more particularly, for it seemed not impossible that owing to the
bending of the glass-work some displacement might occur. But a rather
troublesome measurement of the actual distance in various positions by means
of microscopes negatived this idea. I would however recommend that this
point be kept specially in view in the design of any subsequent apparatus of
this kind.
Experiments to determine the Relation of Pressure and Volume at given
Temperature.
In order to test Boyle's law one of the lateral branches C is connected to
the air-pump and the other to the chamber in which the gas is contained.
The pump is of the Topler form, and is provided with a bulb containing
phosphoric anhydride. No tap or contracted passage intervenes between the
pump-head and B. A lateral channel communicates with a three-way tap, by
which this side of the apparatus can be connected with the gas-generating
vessel. The third way leads to a blow-off under mercury more than a
barometer-height below.
The two sides of the apparatus are connected by a cross-tube which can
be closed or opened by means of a tap. The plug of this tap is provided
with a wide bore. When it is intended to read the zero, the tap is open.
If desired, the mercury may be raised in the Topler so as to prevent the
penetration of gas into the pump-head. When pressures are to be observed,
the tap of the cross-tube is closed, and a good vacuum is made on the pump
side. No particular difficulty was experienced with the vacuum. In the use
of the Topler the mercury was allowed to flow out below, and was trans-
ferred at intervals to the movable reservoir. The latter was protected from
atmospheric moisture by a chloride of calcium tube. When, after standing
five or ten minutes, the mercury was put over, and, on impact, gave a hard
metallic sound with inclusion of no more than a small speck of gas, the
vacuum was nearly sufficient, and no further change could be detected at the
manometer. The capacity of the pump-head was two or three times that of
the remaining space to be exhausted.
In the earlier experiments the gas-containing tube, placed vertically, was
graduated to 50 cub. centims. at intervals of 10 cub. centims. Prolonged
below by more than a barometer-height of smaller tubing, it terminated in a
hose and mercury reservoir, the latter protected by chloride of calcium. In
520 ON A NEW MANOMETER, AND ON THE LAW OF THE PRESSURE [266
order to get rid of most of the adherent moisture and carbonic anhydride,
the tubes on both sides of the apparatus were heated pretty strongly in a
vacuous condition. The first trial was with oxygen, in the hope of at once
obtaining a confirmation of Bohr's anomaly; but not succeeding in this,
I fell back upon nitrogen and hydrogen. With a vacuum on the pump side,
readings of pressure were taken with the mercury in the chamber at 0 and
at 50 cub. centims., and the ratio of pressures (about 2:1) was deduced.
When this had been repeated, some of the gas was allowed to escape by
opening the cross-tap, the zero was again observed, and the vacuum re-
established on the pump side. Another ratio of pressures could now be
obtained, corresponding to the same (unknown) volumes as before, but to a
different total pressure.
In utilising the ratios of pressure thus obtained, it was of course necessary
to consider how far the temperature could be assumed to be unchanged
within each pair of pressures brought into comparison. The general tem-
perature of the cellar was extremely uniform, and no difference could be
read upon a thermometer worth taking into account. Passing over this
question for the present, we may consider how far the results conformed to
Boyle's law. The agreement of the ratios, except, perhaps, at the highest
pressures of about 1^ millims. of mercury, was sufficiently good, and of itself
goes a long way to confirm Boyle's law. In strictness, all that the constancy
of the ratio can prove is that the relation between pressure (p) and density
(p) is of the form
p = *pn, (1)
where n is some numerical quantity. To limit n to the value unity, the
constancy of the ratios might be followed up into the region of pressure
for which Boyle's law is known to hold, but this can scarcely be said to
have been done here. Otherwise, we need to know what the ratio of densities
in the two positions of the mercury really is, and not merely that it remains
constant.
In the case of the original volume chamber the first was the method
employed. The smaller volume, defined by the upper mark in the volume
tube and by the "point" in the manometer, was filled with dry air at a
known atmospheric pressure. The included air was then isolated and
expanded until it occupied the larger (approximately double) volume, and
the new pressure determined by observation of the difference of levels in
the tube and in a mercury reservoir similarly fashioned. The operation was
rather a difficult one, and the result was only barely accurate enough. The
ratio of volumes thus determined by use of Boyle's law, as applied to air
at atmospheric and half atmospheric pressures, agreed sufficiently well with the
ratio of pressures found by the manometer for rare hydrogen and nitrogen ;
1901] OF GASES BETWEEN 1'5 AND O'Ol MILLIMETRES OF MERCURY. 521
and thus Boyle's law may be considered to be extended to these rare gases.
The rarefaction was carried down to a total pressure of only '02 millim. At
this stage discrepancies of the order of 5 per cent, are to be expected.
Having obtained fairly satisfactory results with hydrogen and nitrogen,
I returned to oxygen, fully expecting to verify the anomalous behaviour
described by Bohr. In this I have totally failed. The gas was prepared by
heating permanganate of potash, dried by phosphoric anhydride, and may be
regarded as fairly pure. The region of pressure round '7 millim. was carefully
examined, use being made of the intermediate divisions of the 50 cub.
centims. range of volume. No unsteadiness of the kind indicated by Bohr,
or appreciable departure from Boyle's law, was detected. And when the
pressures were diminished down to a few hundredths of a millimetre, there
was no falling off in the product of pressure and volume. The observations
were repeated a second time with a fresh supply of oxygen.
The experience gained up to this date (August, 1900) showed that the
manometer worked well, and that there was no difficulty about the vacuum,
but I was not altogether satisfied with the way in which the volumes had
been determined. There was some want of elegance, to say the least, in
using Boyle's law for this purpose, and barely adequate accuracy in the appli-
cation itself. The latter objection might have been overcome by the use of
a suitable cathetometer, but such was not to hand. The most direct method
by actually gauging with mercury the spaces concerned being scarcely
feasible, I devised another method which has the advantage of easy execution
and* is practically independent of the assumption of Boyle's law. The
opportunity was taken to increase the range over which the volume could be
varied.
The new chamber, composed mainly of tubing of 18 millims. diameter, is
graduated at intervals of 10 cub. centims. over a total range of 200 cub.
centims. It is prolonged above and below by narrow tubing in order to
connect it with the sloping manometer bulb and with the hose and mercury
reservoir as before. The zero mark is situated on the upper tube a few
centimetres above its junction with the wider one. It is scarcely necessary
to say that no rubber was employed except for the hoses, and that these
were always occupied by mercury under a pressure above atmosphere. The
mercury reservoirs themselves were protected against damp by chloride of
calcium.
If we call the ungauged volume (from the zero mark to the bulb of the
sloping manometer with "point" set) V, and the gauged volume v, the total
volume occupied by the gas is V + v ; and the problem is how to determine V.
If we may assume the correctness of Boyle's law for rare gases and may rely
upon the sloping manometer, the process is simple enough. We have only to
find the pressures exerted by the included gas at volumes V and V + v, whence
522 ON A NEW MANOMETER, AND ON THE LAW OF THE PRESSURE [266
by Boyle's law the ratio of these volumes is known and thus V determined
in terms of v. In order to avoid the use of Boyle's law, further observations
are necessary.
The requisite data can be obtained by changing the quantity of gas.
Suppose that with the original quantity of gas certain pressures, P, P', corre-
spond to total volumes, V + vlt V + v2, and that with a reduced amount of gas
the same pressures are recorded with volumes V+v3, V + v4. Since the
pressure is a function of the density, whether Boyle's law be applicable or not,
it must follow that
V + v, V+v3
r+^-FTv .............................. 9
whence V is determined in terms of the known volumes vlt v2, v3, v4. It may
be remarked that this argument does not assume even the correctness of the
scale of pressures.
In carrying out the method practically it was necessary to work to the
fixed marks of the volume chamber, and thus the same pressures could not be
recovered exactly. But the use of Boyle's law in order to make what is
equivalent to small corrections is unobjectionable.
With this explanation it may suffice to give the details of an actual
determination executed with nitrogen. With the original quantity of gas,
volumes F+70, F+170 gave pressures proportional to 345'4, 184'9.
Sufficient gas was now removed to allow the remainder to give nearly the
same higher pressure as before with v = 0. Thus, corresponding to volumes
F+ 0, V + 40 the pressures were 344'9, 183'3. We have now only to calcu-
late V from the equation
V + 40 344-9 184-9 F+170
345^4 F+70 '
or F» + 110F+ 2800 = 1-0072 (F2 + 170F) ;
whence F = 45'5 cub. centims.
The adopted value, derived from observations upon nitrogen and hydrogen, is
F = 45 '6 cub. centims.
In charging the apparatus, the first step is to make a good vacuum
throughout, the cross-tap being open. The gas supply being started, the
first portions are allowed to blow off from under mercury, and then, by use
of the three-way tap, a sufficiency is introduced into the apparatus to an
absolute pressure of, perhaps, 10 centims. of mercury. The gas-leading tube
would then be sealed off. Ultimately the remainder of the supply tube and
the blow-off tube were exhausted to diminish the risk of leakage.
1901] OF GASES BETWEEN 1'5 AND O'Ol MILLIMETRES OF MERCURY. 523
The " nitrogen" was prepared from air by passage over red-hot copper and
desiccation with phosphoric anhydride. Accordingly it contained argon to
the amount of about 1 per cent.
In taking a set of observations the procedure would be as follows.
Assurance having been obtained that the vacuum was good, the next step
would be to set the mercury in the volume chamber so that v = 190 cub.
centims., then after a few minutes to adjust the sloping manometer and to
read the telescope scale. It was of course necessary to ensure that sufficient
time was allowed for uniformity of pressure to establish itself, and observa-
tions were frequently renewed after a quarter of an hour or longer. In the
case of oxygen, to be considered later, several hours were sometimes allowed.
If operations were leisurely conducted, with first a rough setting of the
volume and then a rough setting of the manometer followed by accurate
settings in the same order, little or no change could afterwards be detected.
Indeed I was rather surprised to find how rapidly equilibrium seemed to be
established. The next smaller volume, e.g., v = 150, would then be observed,
and so on until v = 40. In observations to be used for the examination of
Boyle's law v was not further reduced, as too much stress might thereby be
thrown upon the accuracy of V. The same observations were then repeated
in reverse order and the mean taken. The numbers recorded are thus the
mean of two settings only of the manometer.
The next step was to allow about half the gas to escape. The mercury
at the pump was allowed to rise so as to cut off the pump-head and V + v was
so adjusted as to be equal to the volume remaining upon the other side, about
130 cub. centims. The cross-tap was then opened, and after a sufficient interval
of time the zero, corresponding to no pressure, was read. In the course of the
observations upon nitrogen, extending over ten days, the zero varied from
43'5 to 43'8. Whenever possible the zero used for a set was the mean of
values found before and after.
The annexed tables give the results for nitrogen in detail. In Table I.,
dealing with the highest quantity of gas, the first column gives the volume
( V = 45'6 cub. centims.) ; the second represents the pressure, being the mean
of the two actually read numbers (expressing millimetres of telescope scale)
less the zero reading 437 and corrected to infinitely small arcs as already
explained. The third column is the logarithm of the product of the first
two, and should be constant if Boyle's law holds. The fourth column gives
the approximate value of the pressure in millimetres of mercury; the
fifth the deviation of pv from the mean taken as unity. In the sixth
column is shown the amount by which the observed value of p exceeds
that requisite in order to make pv constant, expressed in millimetres of
mercury.
524 ON A NEW MANOMETER, AND ON THE LAW OF THE PRESSURE [266
TABLE I. — Nitrogen.
November 9-11, Zero = 43'7.
Volume in
cub. ceiitims.
Pressure in
scale divisions
Log. product
Pressure in
millims. Hg
Deviation
of pv
Error of p
in millims.
V+ 70
345-4
•6013
1-49
+ •0002
+ •0003
V+ 80
318-3
•6018
1-38
+ •0014
+ •0019
V+ 90
294-1
•6007
1-27
-•0012
-•0015
F+110
256-8
•6016
1-11
+ •0009
+ •0010
F+130
227-4
•6013
•98
+ •0002
+ •0002
F+150
203-7
•6004
•88
-•0018
-•0016
F-f 170
184-9
•6005
•80
-•0014
-•0011
F+190
169-8
•6021
•73
+ -0021
+ •0015
•6012
TABLE II. — Nitrogen.
November 11-12, Zero = 437.
Volume in
cub. centims.
Pressure in
scale divisions
Log. product
Pressure in Deviation
millims. Hg \ of pv
Error of p
in millims.
F+ 0
344-9
•1966
1-49
+ •0007
+ •0010
F+ 10
282-3
•1958
1-22
-•0012
-•0015
F+ 20
239-5
•1962
1-04
-•0002
-•0002
F+ 40
183-3 .
•1956
•79
-•0016
-•0013
F+ 60
148-8
•1963
•64
•0000
•oooo
F+ 80
125-2
•1966
•54
+ •0007
+ •0004
F+110
101-1
•1968
•44
+ O012
+ •0005
F+150
80-2
•1955
•35
-•0018
-•0006
F+190
66-9
•1976
•29
+ •0030
+ •0009
•1963
1901] OF GASES BETWEEN 1'5 AND O'Ol MILLIMETRES OF MERCURY. 525
TABLE III. — Nitrogen.
November 13, Zero = 43'6.
Volume in
cub. centims.
Pressure in
scale divisions
Log. product
Pressure in
inillinis. Hg
Deviation
of pv
Error of p
in millims.
V+ 40
91-1
•892
•394
•ooo
•oooo
V+ 60
73-9
•892
•320
•ooo
•oooo
V+ 80
62-3
•893
•269
+ •002
+ •0005
F+110
50-2
•893
•217
+ •002
+ •0004
F+ 150
39-6
•889
•171
-•007
-•0012
F+190
33-1
•892
•892
•143
•ooo
•oooo
TABLE IV.— Nitrogen.
November 14, Zero = 43*5.
Volume in
cub. centims.
Pressure in
scale divisions
Log. product
Pressure in
millims. Hg
Deviation
of pv
Error of p
in millims.
F+ 40
46-0
•595
•199
+ •005
+ •0010
F+ 60
37-1
•593
•160
•ooo
•oooo
F+ 80
31-1
•592
•135
-•002
-•0003
F+110
25-1
•592
•109
-'002
-•0002
F+150
20-1
•595
•087
+ •005
+ •0004
F+190
16-5
•590
•071
-•007
-•0005
•593
TABLE V. — Nitrogen.
November 16, Zero = 43'5.
Volume in
cub. centims.
Pressure in
scale divisions
Log. product
Pressure in
millims. Hg
Deviation
of pv
Error of p
in millims.
F+ 40
22-8
•290
•099
•ooo
•oooo
F+ 60
18-6
•293
•081
+ •007
+ •0006
F+ 80
15-6
•292
•067
+ •005
+ •0003
F+110
127
•296
•055
+ •014
+ •0008
F+150
9-9
•287
•043
-•007
-•0003
F+190
8-15
•283
•290
•035
-•016
-'0006
526 ON A NEW MANOMETER, AND ON THE LAW OF THE PRESSURE [266
TABLE VI.— Nitrogen.
November 17-18, Zero = 43'7.
Volume in
cub. centims.
Pressure in
scale divisions
Log. product
Pressure in
millims. Hg
Deviation
oipv
Error of p
in millims.
V+ 40
11-40
•989
•049
+ •005
+ '0002
V+ 60
9-10
•983
•039
-•009
- -0004
V+ 80
7-65
•983
•033
-•009
-•0003
7+110
6-25
•988
•027
+ •002
+ •0001
7+150
5-10
•999
•022
+ 028
+ •0006
7+190
4-05
•980
•017
-•016
- '0003
•987
TABLE VII.— Nitrogen.
November 18-19, Zero = 43'8.
Volume in
cub. centims.
Pressure in
scale divisions
Log. product
Pressure in
millims. Hg
Deviation
of pv
Error of p
in millims.
F+ 40
5-90
•703
•026
+ •014
+ •0004
7+ 60
4-60
•686
•020
-•026
-•0005
7+ 80
4-15
•717
•018
+ '047
+ •0008
7+110
3-10
•683
•013
-•033
-•0004
7+150
2'55
•698
•on
+ •002
•oooo
•697
In the second set the quantity of gas had been adjusted to give a suitable
pressure with v = 0. It is from it and from Table I. that the data were
obtained for the calculation of V already given.
These tables give a fairly complete account of the behaviour of nitrogen
from a pressure of about 1'5 millims. down to '01 millim. of mercury. In
each set the range of pressure is nearly in the ratio of 3 : 1, and overlaps the
range of the preceding and following sets. An examination of the fifth
1901] OF GASES BETWEEN 1'5 AND O'Ol MILLIMETRES OB^ MERCURY. 527
column shows no indication of departure from Boyle's law. The sixth
column allows a judgment to be formed of the degree of accuracy to which
the law is verified. It gives the amount by which p exceeds the value
necessary in order that pv should be absolutely constant, expressed in milli-
metres of mercury. The errors thus exhibited include not only those arising
in the setting of the manometer and the reading of the telescope, but also
those entailed in the measurements of volume, and in consequence of fluctua-
tions of temperature. The latter source of error is of course more important
at the higher pressures. It will be seen that the accuracy attained is very
remarkable. Even at the higher pressures the mean error is only about
•001 millim., while at the lower pressures of Tables III. — VII. the mean error
is less than '0004 millim. And it must be remembered that the numbers to
which these errors relate are the means of two observations only.
As a means of dealing with very small pressures, the sloping manometer
has proved itself in a high degree satisfactory, the performance being some
twenty-five times better than Amagat's standard. It could hardly have been
expected that the mean error would prove to be less than one wave-length of
yellow light*. Considered as a pressure, the mean error corresponds to the
change of barometric pressure accompanying an elevation of 4 millims.
On hydrogen more than one series of observations have been carried
out. The specimen that will be given is not in some respects the most
satisfactory, but it is chosen as having been pursued to the greatest rare-
factions. The gas was dried carefully with phosphoric anhydride and was
introduced into the apparatus as already described. It is thought sufficient
to record only numbers corresponding to the three last columns of Tables
I. — VII., the first column giving the pressure in millims. of mercury, the
second the deviation of pv from the mean value of the set taken as unity,
the third the error in p from what would be required to make pv absolutely
constant.
In several of the sets of observations recorded in Table VIII., there would
seem to be a tendency for the positive errors to concentrate towards the
beginning, i.e., for pv to diminish slightly with p. It was at this stage that
a suspicion arose that the distance between the glass points of the mano-
meter might not be quite constant, but, as has been related, the suspicion
was not verified. It is just possible that at the higher pressures and smaller
* I had at one time contemplated an apparatus from which a further ten-fold increase in
accuracy might be expected. Two beams of light, reflected nearly perpendicularly from the
mercury surfaces, would be brought to interference by an arrangement similar to that used
in investigating the refractivity of gases (Proc. Roy. Soc. Vol. LIX. p. 200, 1896 [Vol. iv. p. 218] ;
Vol. LXIV. p. 97, 1898 [Vol. iv. p. 364]). Preliminary trials proved that the method is feasible ;
but the delicacy is excessive in view of the fact that according to Hertz the pressure of mercury
vapour at common temperatures itself amounts to '001 millim.
528 ON A NEW MANOMETER, AND ON THE LAW OF THE PRESSURE [266
TABLE VIII.— Hydrogen.
October— November, 1900.
Pressure in
millims. Hg
Deviation
of pv
Error of p
in millims.
Pressure in
millims. Hg
Deviation
of pv
Error of p
in millims.
1-43
+ •0025
+ 0036
1-44
+ O018
+ •0026
1-31
+ •0030
+ •0039
1-18
-•0005
-•0006
1-20
+ •0002
+ •0002
i-oo
+ •0009
+ •0009
1-11
-0012
-•0013
•87
+ •0007
+ •0006
•97
-•0005
-•0005
•77
+ •0005
+ •0004
•86
-•0002
-•0002
•62
•0000
•oooo
•77
-•0016
-•0012
•57
-•0028
-•0016
•70
-•0025
-•0017
•52
-•0009
-•0005
•64
+ 0007
+ •0004
•48
-•0018
-•0009
—
—
-
•42
+ •0018
+ •0008
•769
+ •0021
+ •0016
•386
•0000
•oooo
•624
+ •0028
+ •0017
•315
+ •0044
+ '0014
•524
•0000
•0000
•264
+ •0023
+ •0006
•423
+ •0002
+ •0001
•213
+ •0014
+ •0003
•335
-•0037
-•0012
•168
-•0072
-•0012
•279
-•0018
-•0005
•140
-•0014
-•0002
•196
+ •0079
+ •0015
•098
-•009
-•0009
•158
+ •0046
+ •0007
•080
•000
•oooo
•133
+ •0053
+ •0007
•068
+ •005
+ •0003
•106
-•0053
-•0006
•055
+ •007
+ •0004
•085
-•0037
-•0003
•044
+ •007
+ O003
•070
-•0083
-•0006
•036
-•005
-•0002
•051
+ •007
+ •0004
•027
-•047
-•0013
•041
+ •002
+ •0001
•023
+ •016
+ •0004
•034
-•009
-•0003
•018
-•054
-•0010
•027
-•023
-•0006
•016
+ •021
+ •0003
•023
+ •036
+ •0009
•013
+ •040
+ •0005
•018
-•014
-•0003
•010
+ •019
+ •0002
volumes the temperature changes were not insensible. It is probable that
they would operate in the direction mentioned, inasmuch as at the smaller
volumes a larger proportion of the gas would be in the connecting tubes
1901] OF GASES BETWEEN 1'5 AND O'Ol MILLIMETRES OF MERCURY. 529
at a higher level in the room, and therefore warmer. Considerable pre-
caution was taken, and I was not able to satisfy myself that disturbance due
to temperature really existed. In another series of observations on hydrogen
the tendency was scarcely apparent, and it remains doubtful whether there
is any real indication of departure from Boyle's law. It may be noted that
interest was concentrated rather upon the lower pressures, and that perhaps
less pains were taken over the readings of the higher pressures, where in
any case the error would be a smaller proportion of the whole. Also some
of the observations were not repeated. Another point that may be noted
is that the means are chosen with respect to the values of pv, and that a
different choice would in many cases materially reduce the mean error in
the last column.
Having thoroughly tested the apparatus and the method of experimenting
with hydrogen and nitrogen, I returned with curiosity to the case of oxygen.
Special pains were taken to ensure that the gas should be pure and above
all dry. To this end glass tubes were prepared containing permanganate of
potash and phosphoric anhydride, and these were connected by sealing to
one of the branches of the three-way tap. A high vacuum having been made
throughout, heat was gradually applied, and some of the oxygen allowed to
blow off. The phosphoric tube (of considerable capacity) was then allowed
to stand full of gas for some little time, after which the necessary gas to a
pressure of about 10 centims. was allowed to enter the apparatus by means
of the three-way tap. With regard to the maintenance of the purity of the
gas under rarefaction, it may be remarked that the method of experimenting
was favourable, inasmuch as the last stages were not reached until the
apparatus had been exposed to the gas under trial for a week or two. Any
contamination that might be communicated from the glass during the first
few days would for the most part be removed before the final stages were
reached.
Before the regular series was commenced, special observations extending
over several days were made in the region of pressure (from 1 millim. to
•5 millim.) where Bohr found anomalies. No unsteadiness could be detected.
Whatever reading was obtained within a few minutes of a change of pressure
was confirmed after an interval of an hour or more. For example, on
November 29, at 12h 25m the pressure which had stood for some time at '80
millim. was lowered to '65 millim. At 8h Om the pressure was unaltered. In
no case was the behaviour in any way different to that which had been
observed with the other gases. It is true that when the observations were
reduced one preliminary set showed an excess of pressure at the smaller
volumes similar to that recorded in the case of hydrogen, but the tendency is
scarcely visible in the regular series now to be given, which extended from
November 27 to December 9.
B. iv. 34
530 ON A NEW MANOMETER, AND ON THE LAW OF THE PRESSURE
An examination of the numbers in the Table IX. shows that Boyle's law
was observed, practically up to the limits of the accuracy of the measure-
ments, and in particular that there was no such falling off in the value of pv
at low pressures as was encountered by Bohr. What can be the cause of the
difference of our experiences I am at a loss to conjecture. I can only suppose
that it must be connected somehow with the quality of the gas, complicated
perhaps by interaction with the glass or with the mercury.
TABLE IX.— Oxygen.
Pressure in
millims. Hg
Deviation
of pv
Error of p
in millims.
Pressure in
millims. Hg
Deviation
ofpv
Error of p
in millims.
1-53
+ •0016
+ •0024
•580
-•0035
-•0020
1-17
-•0012
-•0014
•472
+ •0005
+ •0002
•95
+ •0005
+ •0005
•396
-•0007
-•0003
•80
+ •0007
+ •0006
•321
+ •0016
+ •0005
•65
+ •0012
+ •0008
•255
+ •0012
+ •0003
•57
-•0009
-•0005
•212
+ •0016
+ •0003
•51
-•0014
-•0007
—
—
—
•47
-•0014
-•0007
—
—
—
•43
+ •0009
+ •0004
—
-
—
•288
+ •002
+ •0007
•142
+ •005
+ •0007
•233
•ooo
•0000
•115
+ •009
+ •0011
•196
•ooo
•oooo
•094
-•019
-•0018
•159
+ •005
+ •0008
•077
•ooo
•oooo
•125
-•002
-•0003
•062
+ •012
+ •0007
•103
-•009
-•0010
•051
-•012
-•0006
•068
-•002
-•0002
•034
•ooo
•oooo
•056
+ 005
+ •0003
•029
+ •059
+ •0017
•048
+ •019
+ •0009
•022
-•042
-0009
•038
+ •009
+ •0004
•019
+ •023
+ •0004
•029
-•019
-•0005
•014
-•035
-•0005
•025
-'009
-•0002
—
—
—
The final result of the observations on the three gases may be said to
be the full confirmation of Boyle's law between pressures of 1*5 millims. and
•01 millim. of mercury. If there is any doubt, it relates to the case of
hydrogen, which appears to press somewhat in excess at the highest
1901] OF GASES BETWEEN 1'5 AND O'Ol MILLIMETRES OF MERCURY. 531
pressures. But when we consider the smallness of the amount and the
various complications to which it may be due, as well as a priori probabilities,
we may well hesitate to accept the departure from Boyle's law as having a
real existence.
So far as the present results can settle the question, they justify to the
full the ordinary use of McLeod's gauge within the limits of pressure
mentioned and for nitrogen and hydrogen gases. The same might be said for
oxygen; but until the discrepancy with the conclusions of Bohr can be
explained, the necessity for some reserves must be admitted.
In any case the new manometer has done its work successfully, and is
proved to be capable of measuring small pressures to about -%^ of a milli-
metre of mercury. It was constructed under my direction by Mr Gordon.
34—2
267.
ON A PROBLEM RELATING TO THE PROPAGATION OF
SOUND BETWEEN PARALLEL WALLS.
[Philosophical Magazine, I. pp. 301—311, 1901.]
THE influence of viscosity and heat conduction in modifying the propaga-
tion of sound in circular tubes of moderate dimensions has been treated by
Kirchhoff * in his usual masterly style, but he passes over the case when the
diameter is very large. In my book on the Theory of Sound, 2nd edition,
§ 348, I have given a full statement of Kirchhoff's theory, and have indicated
the alterations required when the boundary is supposed to take the form of
two parallel planes instead of a cylindrical surface. In any case the action
of the wall is supposed to be such as to annihilate variation of temperature,
and tangential as well as normal motion. In connexion with the problem
of the propagation of sound over water I recently had occasion to extend
the analysis to the case of a layer of very great thickness; and though, as
the result showed, the solution fails to answer the question which I had
then in view, it is of some interest in itself. In this case the practical
question differs somewhat from that proposed by KirchhofF, who assumes
not only complete periodicity with respect to time, but also a quasi-periodicity
with respect to x, the direction of propagation, all the functions being sup-
posed proportional to emx, where m is a complex constant, and not otherwise
to depend upon x. This assumption is retained in the present paper. It
seems advisable to give a brief recapitulation of Kirchhoff's theory, referring
for more detailed explanation to the original paper or to the account of it
in Theory of Sound.
* Pogg. Ann. Vol. cxxxiv. 1868; Collected Memoirs, p. 540.
1901] ON THE PROPAGATION OF SOUND BETWEEN PARALLEL WALLS. 533
The condition of the gas at any point x, y, z being defined by the
component velocities u, v, w, and 6', where 6' is proportional to the excess
of temperature, the equation for 0' is found to be
KB' -{a? + h O' + p" + v)} V20' + j- {b2 + h (// + /')] V40' = 0. . . .(1)
In this equation V2 stands for d2fdx2 + d*/dy* + d*ldz* ; h is such that all the
variables (u, v, w, 9') are proportional to e**; a is the velocity of sound as
reckoned on Laplacean principles, 6 the corresponding Newtonian value ;
p, p", v are coefficients of viscosity and of heat conduction.
A solution of (1) may be obtained in the form
where Q1} Q2 are functions of x, y, z satisfying respectively
Xj, X2 being the roots of
Aa - (a2 + h (p! + p" + v)} X + \ {62 + h (p! + //')} X2 = 0 ; (4)
fl
while Alt Az denote arbitrary constants.
In correspondence with this value of 0', particular solutions are obtained
by equating u, v, w to the differential coefficients of
AQi + B*Q*,
taken with respect to x, y, z. The relation of the constants Blt B2 to Alf A.2 is
h
More general solutions may be obtained by addition to u, v, w respectively
of u', v', w', where u', v', w' satisfy
W = 4w', W = A«', VW = ^w' .......... (6)
fJL H p.
Thus
u = u + B.dQ.jdx + B,dQ,ldx,
v= v' + B1dQl/dy + B2dQ,/dy, , .................. (7)
w = w + B.dQ./dz + B2dQ,/dz,
where Blt B2 have the values given above.
It appears that
534 ON A PROBLEM RELATING TO THE PROPAGATION [267
These results were applied by Kirchhoff to the case of plane waves,
supposed to be propagated in infinite space in the direction of +x, and
it may conduce to clearness to deal first with this case. Here v' and w
vanish, while u', Q1} Q2 are independent of y and z. It follows from (8) that
u' also vanishes. The equations for Ql and Q.2 are
\Q1> ffiQt/da? = \&; .................. (9)
so that we may take
where the signs of the square roots are to be so chosen that the real parts
are positive. Accordingly
(11)
(12)
in which the constants A^, A2 may be regarded as determined by the values
of u and 6' when x = 0.
The solution, as expressed by (11), (12), is too general for our purpose,
providing as it does for arbitrary communication of heat at x = 0. From
the quadratic (4) in X we see that if //, p", v be regarded as small quantities,
one of the values of A,, say \, is approximately equal to h2/a2, while the other
(\a) is very great. The solution which we now require is that corresponding
to \i simply. The second approximation to \ is by (4)
that
If we now write in for h, we see that the typical solution is
........................... (14)
where
In (14) an arbitrary multiplier and an arbitrary addition to t may, as
usual, be introduced ; and, if desired, the solution may be realized by omitting
the imaginary part.
In passing on to consider the influence of walls, by which gas is confined,
upon the propagation of sound, it is here proposed to take the case of two
dimensions, rather than the tube of circular section treated by Kirchhoff.
1901] OF SOUND BETWEEN PARALLEL WALLS. 535
The analysis, however, is nearly the same. We suppose that sound is propa-
gated in the layer of gas bounded by fixed walls at y = ± y1 , so that w = 0,
while u, v, 0' are functions of x and y only. The like may be assumed
respecting u', v', Ql} Qz. We suppose further that as functions of x these
quantities are proportional to einx, where m is a complex constant to be
determined. The equations (3) for Q1} Q2 become
(\l-m*)Ql, d«Qa/djf = (*,-!»«)&. ...(16,17)
For u'. v' equations (6), (8) give
*wj/ 4- —0 (~\R 1Q 9fh
— > — 7/& I ct . ~^; — ~ — \ / — '/f i f j //e-u- ~t* 7^" — v/. • * » •»*i -LO, X t/. ^W I
fi I dyz \p J ' dy
These three equations are satisfied if u' be determined by means of the
first, and v' is chosen so that
v^-j-^-- ~, (21)
h/fju - m? dy
a relation obtained by subtracting from (19) the result of differentiating (20)
with respect to y. The solution of (18) may be written u' = AQ, in which
A is a, constant, and Q a function of y satisfying
Thus, by (5), (7),
AH (k
dy
Qi — A%m
f±-pk
•••v-/
...(23)
m
dQ
A (k
U. ) *
\ dQl , (h \ e
^2 ("^
h/fj, — m?
dy
6'
Ms
J dy \\2 /
A,Q2.
dy ' -(24)
...(25)
On the walls at y = ± yl} u, v, 9' must satisfy certain conditions. It will
here be supposed that there is neither motion of the gas nor change of
temperature ; so that when y = ±yi,u,v, 0' vanish. The condition of which
we are in search is thus expressed by the evanescence of the determinant
of (23), (24), (25), viz. :
which is to be satisfied when y— ± yv.
Since u is an even function of y, we have from (3), (22),
Q =c
536 ON A PROBLEM RELATING TO THE PROPAGATION [267
From (23), (25), and from the fact that w = 0 when y = yi, we get as
the general value of u, without regard to the constant multiplier,
M=s'5(£) + A/x1-A/x1 of^O'Ajx^A/x, sferr -(28)
In equation (26) the values of \1} X^ are independent of ylt being deter-
mined by (4). In the application to air under normal conditions /u/, p", v
may be regarded as small, and we have approximately
X1==A2/a2, X-2 = ha2/vb? (29)
A second approximation to the value of X^ is given in (13). It is here
assumed that the velocity of propagation of viscous effects of the pitch in
question, viz. V(2/A'w), is small in comparison with that of sound, so that
inp'/a?, or hp'/a?, is a small quantity — a condition abundantly satisfied in
practice.
In interpreting the solution we limit ourselves here to the case which
arises when //, ft", v are treated as very small — so small that the layer of gas
immediately affected by the walls is but an insignificant fraction of the whole.
When // &c. vanish, we have
Xj == h?ja?, m2 = A2/a2,
so that unless y be great y^/(iftii — \l) is small. On the other hand,
yiV/(m2 — A //A'), 2h v' (m2 — X,) are large. For the moment we leave the value
of y*/(m? - \) open, and merely introduce the simplifications arising out
of the largeness of the arguments in Q and Q2.
If z be a complex quantity of the form £ + iy, we have in general
cos z = cos £ cosh r\ — i sin £ sinh ?;, (30)
_ sin 2f + i sinh 2 17
~ cos 2f+ cosh 2*, '
Thus, when i) is large,
d log cos z
— ^ = - tan z = — i ;
dz
so that when y = yi, since A is a pure imaginary,
d\ogQ_ //A\ dlogQ,_
The introduction into (26) of these values and those of Xj and X, from
(29), gives
dlogQ1= y'tf
dy a? '
where
y^vy + ^/fc-WV*; ..................... (33)
1901] OF SOUND BETWEEN PARALLEL WALLS. 537
or, if z = yl ^(m? — Xj),
ztanz^y'tf-y./a- ............................ (34)
This is the equation by which z, and thence m2, is to be determined.
In the case corresponding to that treated by Kirchhoff, yl is not so large
but that the right-hand member of (34) is a small quantity. The solution
of (34) is then
*• = •/%!/«•; .............................. (35)
whence
(36)
We now write k = ni, so that the frequency is 71/2-Tr. Thus
V/*=V(i~).(l+») ........................... (37)
and
m — ± (m' + im"), ........................... (38)
where by (36)
The solution differs from that found by Kirchh6ff for a circular tube
of radius r merely by the substitution of ^yl for r*.
So far, then, as it depends on t and x, the typical solution is
aint g—m'x—im"x
or when realized,
e~m'x cos (nt - m"x), ........................ (40)
where m, m" have the values given in (39). This is for waves travelling
in the positive direction.
As a function of y, u is given by (28); but this may now be simplified
in virtue of the supposition that the layer directly influenced by the viscosity
is but a small fraction of ylt By (27)
cos (y^n/f . V^t) = cos {yx Vn/V . (1 - »)}
irin(*V&)\ (41)
use being made of (30), in which 17 is large. In consequence of (41),
Q(y) + Q(yi) vanishes unless y be nearly equal to ylt viz. unless the point
considered be within the frictional layer. In like manner, and under the
same restriction, Q2 (y) -r- Q.z (y^ may be neglected. Except in the immediate
neighbourhood of the walls, (28) now reduces to
Theory of Sound, 2nd edit. § 350.
538 ON A PROBLEM RELATING TO THE PROPAGATION [267
In the case considered by Kirchhoff, where the argument of Ql is small,
we have from (27) approximately
&(</) = & (2A) =1,
and accordingly u = — 1. To this approximation the velocity is uniform
across the whole section until it begins to fall off as the walls are closely
approached.
As a first step towards the consideration of what occurs when yl is great,
we may proceed to a second approximation. Thus, from (27),
-^)=i-^2; ............... (43)
*/]-
so that
yI
This equation expresses the dependence of u upon y. The dependence
on t and a; is given by the factors already considered, viz.
That (44) is complex indicates that the phase varies with y. The realized
expression will be
from which we infer (i) that the intensity is least in the middle where
y = 0, and increases towards the walls until the frictional layer is approached;
(ii) that, as y2 increases from the centre, constancy of phase demands a
diminishing x, or, in other words, that the wave-surface is convex towards
+ x and the wave divergent.
We have now to trace the solution of (34) when the right-hand member
becomes large. Writing it in the form
(46)
we have to find such a complex value of z, say £ + irj, that the function
on the left is real. Initially, when yl is small,
% + i<n=p (cos 0 + i sin 0) = p (cos 67£° + i sin 67£°),
and
1901] OF SOUND BETWEEN PARALLEL WALLS. 539
If \ve retain the angle 67£° and increase p, we find, calculating by means
of (31), that i~**. ten* becomes complex with imaginary part positive.
Thus if p = 1, we get
i~*z. tan z = "80 (cos 9° 54' + i sin 9° 54').
This is a sign that 6 must be reduced. If we take p = 1, 0 = 60°, we find
i-*z . tan z = '83 (cos 2° 7' - i sin 2° 7').
If p = 1-5, while 0 = 60°, we get in detail
z = £ + irj = -75 + i x 1-299 ;
sin 2£ = sin 85° 57' = '998, sinh fy = 6'695,
cos 2£ = cos 85° 57' = "071, cosh ty = 6'769,
whence
so that
i-*s . tan $ = 1-5 (cos 6° 31' + t sin 6° 31').
The course of the calculation makes it clear that as z increases, tan z
approaches the limit i, so that ultimately B = ^TT, or the angle reduces from
67£° to 45°. Hence
ft-tVnfy*/*1' .............................. (47)
and
rn? = h*la?- indict, ........................ (48)
independent of y^
In order to obtain u as a function of y, we have now to interpret (42)
for the case where yl is great. It may be written
where e, given by (47), is
„ — , ...(49)
COS 3
By (30), since 17 is large, we have approximately
cos z = ^(cos f + i sin £), (51)
where
Accordingly, cos 2 is large. If, as near the middle of the layer of gas, y be
not large, cos (2y/yl) = 1, and
u-- I/cos z, (53)
a small quantity. When y is so large that 2y/ya is large, as well as z, we
may write
M = -e*'-1>.ei<*'-*>, (54)
where
540 ON A PROBLEM RELATING TO THE PROPAGATION [267
As the walls are approached u rapidly increases, and at last e^'-v becomes
nearly equal to unity. We must bear in mind, however, that (42) and there-
fore (54) must not be applied within the frictional layer lying quite close
to the walls, so that we are not at liberty to suppose y actually equal to yt.
Under normal conditions the thickness of the frictional layer is very small.
If in C.G.s. measure we take /*' = '16, w = 2?r x 256, we find V(™/ V) = 67.
Thus if we suppose the thickness of the frictional layer in (41) to be
defined by
we get
y\ — y = '15 millim.
If the point under consideration be a few multiples of this (say 1 millim.)
from the walls, the ratio Q (y) -r- Q (y^) may be neglected.
The thickness of the layer through which Q2 (y} + Q2 (y^ is sensible
is of the same order of magnitude.
Let us next consider what value of (yl — y} makes (?/ — 77) in (54) equal
to unity. By (52), (55)
If we take p ='16, v = '256, we find from (33) 7' ='6; and a =33200; so
that for a frequency of 256 we get
V 2. 33200 2
For air and for a sound of this pitch the falling off becomes important
at a distance of about 400 metres from the walls.
As has already been suggested, this solution fails to answer the practical
question for the sake of which it was originally attempted. It was desired
to know whether in the propagation of sound for long distances over smooth
water, there was any important shadow formed near the surface under the
influence of viscosity and heat conduction. It would apparently be a matter
of some difficulty to formulate and solve a definite problem in which this
question is involved. But, as Lord Kelvin has pointed out to me, a sufficient
answer to the practical question may be arrived at by very simple reasoning
on the basis of a solution originally given by Stokes (see Theory of Sound,
§ 347). If U be the tangential velocity of a plane vibrating rigidly in an
atmosphere of viscous fluid with a frequency H/^TT, the work required to
maintain the motion is, for unit of area,
1901] OF SOUND BETWEEN PARALLEL WALLS. 541
where Um denotes the maximum value of U during the period. The same
expression may be applied to find the work lost by the presence of a
fixed plane in air vibrating with velocity U. The energy of this motion
is, per unit of volume, $pUm*, or for a stratum of height y resting upon unit
of area,
If we equate the two expressions we get a superior limit to the thickness
of the stratum whose energy could be absorbed in time t. We find thus
yrsV \*P
or, if we take n = ZTT x 256, and as for air p/p = p = '16, T/ = lit.
Thus in 9 seconds the thickness of a stratum of shadow could not reach
1 metre, and must, in fact, be very much less. It would appear therefore
that this effect may be neglected in practice, unless it be in the case of
an observer extremely close to the water.
268.
POLISH.
[Proceedings of the Royal Institution, xvi. pp. 563—570, 1901 ;
Nature, LXIV. pp. 385—388, 1901.]
THE lecture commenced with a description of a home-made spectroscope
of considerable power. The lens, a plano-convex of 6 inches aperture and
22 feet focus, received the rays from the slit, and finally returned them to
a pure spectrum formed in the neighbourhood. The skeleton of the prism
was of lead; the faces, inclined at 70°, were of thick plate-glass cemented
with glue and treacle. It was charged with bisulphide of carbon, of which
the free surface (of small area) was raised above the operative part of the
fluid. The prism was traversed twice, and the effective thickness was
5£ inches, so that the resolving power corresponded to 11 inches, or 28 cm.,
of CS2. The liquid was stirred by a perforated triangular plate, nearly
fitting the prism, which could be actuated by means of a thread within reach
of the observer. The reflector was a flat, chemically silvered in front.
So far as eye observations were concerned, the performance was satis-
factory, falling but little short of theoretical perfection. The stirrer needed
to be in almost constant operation, the definition usually beginning to fail
within about 20 seconds after stopping the stirrer. But although the stirrer
was quite successful in maintaining uniformity of temperature as regards
space, i.e. throughout the dispersing fluid, the temperature was usually some-
what rapidly variable with time, so that photographs, requiring more than
a few seconds of exposure, showed inferiority. In this respect a grating is
more manageable.
The lens and the faces of the prism were ground and polished (in 1893)
upon a machine kindly presented by Dr Common. The flat surfaces were
tested with a spherometer, in which a movement of the central screw through
! 60*000 inch could usually be detected by the touch. The external surfaces
1901] POLISH. 543
of the prism faces were the only ones requiring accurate flatness. In polish-
ing, the operation was not carried as far as would be expected of a professional
optician. A few residual pittings, although they spoil the appearance of
a surface, do not interfere with its performance, at least for many purposes.
In the process of grinding together two glass surfaces, the particles of
emery, even the finest, appear to act by pitting the glasses, i. e. by breaking
out small fragments. In order to save time and loss of accuracy in the
polishing, it is desirable to carry the grinding process as far as possible, using
towards the close only the finest emery. The limit in this direction appears
to depend upon the tendency of the glasses (6 inches diameter) to seize, when
they approach too closely, but with a little care it is easy to attain such
a fineness that a candle is seen reflected at an angle of incidence not exceed-
ing 60°, measured as usual from the perpendicular.
The fineness necessary, in order that a surface may reflect and refract
regularly without diffusion, viz. in order that it may appear polished, depends
upon the wave-length of the light and upon the angle of incidence. At
a grazing incidence all surfaces behave as if polished, and a surface which
reflects red light pretty well may fail signally when tested with blue light at
the same angle. If we consider incidences not too far removed from the per-
pendicular, the theory of gratings teaches that a regularly corrugated surface
behaves as if absolutely plane, provided that the wave-length of the corruga-
tions is less than the wave-length of the light, and this without regard to the
depth of the corrugations. Experimental illustrations, drawn from the sister
science of Acoustics, were given. The source was a bird-call from which
issued vibrations having a wave-length of about 1'5 cm., and the percipient
was a high-pressure sensitive flame. When the bird-call was turned away,
the flame was silent, but it roared vigorously when the vibrations were re-
flected back upon it from a plate of glass. A second plate, upon which small
pebbles had been glued so as to constitute an ideally rough surface, acted
nearly as well, and so did a piece of tin plate suitably corrugated. In all
these cases the reflection was regular, the flame becoming quiet when the
plates were turned out of adjustment through a very small angle. In another
method of experimenting the incidence was absolutely perpendicular, the
flame being exposed to both the incident and the reflected waves. It is
known that under these circumstances the flame remains quiescent at the
nodes and flares most vigorously at the loops. As the reflector is drawn
slowly back, the flame passes alternately through the nodes and loops, thus
executing a cycle of changes as the reflector moves through half a wave-
length. The effects observed were just the same whether the reflector were
smooth or covered with pebbles, or whether the corrugated tin plate were
substituted. All surfaces were smooth enough in relation to the wave-length
of the vibration to give substantially a specular reflection.
544 POLISH. [268
Finely-ground surfaces are still too coarse for perpendicular specular re-
flection of the longest visible waves of light. Here the material may be
metal, or glass silvered chemically on the face subsequently to the grinding.
But experiment is not limited by the capabilities of the eye ; and it seems
certain that a finely ground surface would be smooth enough to reflect with-
out sensible diffusion the longest waves, such as those found by Rubens to be
nearly 100 times longer than the waves of red light. An experiment may be
tried with radiation from a Leslie cube containing hot water, or from a
Welsbach mantle (without a chimney). In the lecture the latter was em-
ployed, and it fell first at an angle of about 45° upon a finely ground flat
glass silvered in front. By this preliminary reflection, the radiation was
purified from waves other than those of considerable wave-length. The
second reflection (also at 45°) was alternately from polished and finely ground
silvered surfaces of the same size, so mounted as to permit the accurate sub-
stitution of the one for the other. The heating-power of the radiation thus
twice reflected was tested with a thermopile in the usual manner. Repeated
comparisons proved that the reflection from the ground surface was about
•76 of that from the polished surface, showing that the ground surface re-
flected the waves falling upon it with comparatively little diffusion. A slight
rotation of any of the surfaces from their proper positions at once cut off the
effect. It is probable that the device of submitting radiation to preliminary
reflections from one or more merely ground surfaces might be found useful in
experiments upon the longest waves.
In view of these phenomena we recognise that it is something of an
accident that polishing processes, as distinct from grinding, are needed at all ;
and we may be tempted to infer that there is no essential difference between
the operations. This appears to have been the opinion of Herschel*, whom
we may regard as one of the first authorities on such a subject. But,
although, perhaps, no sure conclusion can be demonstrated, the balance of
evidence appears to point in the opposite direction. It is true that the same
powders may be employed in both cases. In one experiment a glass surface
was polished with the same emery as had been used effectively a little earlier
in the grinding. The difference is in the character of the backing. In
* Enc. Met., Art. Light, p. 447, 1830: "The intensity and regularity of reflection at the
external surface of a medium is found to depend not merely on the nature of the medium, but
very essentially on the degree of smoothness and polish of its surface. But it may reasonably be
asked, how any regular reflection can take place on a surface polished by art, when we recollect
that the process of polishing is, in fact, nothing more than grinding down large asperities into
smaller ones by the use of hard gritty powders, which, whatever degree of mechanical comminu-
tion we may give them, are yet vast masses, in comparison with the ultimate molecules of matter,
and their action can only be considered as an irregular tearing up by the roots of every projection
that may occur in the surface. So that, in fact, a surface artificially polished must bear some-
what of the same kind of relation to the surface of a liquid, or a crystal, that a ploughed field does
to the most delicately polished mirror, the work of human hands."
Fig. 1.
Fig. 2.
1901] POLISH. 545
grinding, the emery is backed by a hard surface, e.g. of glass, while during
the polishing the powder (mostly rouge in these experiments) is imbedded in
a comparatively yielding substance, such as pitch. Under these conditions,
which preclude more than a moderate pressure, it seems probable that no
pits are formed by the breaking out of fragments, but that the material is
worn away (at first, of course, on the eminences) almost molecularly.
The progress of the operation is easily watched with a microscope, pro-
vided, say, with a £-inch object-glass. The first few minutes suffice to effect
a very visible change. Under the microscope it is seen that little facets,
parallel to the general plane of the surface, have been formed on all the more
prominent eminences*. The facets, although at this stage but a very small
fraction of the whole area, are adequate to give a sensible specular reflection,
even at perpendicular incidence. On one occasion five minutes' polishing of
a rather finely ground glass surface was enough to qualify it for the formation
of interference bands, when brought into juxtaposition with another polished
surface, the light being either white or from a soda flame ; so that in this
way an optical test can be applied almost before the polishing has begun f.
As the polishing proceeds, the facets are seen under the microscope to
increase both in number and in size, until they occupy much the larger part
of the area. Somewhat later the parts as yet untouched by the polisher
appear as pits, or spots, upon a surface otherwise invisible. Fig. 1 represents
a photograph of a surface at this stage taken with the microscope. The
completion of the process consists in rubbing away the whole surface down
to the level of the deepest pits. The last part of the operation, while it
occupies a great deal of time, and entails further risk of losing the " truth "
of the surface, adds very little to the effective area, or to the intensity of the
light regularly reflected or refracted.
Perhaps the most important fact taught by the microscope is that the
polish of individual parts of the surface does not improve during the process.
As soon as they can be observed at all, the facets appear absolutely structure-
less. In its subsequent action the polishing tool, bearing only upon the parts
already polished, extends the boundary of these parts, but does not enhance
their quality. Of course, the mere fact that no structure can be perceived
does not of itself prove that pittings may not be taking place of a character
too fine to be shown by a particular microscope or by any possible microscope.
But so much discontinuity, as compared with the grinding action, has to be
admitted in any case, that one is inevitably led to the conclusion that in all
probability the operation is a molecular one, and that no coherent fragments
* The interpretation is facilitated by a thin coating of aniline dye which attaches itself
mainly to the hollows.
t With oblique incidence, as in Talbot's experiments (see Phil. Mag. xxvm. p. 191, 1889
[Vol. in. p. 308]), achromatic bands may be observed from a surface absolutely unpolished,
but this disposition would not be favourable for testing purposes.
R. iv. 35
546 POLISH. [268
containing a large number of molecules are broken out. If this be so, there
would be much less difference than Herschel thought between the surfaces
of a polished solid and of a liquid.
Several trials have been made to determine how much material is actually
removed during the polishing of glass. In one experiment a piece 6 inches
in diameter, very finely ground, was carefully weighed at intervals during the
process. Losses of '070, '032, '045, '026, '032 gms. were successively registered,
amounting in all to '205 gms. Taking the specific gravity of the glass as 3,
this corresponds to a thickness of 3'6 x 10~4 cm., or to about 6 wave-lengths
of mean light, and it expresses the distance between the original mean
surface and the final plane. But the polish of this glass, though sufficient
for most practical purposes, was by no means perfect. Probably the 6 wave-
lengths would have needed to be raised to 10 in order to satisfy a critical eye.
It may be interesting to note for comparison that, in the grinding, one charge
of emery, such as had remained suspended in water for seven or eight
minutes, removed a thickness of glass corresponding to 2 wave-lengths.
In other experiments the thickness removed in polishing was determined
optically. A very finely ground disc was mounted in the lathe and polished
locally in rings. Much care was needed to obtain the desired effect of a ring
showing a continuously increasing polish from the edges inwards. To this
end it was necessary to keep the polisher (a piece of wood covered with
resin and rouge) in constant motion, otherwise a number of narrow grooves
developed themselves.
The best ring was about half an inch wide. When brought into contact
with a polished flat and examined at perpendicular incidence with light from
a soda flame, the depression at its deepest part gave a displacement of three
bands, corresponding to a depth of 1£X. On a casual inspection this central
part appeared well polished, but examination under the microscope revealed
a fair number of small pits. Further working increased the maximum depth
to 2£\, when but very few pits remained. In this case, then, polish was
effected during a lowering of the mean surface through 2 or 3 wave-lengths,
but the grinding had been exceptionally fine.
It may be well to emphasize that the observations here recorded relate to
a hard substance. In the polishing of a soft substance, such as copper, it is
possible that material may be loosened from its original position without
becoming detached. In such a case pits may be actually filled in, by which
the operation would be much quickened. Nothing suggestive of this effect
has been observed in experiments upon glass.
Another method of operating upon glass is by means of hydrofluoric acid.
Contrary to what is generally supposed, this action is extremely regular, if
proper precautions are taken. The acid should be weak, say one part of
commercial acid to two hundred of water, and it should be kept in constant
1901] POLISH. 547
motion by a suitable rocking arrangement. The parts of the glass not in-
tended to be eaten into are, as usual, protected with wax. The effect upon
a polished flat surface is observed by the formation of Newton's rings with
soda-light. After perhaps three-quarters of an hour, the depression corre-
sponds to half a band, i.e. amounts to ^X, and it appears to be uniform over
the whole surface exposed. Two pieces of plate glass, 3 inches square, and
flat enough to come into fair contact all over, were painted with wax in
parallel stripes, and submitted to the acid for such a time, previously ascer-
tained, as would ensure an action upon the exposed parts of J X. After
removal of the wax, the two plates, crossed and pressed into contact so as to
develop the colours, say of the second order, exhibited a chess-board pattern.
Where two uncorroded, or where two corroded parts, are in contact, the
colours are nearly the same, but where a corroded and an uncorroded surface
overlap, a strongly contrasted colour is developed. The combination lends
itself to lantern projection, and the pattern upon the screen [shown] is very
beautiful, if proper precautions are taken to eliminate the white light reflected
from the first and fourth surfaces of the plates.
In illustration of the action of hydrofluoric acid, photographs* were
shown of interference bands as formed by soda-light between glass surfaces,
one optically flat and the other ordinary plate, upon which a drop of dilute
acid had been allowed to stand (Fig. 2). Truly plane surfaces • would give
bands straight, parallel, and equidistant.
Hydrofluoric acid has been employed with some success to correct ascer-
tained errors in optical surfaces. But while improvements in actual optical
performance have been effected, the general appearance of a surface so treated
is unprepossessing. The development of latent scratches has been described
on a former occasion f.
A second obvious application of hydrofluoric acid has hitherto been less
successful. If a suitable stopping could be found by which the deeper pits
could be protected from the action, corrosion by acid could be used in sub-
stitution for a large part of the usual process of polishing.
In connexion with experiments of this sort, trial was made of the action
of the acid, upon finely ground glass, such for example as is used as a backing
for stereoscopic transparencies, and very curious results were observed. For
this purpose the acid may conveniently be used much stronger, say one part
of commercial acid to 10 parts of water, and the action may be prolonged
for hours or days. The general appearance of the glass after treatment is
smoother and more translucent, but it is only under the microscope that the
remarkable changes which the surface has undergone become intelligible.
Fig. 3 is from a photograph taken in the microscope, the focus being upon
the originally ground surface itself. The whole area is seen to be divided
* The plates were sensitised in the laboratory with cyanine.
t Proc. Roy. Inst. March 1893. [Vol. iv. p. 59.]
35—2
548
POLISH.
[268
into cells. These cells increase as the action progresses, the smaller ones
being, as it were, eaten up by the bigger. The division lines between the
cells are ridges, raised above the general level, and when seen in good focus
appear absolutely sharp. The general surface within the cells shows no
structure, being as invisible as if highly polished.
That each cell is in fact a concave lens, forming a separate image of the
source of light, is shown by slightly screwing out the object-glass. Fig. 4
was taken in this way from the same surface, the source of light being the
flame of a paraffin lamp, in front of which was placed a cross cut from sheet-
metal.
The movement required to pass from the ridge to the image of the source,
equal to the focal length (/) of the lens, may be utilised to determine the
depth (t) of a cell. In one experiment the necessary movement was '005 inch.
The semi-aperture (y) of the " lens " was '0015 inch, whence by the formula
y* =ft, we find t = '00045 inch. This represents the depth of the cell, and it
amounts to about 8 wave-lengths of yellow light.
Fig. 5.
The action of the acid seems to be readily explained if we make the very
natural supposition that it eats in everywhere, at a fixed rate, normally to
the actual surface. If the amount of the normal corrosion after a proposed
time be known, the new surface can be constructed as the " envelope " of
spheres having the radius in question and centres distributed over the old
surface. Ultimately, the new surface becomes identified with a series of
spherical segments having their centres at the deeper pits of the original
surface. The construction is easily illustrated in the case of two dimensions.
In the figure A is supposed to be the original surface ; B, C, D, E surfaces
formed by corrosion, being constructed by circles having their centres on A.
In B the ridges are still somewhat rounded, but they become sharp in D
and E. The general tendency is to sharpen elevations and to smooth off
depressions.
Fig. 3.
Fig.
269.
DOES CHEMICAL TRANSFORMATION INFLUENCE WEIGHT?
[Nature, LXIV. p. 181, June, 1901.]
CAREFUL experiments by Heydweiller, published in the last number of
Drude's Annalen (Vol. v. p. 394), lead their author to the conclusion that in
certain cases chemical action is accompanied by a minute, but real, alteration
of weight. The chemical actions here involved must be regarded as very
mild ones, e.g. the mere dissolution of cupric sulphate in water, or the sub-
stitution of iron for copper in that salt.
The evidence for the reality of these changes, which amount to 0*2 or
0'3 mg., and are accordingly well within the powers of a good balance to
demonstrate, will need careful scrutiny; but it may not be premature to
consider what is involved in the acceptance of it. The first question which
arises is — does the mass change as well as the weight ? The affirmative
answer, although perhaps not absolutely inconsistent with any well ascer-
tained fact, will certainly be admitted with reluctance. The alternative —
that mass and weight are not always in proportion — involves the conclusion,
in contradiction to Newton, that the length of the seconds' pendulum at
a given place depends upon the material of which the bob is composed.
Newton's experiment was repeated by Bessel, who tried a number of metals,
including gold, silver, lead, iron, zinc, as well as marble and quartz, and whose
conclusion was that the length of the seconds' pendulum formed of these
materials did not vary by one part in 60,000. At the present day it might
be possible to improve even upon Bessel, or at any rate to include more
diverse substances in the comparisons ; but in any case the accuracy obtain-
able would fall much short of that realized in weighings.
As regards Heydweiller's experiments themselves, there is one suggestion
which I may make as to a possible source of error. Is the chemical action
sufficiently in abeyance at the time of the first weighing ? If there is copper
sulphate in one branch of an inverted U and water in the other, the equi-
librium can hardly be complete. The water all the time tends to distil over
into the salt, and any such distillation must be attended by thermal effects
which would interfere with the accuracy of the weighing.
[See further Nature, May 15, 1902.]
270.
ACOUSTICAL NOTES.— VI.
[Philosophical Magazine, n. pp. 280—285, 1901.]
Forced Vibrations.
IF free vibrations be represented by cos nt, and if the forced vibration
due to a force acting in a very long period be cospt, then the actual forced
vibration will be
n2 cos pt
ri1 -p2 '
It is here implied : —
(1) That in all cases the forced vibration takes its period from the force,
whatever may be the natural period.
(2) That if the forced vibration be the slower, viz. if p < n, the phase is
the same as if the vibration were infinitely slow, in which case the vibrator
would be situated at any instant of time in the position where the momentary
force would permanently maintain it.
(3) That if the forced vibration be the quicker (p > n), the phase of the
actual vibration is the opposite of that defined in (2).
(4) That if the force have nearly the period of the free vibrations, the
effect is much enhanced. Indeed, according to the formula it would become
infinite, which means that forces of a viscous character, never really absent,
must now be brought into the reckoning.
So far as I am aware, illustrations of this important theory* have usually
been wanting in lecture demonstrations, except as regards (4). I have found
that if we employ as vibrator a magnet with attached mirror, as used for
example in Thomson galvanometers, the whole may readily be brought before
a large audience.
* Young's Lectures on Natural Philosophy, p. 578 (1807).
1901] ACOUSTICAL NOTES. 551
With the aid of an external magnet, whose distance could be varied, the
frequency of (complete) vibration was adjusted to 10 per minute, the vibra-
tions being manifested by the motion of a spot of light reflected from the
mirror on to a scale in the usual manner. The force brought to bear upon
the vibrator had its origin in the revolution of a rather long permanent
magnet, situated at some little distance, and so mounted as to be capable of
rotation. No particular situation is necessary, but the action of the magnet
is simplest in certain special cases, as when its centre is at the level of the
suspended magnet and in the direction of the screen. The plane of revolu-
tion being horizontal, the deflecting action is then greatest when the revolving
magnet points towards the suspended magnet. In one of these positions,
say when the spot is deflected to the right, a bell rings automatically.
Uniform rotation at any desired speed is maintained by hand with the aid
of gearing, diminishing the speed in the ratio of 5:1, and of a metronome
set as required.
To illustrate propositions (1) and (2) the long magnet is caused to rotate
with a frequency of 8 per minute, i.e. with a frequency somewhat less than
that natural to the suspended system. At first the phenomenon is com-
plicated by the interaction of natural and forced vibrations ; but the former
soon die away. It is then recognised that the vibrations observed upon the
screen are isochronous with the revolution of the magnet, and that the bell
rings at the moment when the spot of light attains its greatest elongation
towards the right.
In the next experiment the speed of revolution is altered to 12 per
minute, so as to bring about the condition of things contemplated in (3).
After a little interval of settling down the bell rings always at the moment
when the spot is most deflected to the left, showing that the phase has been
altered by half a period.
To illustrate (4) the speed of revolution may now be adjusted to 8 per
minute. The arc of vibration is seen gradually to increase until it reaches
a large value, the bell now ringing, not at either extreme elongation, but as
the spot passes from left to right through its position of equilibrium.
Vibrations of Strings.
At the Royal Institution it is usual to illustrate this subject by ex-
periments after the method of Melde and Tyndall. The string is connected
with a large tuning-fork, whose prongs stand vertically, and the vibrations
are maintained electrically in the well-known manner. The electric contact
is between solids (of platinum), one attached to the prong, the other forming
the point of an adjustable screw carried by the framework.
552 ACOUSTICAL NOTES. [270
The string, 10 feet long, is stretched horizontally and the tension is
adjusted until a vigorous vibration ensues, which happens when one of the
modes of vibration has a period in simple relation to that of the fork. There
is here an important distinction according as the length of the string is
parallel or perpendicular to the motion of the point of attachment. In the
latter case the vibrations are of the character commonly classified as forced,
and the period is the same as that of the fork. But if the fork be so situated
that the motion of the point of attachment is along the length of the string,
the vibrations are of an entirely different character, and are executed in
a period the double of that of the fork. The theory of vibrations of this
class was discussed in a paper on Maintained Vibrations* published many
years ago, reference to which must here suffice.
A convenient device for demonstrating the relationship of periods is to
illuminate the string by sparks synchronous with the vibrations of the fork
itself. For this purpose an induction-coil is included in the circuit by which
the fork is driven, so that every break at the fork causes a spark between the
secondary terminals, to which a small jar is connected in the usual manner.
If then the vibrations of the string be isochronous with the fork, and there-
fore with the sparks, the intermittent illumination exhibits what is ordinarily
seen as a gauzy spindle resolved into the appearance corresponding to a single
phase of the vibration ; that is, the string is seen apparently fixed (in a dis-
placed position) and single. But if, as when the point of attachment moves
parallel to the length of the string, the vibrations are only half as fast as
those of the fork, the string is found in two (opposite) phases at the moments
of illumination, and is consequently seen double. The effect is improved by
a piece of ground glass, which may be held either between the sparks and
the string, or between the string and the eye. In the latter case it is a
shadow that is seen. It is desirable to retain enough continuous light to
allow the form of the gauzy spindle to remain visible. In this way the
difference between the two kinds of vibration may be exhibited to many
persons at once. [1902. The stroboscopic method of observation had already
been very similarly applied to this experiment by Costing, Onder houden
trillungen van gespannen draden, Helder, 1889.]
A detail of some importance relates to the use of the condenser, associated
as usual with the primary circuit of the coil. If its poles be connected
simply with the outer terminals of the fork-apparatus regarded as an in-
terrupter, the secondary sparks will be inferior or may fail altogether. The
explanation is to be sought in the self-induction of the magnet associated
with the fork, which apparently interferes with the suddenness of the break.
The poles of the condenser should be connected as directly as possible with
the two pieces of metal between which the break takes place. In the
* Phil. Mag. Vol. xv. p. 229 (1883) ; Scientific Papers, Vol. n. p. 188.
1901] ACOUSTICAL NOTES. 553
apparatus at the Royal Institution it makes all the difference on which side
of the small electromagnet the pole of the condenser is attached.
Beats of Sounds led to the Two Ears separately.
When two approximately pure tones, of equal intensity and of approxi-
mately equal frequency, are conveyed to one ear, beats are perceived according
to a well-known elementary theory, the frequency of the beats being the
difference of the frequencies of the tones. When the beats are somewhat
slow, the phase of silence is distinctly recognisable, and indeed the moment
of the occurrence of this phase is capable of being fixed with great accuracy.
The question whether the beats are still audible when one sound is led
to one ear alone, and the second sound to the second ear alone, is of great
importance. A careful experiment of this sort is described by Prof. S. P.
Thompson *, in which the sounds were conveyed to the ears by rubber tubes ;
and the conclusion was that in spite of all precautions the beats were most
distinctly heard, although there was no phase of " silence," such as is per-
ceived when both sounds are conveyed to the same ear.
I have lately tried a somewhat similar experiment, using telephones and
electrical conveyance, by which perhaps the risk of the sounds reaching the
wrong ears is reduced to a minimum. Two entirely independent, electrically
driven, forks of about 128 vibrations per second were the sources of sound.
Near the electromagnet of each fork was placed a small coil of wire in
connexion with a telephone. The higher harmonics were greatly moderated
by the interposition of thick sheets of copper ; but the sounds were doubtless
no more than rough approximations to pure tones. Both forks were placed
at a great distance from the observer ; and in one case the double connecting
wire was passed through a hole in a thick wall specially arranged many years
ago for this sort of experimenting. When the telephones were pressed closely
to the ears, the utmost possible was done to secure that each sound should
have access only to its proper ear.
The results depended somewhat upon the frequency of the beats. When
this exceeded one per second, the beats were very easily audible. When, on
the other hand, the frequency was reduced to \ or \ beat per second, the
beats were not easily perceived at first. After a little while the attention
seemed to concentrate itself upon the variable element in the aggregate
effect, and the cycle became clear. But even after some practice neither
Mr Gordon nor I could hear slow beats during the first 10 or 15 seconds
of observation.
The general results of the experiments do not appear to me to exclude
the view that the comparatively feeble beats heard under these conditions
* Phil. Mag. Vol. iv. p. 274 (1877).
554 ACOUSTICAL NOTES. [270
may be due to the passage of sound from one ear to the other through the
bones of the head or perhaps through the Eustachian tube.
Loudness of Double Sounds.
Observations upon the double syrens (with separate horns) used by the
Trinity House have given the impression that as heard from a distance the
two syrens are no better than one, even though the horns are parallel, and
the observer situated in the direction of the axis. Dr Tyndall's experience
was similar. In his Report of 1874 he remarks (June 2), " There was no
sensible difference of intensity between the single horn and the two horns " ;
and again (June 10), " Subsequent comparative experiments even proved
the sound of the two horns to be more effective than that of the three."
These conclusions are rather startling, suggesting the query as to what
then can be the use of multiplying pipes in an organ or voices in a chorus.
In order to clear the ground a little, I have recently tried some small-scale
experiments with organ-pipes.
Two stopped pipes of pitch about 256 were mounted near the window
of a room on the ground-floor. When the window was open the sounds could
be heard (over grass) to about 200 metres ; but when the window was closed
the range was much less. Some difficulty was experienced in getting equal
effects from the two pipes. According to the instructions of the observer,
one or other supply-pipe was more or less throttled with wax.
With approximate equality of intensities and with such tuning that the
beats were at the rate of about two per second, the results were very distinct.
The beats were much more easily audible than either of the component
sounds. Doubtless part of the advantage was due to the contrast provided
by the silences ; but it was thought that, apart from this, the swell of the
beat was distinctly louder than either sound alone.
The result of the experiment is, of course, just what was to be expected
from a mechanical point of view. According to theory the intensity (reckoned
according to energy propagated) at the loudest part of the beat should be
four times that of the (equal) component sounds heard separately.
In another set of experiments the pipes were mistuned until the interval
was about a minor third, no distinct beats being audible. In this case the
intensity of the compound sound might be expected to be double of that of
the (equal) component sounds. The impression upon the observer hardly
corresponded to this anticipation. It was difficult to say that the compound
sound was decidedly the louder ; although the accession of the second sound
as an addition to the first could always be distinguished, and this whether
the higher or the lower sound were the one added. It may be remarked
that the question involved in this experiment is partly physiological, and not
merely mechanical as in the case of sounds nearly in unison.
271.
ON THE MAGNETIC ROTATION OF LIGHT AND THE SECOND
LAW OF THERMODYNAMICS.
[Nature, LXIV. pp. 577, 578, 1901.]
IN a paper published sixteen years ago I drew attention to a peculiarity
of the magnetic rotation of the plane of polarisation arising from the cir-
cumstance that the rotation is in the same absolute direction whichever way
the light may be travelling. "A consequence remarkable from the theoretical
point of view is the possibility of an arrangement by which the otherwise
general optical law of reciprocity shall be violated. Consider, for example,
a column of diamagnetic medium exposed to such a force that the rotation is
45°, and situated between two Nicols whose principal planes are inclined to
one another at 45°. Under these circumstances light passing one way is
completely stopped by the second Nicol, but light passing the other way is
completely transmitted. A source of light at one point A would thus be
visible at a second point B, when a source at B would be invisible at A ; a
state of things at first sight* inconsistent with the second law of thermo-
dynamics." (Phil. Trans. CLXXVI. p. 343, 1885; Scientific Papers, Vol. II.
p. 360.) It is here implied that the inconsistency is apparent only, but I did
not discuss it further.
In his excellent report (" Les Lois theoriques du Rayonnement, Rapports
presented au Congres International de Physique," Paris, 1900, Vol. II. p. 29),
W. Wien, considering the same experimental combination of Nicols and
magnetised dielectric, arrives at a contrary conclusion. It may be well to
quote his statement of the case. " La rotation magne'tique du plan de
polarisation constitue un cas exceptionnel digne de remarque, et Ton pourrait
ici imaginer un dispositif qui mettrait en echec le principe de Carnot s'il
n'existait pas une compensation inconnue.
* The italics are in the original That magnetic rotation may interfere with the law of
reciprocity had already been suggested by Helmholtz.
556 MAGNETIC ROTATION OF LIGHT. [271
" Faisons, en effet, les suppositions suivantes : Deux corps de temperature
egale sont entoures d'une enveloppe adiabatique. Les rayons qu'ils s'en-
voient r^ciproquement traversent deux prismes de nicol. Entre ces prismes
se trouve une substance non absorbante sur laquelle agissent des forces
magnetiques qui font tourner le plan de polarisation d'un angle determine.
La radiation emanant du corps 1 penetre dans le nicol 1. Nous supposerons
que le rayon subissant la reflexion totale n'est pas absorbe, mais renvoye"
dans sa propre direction par des miroirs convenablement disposes. Admettons
que le plan de polarisation soit tourne" de 45° par les forces magnetiques.
La section principale du deuxieme nicol etant orientee dans la direction
parallele au plan de polarisation du rayon emergent, toute la lumiere trans-
raise par la substance absorbante (sic) traversera le nicol. Par consequent,
la moiti£ des rayons e"mis par le corps 1 frappera le corps 2.
" Les rayons emis par le corps 2 se divisent en deux parties egales, dans
le nicol 2. Une moitie est, comme precedemment, renvoyee par reflexion.
L'autre moitie, apres que son plan de polarisation a subi une rotation de 45°
dans le meme sens que les rayons emis par le corps 1, vient frapper le premier
nicol. La section principale de ce nicol etant perpendiculaire au plan de
polarisation, aucune radiation ne le traverse, et nous pouvons renvoyer toute
la lumiere au corps 2.
" Le corps 2 re9oit ainsi trois fois plus d'energie que le corps 1. [That is,
2 receives the whole of its own radiation and the half of that of 1, while 1
receives only the half of its own radiation.] L'un de ces corps s'echauffera
par consequent de plus en plus aux depens de 1'autre."
Wien then suggests certain ways of escape from this conclusion, but it
appears to me that the difficulty itself depends upon an oversight. It is not
possible to send back to 2 the whole of its radiation in the manner proposed.
The second half, which after passage of Nicol 2 is totally reflected at Nicol 1
and then returned upon its course, on its arrival at Nicol 2 is not transmitted
(as Wien seems to suppose) but is totally reflected. When again returned
upon its course by a perpendicular reflector, and again reflected through 45°
by the magnetised medium, it is in a condition to be completely transmitted
by Nicol 1, and thus finds its way to body 1, and not to body 2 as the
argument requires. The two bodies receive altogether the same amount of
radiation, and there is therefore no tendency to a change of temperature.
Although I have not been able to find any note of it, I feel assured that
the above reasoning was present to my mind when I wrote the passage
already cited.
272.
ON THE INDUCTION-COIL*.
[Philosophical Magazine, n. pp. 581—594, 1901.]
ALTHOUGH several valuable papers relating to this subject have recently
been published by Oberbeck-f-, Walter J, Mizuno§, Beattie||, and KlingelfussIT,
it can hardly be said that the action of the instrument is well understood.
Perhaps the best proof of this assertion is to be found in the fact that, so far
as I am aware, there is no a priori calculation, determining from the data of
construction and the value of the primary current, even the order of mag-
nitude of the length of the secondary spark. I need hardly explain that
I am speaking here (and throughout this paper) of an induction-coil working
by a break of the primary circuit, not of a transformer in which the primary
circuit, remaining unbroken, is supplied with a continuously varying alter-
nating current.
The complications presented by an actual coil depend, or may depend,
upon several causes. Among these we may enumerate the departure of the
iron from theoretical behaviour, whether due to circumferential eddy-currents
or to a failure of proportionality between magnetism and magnetizing force.
A second, and a very important, complication has its origin in the manner of
break, which usually occupies too long a time, or at least departs too much
from the ideal of an instantaneous abolition of the primary current. A third
complication arises from the capacity of the secondary coil, in virtue of which
the currents need not be equal at all parts of the length, even at the same
* From the Jubilee volume presented to Prof. Bosscha.
t Wied. Ann. LXH. p. 109 (1897); LXIV. p. 193 (1898).
J Wied. Ann. LXH. p. 300 (1897) ; LXVI. p. 623 (1898).
§ Phil Mag. XLV. p. 447 (1898).
|| Phil. Mag. L. p. 139 (1900).
f Wied. Ann. v. p. 837 (1901).
558 ON THE INDUCTION-COIL. [272
moment of time. If we ignore these complications, treating the break as
instantaneous, the iron as ideal, and the secondary as closed and without
capacity, the theory, as formulated by Maxwell*, is very simple. In his
notation, if x, y denote the primary and secondary currents, L, M, N the
coefficients of self and mutual induction, the energy of the field is
(1)
If c be the primary current before the break, the secondary current at time t
after the break has the expression
(2)
8 being the resistance of the secondary circuit. The current begins with
a value c . M/N, and gradually disappears.
The formation of the above initial current is best understood in the light
of Kelvin's theorem, as explained by me in an early paperf. For this
purpose it is more convenient to consider the reversed phenomenon, viz., the
instantaneous establishment of a primary current c. The theorem teaches
that subject to the condition x — c the kinetic energy (1) is to be made
a minimum; so that
Me + Ny = 0
gives the initial secondary current. In the case of the break we have merely
to reverse the sign of y.
Immediately after the break, when x — 0 and y has the above value, the
kinetic energy is
Immediately before the break the kinetic energy is \ Z/c2, so that the loss of
energy at break — the energy of the primary spark — is
vanishing when the primary and secondary circuits are closely intertwined —
the case of no " magnetic leakage."
If we maintain the suppositions as to the behaviour of the iron and the
suddenness of the break, the above calculated secondary current may be
supposed to be instantaneously formed, even although the secondary circuit
be not closed. This is most easily seen when a condenser, such as a leyden-
* " Electromagnetic Field," Phil. Trans. 1864 ; Maxwell's Scientific Papers, i. p. 546.
t " On some Electromagnetic Phenomena considered in connexion with the Dynamical
Theory," Phil. Mag. XXXMII. p. 1 (1869) ; Scientific Papers, i. p. 6.
1901] ON THE INDUCTION-COIL. 559
jar, is associated with the ends of the secondary. Even when no jar is
applied, the capacity of the secondary itself acts in the same direction and
allows the formation of the current. Whether partly due to a jar or not, it
will be convenient for the present to regard the capacity as associated with
the terminals only of the secondary wire. Under these circumstances the
secondary current follows the laws laid down by Kelvin in 1853, the same in
fact as govern all vibrations in which there is but one degree of freedom. If
the resistance is not too high, the current is oscillatory. After the lapse of
one quarter of a complete period of these oscillations, the current vanishes,
and the whole remaining energy is the potential energy of electric charge.
If the resistance of the secondary wire can be neglected (so far as its influence
during this short time is concerned), the potential energy of charge is the
equivalent of the original energy of the secondary current at the moment
after the break. In the case of no magnetic leakage, this is again the same
as the energy of the primary current before break.
On these principles it is easy to calculate a limit for the maximum
potential-difference at the terminals of the secondary, or for the spark-length,
so far as this is determined by the potential-difference. For if q be the
capacity at the secondary terminals, V the maximum potential-difference, the
energy of the charge is ^ q V2, and this can never exceed the energy of the
primary current before break, viz., \Lc-. The limit to the value of V is
accordingly
V=c^(L/q\ (4)
and it is proportional to the primary current.
So long as the iron can be treated as ideal, the above formula holds good,
and upon the supposition of a sufficiently sudden break there seems to be no
reason why it should not afford a tolerable approximation to the actual
maximum value of V. The proportionality between spark-length and primary
current was found to. hold good in Walter's experiments over a considerable
range.
When the core is very long in proportion to its diameter, or when it
approximates to a closed circuit, the behaviour of the iron may deviate
widely from that described as ideal, and the quantity denoted by L has no
existence. But the principle remains that the energy of charge at the
moment preceding the secondary spark cannot exceed, though it may some-
what closely approach, the energy of the primary current before break.
We have next to consider how the energy of the primary current is to
be reckoned, and here we encounter questions as to which opinion is not
yet undivided. The general opinion would, I suppose, be that the bodily
magnetization of the iron represents a large store of available energy. If
this be correct, the inference would be irresistible in favour of a very long,
560 ON THE INDUCTION-COIL. [272
or a completely closed, iron core. Some years ago*, reasoning on the basis of
the theory of Warburg and Hopkinson, I endeavoured to show that highly
magnetized iron could not be regarded as a store of energy — that the energy
expended in producing the magnetization was recoverable but to a small
extent, or not at all. Although this conclusion does not appear to have been
accepted, perhaps in consequence of an erroneous application to alternating
current transformers, I still see no means of escape from it. The available
energy of a highly magnetized closed circuit of iron is insignificant. If the
length be limited, there is available energy, in virtue of the free polarity at
the ends.
The theory is best illustrated by the case of an ellipsoid of revolution
exposed to uniform external magnetizing force «£)' acting parallel to the axis.
" If 3 be the magnetization parallel to the axis of symmetry (2c), the de-
magnetizing effect of 3 is N%, where N is a numerical constant, a function
of the eccentricity (e). When the ellipsoid is of the ovary or elongated form,
a = b = c V(l - e*),
becoming in the limiting case of the sphere (e = 0),
N=%7T-
and at the other extreme of elongation assuming the form
2c
/e,x
(5)
" The force actually operative upon the iron is found by subtracting
from that externally imposed, so that
and if from experiments on very elongated ellipsoids (N = 0) we know the
relation between .£) and 3, then the above equation gives us the relation
between .£)' and 3 for any proposed ellipsoid of moderate elongation. If we
suppose that $ is plotted as a function of 3, we have only to add in the
ordinates N%, proper to a straight line, in order to obtain the appropriate
curve for «£)'."
The work expended in magnetizing the iron is per unit of volume
* "On the Energy of Magnetized Iron," Phil. Mag. xxn. p. 175 (1886); Scientific Papers,
n. p. 543.
1901] ON THE INDUCTION-COIL. 561
if we reckon from the condition of zero magnetization. The first part is
practically wasted ; the second, which in most cases of open magnetic circuits
is much the larger, is completely recovered when the iron is demagnetized.
If it appear paradoxical that the large integral electromotive force which
would accompany the disappearance of high magnetization in a closed iron
circuit should be so inefficient, we must remember that the mechanical value
of electromotive force depends upon the magnitude of the current which it
drives, and that in the present case the existence of more than a very small
current is inconsistent with that drop of magnetization upon which the
electromotive force depends.
The considerations above explained are of interest in the present question
as affording a limit depending only upon the iron core and the secondary
capacity. For 3 cannot exceed a value estimated at about 1700 C.G.S., what-
ever may be the magnetizing force of the primary current. Thus if v be the
volume of the core, the maximum energy* is
i^xvx 17002;
and the limit to V is found by equating this to \qV*, so that
) (6)
I have made a rough application of this formula to a coil in my possession,
with results that may be here recorded. The core had a diameter of 3 cm.
and a length of 27 cm., so that v = 180 c.c. From (5), properly applicable
only to an ellipsoid, we get by setting 2a = 3, 2c = 27, N = '30.
The capacity of the secondary is more difficult to deal with. In modern
coils the greater part would appear to arise from the positive and negative
potentials at the ends of the coil as opposed to the zero potential of the
primary wire. The capacity between the primary and secondary wires, con-
sidered as poles of a condenser, can be calculated and in many cases de-
termined experimentally. The axial dimension of the secondary of the coil
above referred to is about 18 cm., and the external diameter of the primary
wire is about 5 cm., making the area of each of the opposed surfaces
270 sq. cm. The interval between the primary and secondary wires is
•25cm.; so that, taking the specific inductive capacity of the intervening
layer at 3, we get for the capacity in electrostatic measure of the condenser
so constituted
J-x3x|5
* The energy of the primary current without a core is here neglected.
t Another coil by Apps, in which the insulation was sufficiently good to allow the application
of electrostatic methods, was tested experimentally. The capacity between primary and secondary
wires was thus found to be 120 cm., less than the half of that calculated for the first coil. But
in this case an ebonite tube separated the two wires.
36
562 ON THE INDUCTION-COIL. [272
Only a fraction of this, however, is operative in the present case. On the
supposition of a coil constructed in numerous sections, the potential in the
middle will be zero, the same as that of the primary wire, and will increase
numerically towards either end. The factor of reduction on this account
r+i
will be I #2 dx, or -^ , so that we may take as the value of q in (6) about
23 cm. — probably rather an underestimate. Calculating from these data, we
get in (6)
F=2600.
This is in electrostatic measure. The corresponding volts are 7'9 x 105. If
we reckon 33,000 volts to the cm., the spark-length will stand at 24 cm.
The coil in question is supposed to be capable of an 8 or 10 cm.' spark, and
doubtless was capable when new. It is remarkable that the limit, fixed by
the iron and secondary capacity alone, should exceed so moderately the
actual capability of the coil.
The limiting formula (6), in which neither the value of the primary
current nor the number of secondary windings appears, is arrived at by
supposing the iron to be magnetically saturated. It illustrates, no doubt
with much exaggeration, the disadvantage of too great a length. If a be
given, while c varies, v and q are both proportional to c, so that V oc *JN.
And JN oc c"1 nearly. In somewhat the same way the increase of effective
capacity explains the comparative failure of attempts to increase spark-length
by combining similar coils in series, in spite of the augmented energy at the
moment of break*.
If the object be a rough estimate rather than a limit, a more practical
formula will be obtained by substituting for 3 in (6) its approximate value
N\ so that
«£)' denoting the external magnetizing force, due to the primary current.
The actual magnetizing force, required to magnetize the soft iron, is here
regarded as relatively negligible. According to (7) the spark-length is pro-
portional, cceteris paribus, to the primary current ; and it increases with the
length of the coil, since N now occurs in the denominator. The application
must not be pushed into the region where the iron becomes approximately
saturated.
In the above discussion the capacity q of the secondary will probably be
thought to play an unexpectedly important part, and the question may be
raised whether it is really this capacity which limits the spark-length in
* I am indebted to Mr Swinton for the details of some experiments in this direction made
for Lord Armstrong.
1901] ON THE INDUCTION-COIL. 563
actual coils. It is not difficult to prove by experiment that capacities of the
order above estimated, applied to the secondary terminals, do in fact reduce
the spark-length, though not, so far as I have seen, to the extent demanded
by the law of q~*. But we must remember that this law has been obtained
on the assumptions, not to be fulfilled in practice, of absolute suddenness of
break and of entire absence of eddy-currents in the iron. If under these
conditions secondary capacity were also absent, it would seem that there
could be no limit to the maximum potential developed. The experiments
of Prof. J. J. Thomson* may be considered to show that even in extreme
cases, such as the present, the iron, as a magnetic body, would not fail to
As regards the eddy-currents, it may be well to consider a little further
upon what their importance depends. If there were no secondary circuit,
the magnetism of each wire of the iron core would be continued at the
moment after break, supposed infinitely sudden, by a superficial eddy-current.
A secondary circuit, closely intertwined, with the primary, would transfer
these eddy- currents to itself, and so continue for the first moment the mag-
netism of the core. But a little later, as the magnetism diminished, eddy-
currents would tend to be formed, and their importance for our purpose
depends upon their duration. If this be short, compared with the time-
constants of the secondary circuit, their influence may be neglected. Other-
wise the electromotive force of the falling magnetism lags, and acts to less
advantage. The time-constant, viz. the time in which the current falls in
the ratio e :l, for the principal eddy-current in a cylinder of radius R is
given by
torfiCK /ox
= (2=464?' •'
where C represents the conductivity and fi the permeability f. If d be the
thickness of a thin sheet having the same time-constant as the wire o'f radius
R, it is easily shown in the same way that
d-.R^-jr: 2-404.
If we take for iron in C.G.S. measure
(7=1/9611,. /A = 500,
we get approximately
.................................... (9)
so that for a wire of 1 mm. diameter T = -^fa second. It may be doubted
whether this would be small enough to prevent the eddy-currents reacting
injuriously upon the secondary circuit.
* Recent Researches, p. 323.
t Brit. Assoc. Rep. p. 446 (1882) ; Scientific Papers, n. p. 128.
36—2
564 ON THE INDUCTION-COIL. [272
We will now consider the third of the causes which impose a limit upon
the secondary spark, viz. want of suddenness in the break, supposed for the
present to be unprovided with a condenser. After the cessation of metallic
contact the primary current is prolonged by the formation of a sort of arc,
the duration of which depends among other things, such as the character of
the metals, upon the magnitude of the current itself. If we again suppose
the behaviour of the iron to be ideal, we may treat the secondary circuit as
a simple vibrator, upon which acts a force ( U) proportional to the rate of fall
of the primary current. The equation of such a vibrator is, as usual,
£ + -*+-- *' <10>
and the solution corresponding to u = 0 (no charge), dujdt = Q (no current),
when £ = 0, is*
u = -1, re-fc <«-«') sin n' (t - t') . Udt', . . .(11)
n J0
where
ri = J(n*-iK*) (12)
The various elements of (11) represent in fact the effects at time t of the
velocities Udt' communicated (t — t') earlier. In the present case we are to
suppose that U is positive throughout, and that fUdt' is given.
The integral simplifies in the case of K = 0, that is of evanescent secondary
resistance. We have then n' = n, and
- I'
n Jo
.(13)
It is easy to see that the integral, representing the potential at the secondary
terminals, is a maximum when U is concentrated at some one time t', and
t is such that
sinn(t-t')=l,
that is, when the break is absolutely sudden and the time considered is one
quarter period later. If the break be not sudden, sin n (t — t') will depart
from its maximum value during part of the range of integration, and the
highest possible value of u will not be attained.
The theory is substantially the same if K be finite. There is some value
of (t — t') for which
is a maximum ; and the greatest value of u will be arrived at by concen-
trating U at some time t', and by so choosing t that (t — t') has the value
above defined. The conclusion is that if the primary current fall to zero
* Theory of Sound, Vol. i. § 66.
1901] ON THE INDUCTION-COIL. 565
from its maximum value without oscillation, the potential at the secondary
terminals will be greatest when this fall is absolutely sudden, and that this
greatest value begins to be sensibly departed from when the break occupies
a time comparable with one of the time-constants of the secondary circuit.
In the case of no resistance we have to deal merely with the time of
secondary oscillation ; but if the resistance is high, the other time-constant,
N /S, may be the smaller (see equation (2)}.
It is here that the character of the secondary coil, especially as regards
the number of its windings, enters into the question. On the supposition of
an absolutely sudden break, we arrived at the rather paradoxical conclusion
that the limit of spark-length depended only upon the capacity of the
secondary without regard to the number of windings — a number which could
be changed in a high ratio without sensibly influencing the capacity. We
see now, at any rate, that a reduction in the number of windings, and the
accompanying diminution in the time of oscillation, would necessitate a
greater and greater suddenness of break, if the full effect is to be retained.
We will now consider the action of the primary condenser — a question, the
reader may be inclined to think, already too long postponed. For it is well
known that in most actual coils the condenser is an auxiliary of the utmost
importance, increasing the spark-length 5 or 10 times, even when the break
is made at pieces of platinum. And, although it has been customary to say,
no doubt correctly, that the condenser acts by absorbing into itself the
primary spark, and so increasing the suddenness of break, it is usual to
attribute to it a further virtue, and not unnaturally when it is remembered
that the effect may be not merely to stop, but actually to reverse, the primary
current. If, however, the theory of the foregoing pages is correct, we shall
be constrained to take a different view.
The action of the condenser, and especially the most advantageous
capacity, has been studied experimentally by Walter and by Mizuno. That
there must be a most advantageous capacity is evident beforehand, inasmuch
as a very small capacity is continuous with no condenser at all, and a very
large capacity is continuous with an uninterrupted flow of the primary
current. It is more instructive that the former observer found the most
advantageous capacity to vary with the manner of break (whether in air or
under oil), and that the latter found a dependence upon the strength of the
primary current, a larger current demanding a larger condenser.
When a condenser is employed, it is important that it be connected as
directly as possible with the points between which the break is made to
occur. A comparatively small electromagnet, included between one of the
break-points and the associated condenser-terminal, suffices to diminish, or
even to annul, the advantage which the use of the condenser otherwise
566 ON THE INDUCTION-COIL. [272
presents*. The explanation is, of course, that the current in an electro-
magnet so situated tends to flow on across the break -gap, and so to establish
an arc, with a force which the condenser is powerless to relieve.
Returning to the theoretical aspect of the question, and inquiring whether
there is any reason for expecting a condenser to give an advantage as
compared with an absolutely sudden cessation of the primary current, it
is difficult to see ground for other than a negative answer. In the case of
no magnetic leakage, somewhat closely approached, one would suppose, in
practice, an instantaneous abolition of the primary current throws the whole
available energy into the secondary circuit, and thus, doing all that is possible,
allows no room for an improvement. Under such conditions a condenser can
only do harm.
In the opposite extreme case of but a relatively small mutual induction
between primary and secondary, it is indeed conceivable that the action of
a condenser may be advantageous. The two currents would then be com-
paratively independent and, if the resistances were low, they might execute
numerous oscillations. If the primary current were simply stopped, the
effect in the secondary would be small ; whereas, especially if there were
synchronism, the vibrations of the primary current rendered possible by the
condenser might cause an accumulation of effect in the secondary. The case
would be that of "intermittent vibrations f," such as may occur when a large
tuning-fork is clamped in a vice. A vibration, started by a blow, in one prong
gradually transfers itself to the other. But it is difficult to believe that any-
thing of this sort occurs in an induction-coil as actually used.
I do not know how far the theoretical arguments here advanced will
convince the reader that the use of a condenser in the primary circuit should
offer no advantage as compared with a sufficiently sudden simple break ; but
I may confess that I should have hesitated to put them forward had I not
obtained experimental confirmation of them. My earlier attempts in this
direction were unsuccessful. A quick break was constructed in which a
spring, bearing upwards against a stop, could be knocked away by a blow
with a staff, or by a falling weight. Although the contacts were of platinum,
but little advantage was gained in comparison with the ordinary platinum
break of the coil. Thus in one set of experiments, where the coil was excited
by a single Grove cell, a break made quickly by hand gave a spark about
8 mm. long. The use of a weight, hung by a cotton thread, and falling
through about 12 feet when the thread was burned, increased the length only
to 8£ mm. This was without a condenser. When the condenser was applied,
the spark-length was 14 mm., and it made no perceptible difference whether
or not the falling weight was employed. Considering that the velocity of the
* PkiL Mag. Vol. n. p. 282 (1901). [Vol. iv. p. 552.]
t Theory of Sound, Vol. i. § 114.
1901] ON THE INDUCTION-COIL. 567
weight at impact must have been about 30 feet per second and that its mass
was large compared with that of the spring, these results were far from pro-
mising. With a stronger primary current the advantage gained from the
condenser was much greater, and the utility of the quicker break, with or
without condenser, seemed to be nil.
But, in spite of the failure of the quick break, one or two observations
presented themselves which seemed worthy of being followed up. It was
noticed that, with one Grove cell in the primary, the spark, although very
inferior when no condenser at all was employed, was improved when the
usual condenser (of large capacity) was replaced by a single sheet of coated
glass (Franklin's pane). And, what was perhaps more instructive still, when
the already weak primary current was further reduced by the insertion of one
or two ohms extra resistance, the spark-length (now very small) was less with
than without the usual coil condenser. This observation was repeated, with
like result, upon another coil (by Apps) and its associated condenser. At
any rate in the case of very weak primary currents, the usual condenser did
harm rather than good.
The view, suggested by the foregoing results, that while the ordinary
break was quick enough in the case of weak currents to allow a condenser
to be dispensed with, the superior arcing power of strong currents demanded
a much more rapid break, encouraged further efforts. An attempt to secure
suddenness by forcibly breaking with a jerk a length of rather thin copper
wire, forming part of the primary circuit, failed entirely, as did also, perhaps
for want of sufficiently powerful appliances, an attempt to blow up a portion
of the primary circuit by electric discharge. Another method, however, at
once allowed an advance to be secured. This consisted in cutting the primary
circuit by a pistol-bullet ; and it was found that this form of break without
condenser was about as efficient as the usual platinum break with condenser,
although the primary current was increased to that supplied by three or four
Grove cells and the spark-length to 40 mm., that is, under about the ordinary
conditions of working.
A further improvement was effected by cutting away about half of the
bullet with the intention of raising its velocity. The following results were
recorded with an Apps' coil excited by three Grove cells. The spark-gap
being 50 mm., the usual platinum break and condenser were not able to send
a spark across. Even with the somewhat more efficient break provided by
a pot of mercury well drowned in oil and condenser, only about one break in
fifteen succeeded. On the other hand, of three bullets fired so as to cut the
primary wire (no condenser) two succeeded ; while for the failure of the third
there was some explanation. The bullet without condenser was now dis-
tinctly superior to the best ordinary break with condenser.
The next step was the substitution of a rifle-bullet, fired from a service
rifle. Here again the bullets were reduced to about one-half, and after
568 ON THE INDUCTION-COIL. [272
cutting the wire were received in a long box packed with wet sawdust. At
60 mm., while the mercury-under-oil break with condenser gave only feeble
brush-discharges, good sparks were nearly uniformly obtained from the bullet
working without a condenser. At 70 mm. the bullet without condenser was
about upon a level with the mercury-under-oil break with condenser at
60 mm. As regards the strength of the primary current, if there was any
difference, the advantage was upon the side of the ordinary break with
condenser, inasmuch as in the case of the bullet the leads were longer and
included about 8 cm. of finer copper wire where the bullet passed.
In the next set of experiments upon the same Apps' coil excited by three
Groves, the bullet was used each time, and the comparison was between the
effect with and without the usual coil condenser. At 55 mm. the bullet
without condenser gave each time a fair or a good spark, while with the
condenser there was nothing more than a feeble brush scarcely visible in
a good light.
The single pane of coated glass was next substituted for the usual con-
denser of the coil, with the idea that possibly this might be useful although
the larger capacity was deleterious. But no distinct difference was detected
when the bullet was fired with this or without any condenser.
In the last set of experiments now recorded the primary current was
raised, six Grove cells being employed partly in parallel, and the wire was
cut each time by a rifle-bullet. At 90 mm. no spark could be got when the
coil condenser was in connexion ; when it was disconnected, a spark, good or
fair, was observed nearly every shot.
Altogether these experiments strongly support the view that the only use
of a condenser, in conjunction with an ordinary break, is to quicken it by
impeding the development of an arc, so that when a sufficient rapidity of
break can be obtained by other means, the condenser is deleterious, operating
in fact in the reverse direction, and prolonging the period of decay of the
primary current. It is hoped that the establishment of this fact will inspire
confidence in the theory, and perhaps suggest improvements in the design of
coils. The first requirement is evidently the existence of sufficient energy at
break, and this implies a considerable mass of iron, well magnetized, and not
forming a circuit too nearly closed. The full utilization of this energy is
impeded by want of suddenness in the break, by eddy-currents in the iron,
and (in respect of spark-length) by capacity in the secondary. It is to be
presumed that in a well-designed coil these impediments should operate
somewhat equally. It would be useless to subdivide the iron, or to reduce
the secondary capacity, below certain limits, unless at the same time the
break could be made more sudden. It would not be surprising if it were
found that the tentative efforts of skilful instrument-makers have already
led to a suitable compromise, at least in the case of coils of moderate size.
The design of larger instruments may leave more to be accomplished.
CONTENTS OF VOLUMES I.-IV.
CLASSIFIED ACCORDING TO SUBJECT.
MATHEMATICS.
PAGE
ri
On the Values of the Integral I QnQn'dfj,, Qn, Qn' being
Jo
Laplace's Coefficients of the orders n, n', with an appli-
cation to the Theory of Radiation. Art. 3. 1870 . . Vol. I. 21
On a Correction sometimes required in Curves professing to
represent the connexion between two Physical Magnitudes.
Art. 12. 1871 . . ... . . . . „ 135
Notes on Bessel's Functions. Art. 15. 1872 . . . ,,140
Note on the Numerical Calculation of the Roots of Fluctuating
Functions. Art. 28. 1874 ,,190
A History of the Mathematical Theories of Attraction and the
Figure of the Earth from the time of Newton to that of
Laplace. By I. Todhunter, M.A., F.R.S. Two Volumes.
(London, Macmillan & Co., 1873.) Art. 29. 1874 . . „ 196
On the Approximate Solution of Certain Problems relating to
the Potential. Art. 39. 1876 ,,272
Questions from Mathematical Tripos Examination for 1876.
Art. 41. 1876 ,,280
On the Relation between the Functions of Laplace and Bessel.
Art. 51. 1878 ,,338
A simple Proof of a Theorem relating to the Potential.
Art. 54. 1878 ,,347
On the Resultant of a large number of Vibrations of the
same Pitch and of arbitrary Phase. Art. 68. 1880 . '. „ 491
On Point-, Line-, and Plane-Sources of Sound. Art. 147. 1888 Vol. III. 44
On James Bernoulli's Theorem in Probabilities. Art. 243.
1899 .. . • Vol. IV. 370
570 CONTENTS OF VOLUMES I. — IV.
GENERAL MECHANICS.
PAGE
Some General Theorems relating to Vibrations. Art. 21. 1873 Vol. I. 170
Section I. The natural periods of a conservative system, vibrating
freely about a configuration of stable equilibrium, fulfil the
stationary condition . , . . . . „ 170
Section II. The Dissipation Function . . . '.'".''*. „ 176
Section III. [The Reciprocity Theorem] . . , . ,,176
On the Vibrations of Approximately Simple Systems. Art. 24.
1873, 1874 ,,185
On the Fundamental Modes of a Vibrating System. Art. 25.
1873 ,,186
A Statical Theorem. Art. 32. 1874, 1875 . . . . „ 223
General Theorems relating to Equilibrium and Initial and
Steady Motions. Art. 34. 1875 ..... „ 232
Uniformity of Rotation. [Phonic Wheel] Art. 56. 1878 . „ 355
On Maintained Vibrations. Art. 97. 1883 . . .. . Vol. II. 188
The Soaring of Birds. Art. 98. 1883 . . . . -f. „ 194
Investigation of the Character of the Equilibrium of an In-
compressible Heavy Fluid of Variable Density. Art. 100.
1883 ,,200
Suggestions for facilitating the Use of a Delicate Balance.
Art. 104. 1883 . . . .- ... . . ,,226
A Theorem relating to the Time-Moduli of Dissipative
Systems. Art. 125. 1885 . t; ..:/... . „ 428
Testing Dynamos. Art. 133. 1886 .... . ,,474
The Reaction upon the Driving-Point of a System executing
Forced Harmonic Oscillations of Various Periods, with
Applications to Electricity. Art. 134. 1886 . . . „ 475
On the Maintenance of Vibrations by Forces of Double
Frequency, and on the Propagation of Waves through a
Medium endowed with a Periodic Structure. Art. 142.
1887 . . . .' ' . . " . . . . Vol. III. 1
The Sailing Flight of the Albatross. Art. 159. 1889. . „ 267
On the Vibrations of an Atmosphere. Art. 166. 1890 . „ 335
CONTENTS OF VOLUMES I. — IV. 571
PAGE
On Huygens's Gearing in Illustration of the Induction of
Electric Currents. Art. 171. 1890 .... Vol. III. 376
The Bourdon Gauge. Art. 172. 1890 „ 379
Experiments in Aerodynamics. [Review of Langley's]
Art. 184. 1891 ,,491
Superheated Steam. Art. 188. 1892 ,,538
Heat Engines and Saline Solutions . . . . . „ 539
Heat Engines and Saline Solutions ..... „ 540
Remarks on Maxwell's Investigation respecting Boltzmann's
Theorem. Art. 190. 1892 ,,554
Grinding and Polishing of Glass Surfaces. Art. 205. 1893 . Vol. IV. 74
On the Propagation of Waves along Connected Systems of
Similar Bodies. Art. 235. 1897 „ 340
On Iso-periodic Systems. Art. 242. 1898 . . . . „ 367
On the Calculation of the Frequency of Vibration of a System
in its GraVest Mode, with an Example from Hydro-
dynamics. Art. 249. 1899 „ 407
The Law of Partition of Kinetic Energy. Art. 253. 1900 . „ 433
The Mechanical Principles of Flight. Art. 257. 1900 . . „ 462
On a Theorem analogous to the Virial Theorem. Art. 262.
1900 .' ,,491
Polish. Art. 268. 1901 . 542
ELASTIC SOLIDS.
On the Nodal Lines of a Square Plate. Art. 22. 1873 . Vol. I. 182
Vibrations of Membranes. Art. 26. 1873 . . . . „ 187
On the Infinitesimal Bending of Surfaces of Revolution.
Art. 78. 1881 ......... 551
On Waves propagated along the Plane Surface of an Elastic
Solid. [With reference to Earthquakes] Art. 130. 1885. Vol.11. 441
On the Bending and Vibration of Thin Elastic Shells, especially
of Cylindrical Form. Art. 152. 1888 . . . . Vol. III. 217
Note on the Free Vibrations of an Infinitely Long Cylindrical
Shell. Art. 155. 1889 , 244
572 CONTENTS OF VOLUMES I. — IV.
PAGE
On the Free Vibrations of an Infinite Plate of Homogeneous
Isotropic Elastic Matter. Art. 156. 1889 . . . Vol.111. 249
On the Uniform Deformation in Two Dimensions of a
Cylindrical Shell of Finite Thickness, with Application
to the General Theory of Deformation of Thin Shells.
Art. 162. 1889 ,,280
On Bells. Art. 164. 1890 . . . . . . . „ 318
Appendix : On the Bending of a Hyperboloid of Kevolution . „ 330
The Bourdon Gauge. Art. 172. 1890 ,,379
On the Stresses in Solid Bodies due to Unequal Heating,
and on the Double Refraction resulting therefrom.
Art. 265. 1901 . Vol. IV. 502
CAPILLARITY.
The Instability of Jets. Art. 58. 1879 Vol.1. 361
The Influence of Electricity on Colliding Water Drops.
Art. 59. 1879 „ 372
On the Capillary Phenomena of Jets. Art. 60. 1879 . „ 377
Appendix I. [Vibrations about a Cylindrical Figure] . . „ 396
Appendix II. [Vibrations about a Spherical Figure] . . „ 400
Further Observations upon Liquid Jets, in Continuation of
those recorded in the Royal Society's ' Proceedings ' for
March and May, 1879. Art. 85. 1882 .... Vol. II. 103
On some of the Circumstances which influence the Scattering of
a nearly Vertical Jet of Liquid ..... „ 103
Influence of Regular Vibrations of Low Pitch ... „ 106
The Length of the Continuous Part . . . . „ 110
Collision of Two Resolved Streams . . . . . ,,112
Collision of Streams before Resolution . . . . . „ 115
On the Equilibrium of Liquid Conducting Masses charged
with Electricity. Art. 90. 1882 „ 130
On the Crispations of Fluid resting upon a Vibrating Support.
Art. 102. 1883 . . . ... . . ,,212
On Laplace's Theory of Capillarity. Art. 106. 1883 . . „ 231
The Form of Standing Waves on the Surface of Running
Water. Art. 109. 1883 ,,258
On the Tension of Recently Formed Liquid Surfaces.
Art. 167. 1890 . Vol. III. 341
CONTENTS OF VOLUMES I. — IV. 573
PAGE
Measurements of the Amount of Oil necessary in order to
check the Motions of Camphor upon Water. Art. 168.
1890 Vol.111. 347
Foam. Art. 169. 1890 . . . „ 351
On the Superficial Viscosity of Water. Art. 170. 1890 . „ 363
Instantaneous Photographs of Water Jets. Art. 174. 1890 „ 382
On the Tension of Water Surfaces, Clean and Contaminated,
Investigated by the Method of Ripples. Art. 175. 1890 „ 383
Postscript. [Optical Effect of greasy Contamination] . . „ 394
On the Theory of Surface Forces. Art. 176. 1890 . . „ 397
Some Applications of Photography. Art. 179. 1891 . . „ 441
On Reflexion from Liquid Surfaces in the Neighbourhood of
the Polarizing Angle. Art. 185. 1892 496
On the Theory of Surface Forces. II. Compressible Fluids.
Art. 186. 1892 . ,,513
Experiments upon Surface- Films. Art. 192. 1892 . . „ 562
The Behaviour of Clean Mercury ...... 562
Drops of Bisulphide of Carbon upon Water „ 563
Movements of Dust ......... 564
Camphor Movements a Test of Surface-Tension „ 565
Influence of Heat ........ „ 567
Saponine and Soap ........ „ 568
Separation of Motes ......... 569
The Lowering of Tension by the Condensation of Ether Vapour „ 570
Breath Figures and their Projection ...... 570
On the Theory of Surface Forces. III. Effect of Slight
Contaminations. Art. 193. 1892 „ 572
On the Instability of a Cylinder of Viscous Liquid under
Capillary Force. Art. 195. 1892 ,,585
On the Instability of Cylindrical Fluid Surfaces. Art. 196.
1892 ,,594
Investigations in Capillarity. Art. 251. 1899 . . . Vol. IV. 415
The Size of Drops ,,415
The Liberation of Gas from Supersaturated Solutions . . „ 420
Colliding Jets ,,421
The Tension of Contaminated Water-Surfaces „ 425
A Curious Observation .... „ 430
574 CONTENTS OF VOLUMES I. — IV.
HYDRODYNAMICS.
PAGE
Notes on Bessel's Functions. Art. 15. 1872 . . . Vol. I. 140
Vibrations of a Liquid in a Cylindrical Vessel. Art. 37. 1875 „ 250
On Waves. Art. 38. 1876 . ..... . „ 251
The Solitary Wave . . . • . . - .... " . „ 256
Periodic Waves in Deep Water ,,261
Oscillations in Cylindrical Vessels . . ; . . . „ 265
On the Approximate Solution of Certain Problems relating to
the Potential. Art. 39. 1876 ..... „ 272
On the Resistance of Fluids. Art. 42. 1876 . . . " „ 287
Notes on Hydrodynamics. Art. 43. 1876 . '.'.'. „ 297
The Contracted Vein . . . . . ... „ 297
Meeting Streams . . ... . . . . ". „ 302
On Progressive Waves. Art. 47. 1877. . . . . ,,322
Note on Acoustic Repulsion. Art. 52. 1878 . . . „ 342
On the Irregular Flight of a Tennis-Ball. Art. 53. 1877 . „ 344
On the Instability of Jets. Art. 58. 1879 . . , . , „ 361
The Influence of Electricity on Colliding Water Drops.
Art. 59. 1879. ........ ,,372
On the Capillary Phenomena of Jets. Art. 60. 1879 . . „ 377
Appendix I. [Vibrations about a Cylindrical Figure] . . „ 396
Appendix II. [Vibrations about a Spherical Figure] . . „ 400
On the Stability, or Instability, of Certain Fluid Motions.
Art. 66. 1880 ,,474
Further Observations upon Liquid Jets, in Continuation of
those recorded in the Royal Society's ' Proceedings ' for
March and May, 1879. Art. 85. 1882 . .' . . Vol.11. 103
On some of the Circumstances which influence the Scattering of
a nearly Vertical Jet of Liquid . , . . . „ 103
Influence of Regular Vibrations of Low Pitch ... „ 106
The Length of the Continuous Part . ... . . ,,110
Collision of Two Resolved Streams . . . . '-..,.. „ 112
Collision of Streams before Resolution . . . . ... { • , „ 115
On the Equilibrium of Liquid Conducting Masses charged
with Electricity. Art. 90. 1882 „ 130
On the Dark Plane which is formed over a Heated Wire in
Dusty Air. Art. 93. 1882 . . . . . . . „ 151
CONTENTS OF VOLUMES I. — IV. 575
PAGE
The Soaring of Birds. Art. 98. 1883 ..... Vol. IT. 194
Investigation of the Character of the Equilibrium of an In-
compressible Heavy Fluid of Variable Density. Art. 100.
1883 . . . .• „ 200
On the Vibrations of a Cylindrical Vessel containing Liquid.
Art. 101. 1883 . ,,208
On the Circulation of Air observed in Kundt's Tubes, and on
some Allied Acoustical Problems. Art. 108. 1883 . „ 239
The Form of Standing Waves on the Surface of Running
Water. Art. 109. 1883 . . . ." . . „ 258
On the Stability or Instability of Certain Fluid Motions. II.
Art. 144. 1887 . . . . . . . . Vol. III. 17
The Sailing Flight of the Albatross. Art. 159. 1889 . . „ 267
On the Vibrations of an Atmosphere. Art. 166. 1890 . „ 335
On the Tension of Water Surfaces, Clean and Contaminated,
Investigated by the Method of Ripples. Art. 175. 1890 „ 383
Some Applications of Photography. Art. 179. 1891 . . „ 441
Experiments in Aerodynamics. [Review of Langley's]
Art. 184. 1891 ,,491
On the Question of the Stability of the Flow of Fluids.
Art. 194. 1892 ,,575
On the Instability of a Cylinder of Viscous Liquid under
Capillary Force. Art. 195. 1892 ,,585
On the Instability of Cylindrical Fluid Surfaces. Art. 196.
1892 ,,594
On the Influence of Obstacles arranged in Rectangular Order
upon the Properties of a Medium. Art. 200. 1892 . Vol. IV. 19
On the Flow of Viscous Liquids, especially in Two Dimensions.
Art. 208. 1893 „ 78
On the Stability or Instability of Certain Fluid Motions. III.
Art. 216. 1895 ......... , . . • • • » 203
On the Propagation of Waves upon the Plane Surface separ-
ating Two Portions of Fluid of Different Vorticities.
Art. 217. 1895 ,,210
576 CONTENTS OF VOLUMES I. — IV.
PAGE
On some Physical Properties of Argon and Helium. Art. 218.
1896 ' . Vol. IV. 215
Density of Argon . . . . . . . . „ 215
The Refractivity of Argon and Helium ..... „ 218
Viscosity of Argon and Helium ...... „ 222
Gas from the Bath Springs ....... „ 223
Buxton Gas . . . . . . . . ,,223
Is Helium contained in the Atmosphere ? . . ' '. ' ' . ' - v „ 224
On the Viscosity of Hydrogen as affected by Moisture.
Art. 234. 1897 . . . . . . . . ,,336
On the Calculation of the Frequency of Vibration of a System
in its Gravest Mode, with an Example from Hydro-
dynamics. Art. 249. 1899 ...... ,,407
On the Viscosity of Argon as affected by Temperature.
Art. 254. 1900 ,,452
The Mechanical Principles of Flight. Art, 257. 1900 .' . „ 462
On the Viscosity of Gases as affected by Temperature.
Art. 259. 1900 ,,481
SOUND.
Remarks on a Paper by Dr Sondhauss. Art. 4. 1870 . Vol. I. 26
On the Theory of Resonance. Art. 5. 1870 . . . „ 33
Introduction ......... ,,33
Part I. ,,37
Several Openings ........ ,,39
Double Resonance . . . . . . . . „ 41
Open Organ-pipes ........ ,,45
Long Tube in connexion with a Reservoir .... ,,48
Lateral Openings ........ ,,50
Part II ,,51
Long Tubes ,,51
Simple Apertures ........ ,,52
Cylindrical Necks ........ ,,53
Potential on itself of a Uniform Circular Disk ... ,,55
Nearly Cylindrical Tubes of Revolution .... ,,62
Upper Limit ......... ,,62
Application to Straight Tube of Revolution whose end lies on
two Infinite Planes ....... ,,64
Tubes nearly Straight and Cylindrical but not necessarily of
Revolution ......... ,,64
Tubes not nearly Straight . . . . . „ 66
Part III ' . - . ... , 67
Experimental ......... ,,67
CONTENTS OF VOLUMES I. — IV. 577
PAGE
On the Vibrations of a Gas contained within a Rigid Spherical
Envelope. Art. 13. 1872 Vol. I. 138
Investigation of the Disturbance produced by a Spherical
Obstacle on the Waves of Sound. Art. 14. 1872 . „ 139
Some General Theorems relating to Vibrations. Art. 21.
1873 „ 170
Section I. The natural periods of a conservative system, vibrating
freely about a configuration of stable equilibrium, fulfil the
stationary condition ....... „ 170
Section II. The Dissipation Function ..... „ 176
Section III. [The Reciprocity Theorem] .... „ 179
On the Nodal Lines of a Square Plate. Art. 22. 1873 . „ 182
On the Vibrations of Approximately Simple Systems. Art. 24.
1873, 1874 ,,185
On the Fundamental Modes of a Vibrating System. Art. 25.
1873 ,,186
Vibrations of Membranes. Art. 26. 1873 .... ,,187
Harmonic Echoes. Art. 27. 1873 ,,188
Mr Hamilton's String Organ. Art. 33. 1875 ... „ 230
Vibrations of a Liquid in a Cylindrical Vessel. Art. 37. 1875 „ 250
On Waves. Art. 38. 1876 ,,251
The Solitary Wave ,,256
Periodic Waves in Deep Water . . . . . . ,,261
Oscillations in Cylindrical Vessels ,,265
Our Perception of the Direction of a Source of Sound. Art. 40.
1876 ,,277
Questions from Mathematical Tripos Examination for 1876.
Art. 41. 1876 ,,280
On the Application of the Principle of Reciprocity to Acoustics.
Art. 44. 1876 ,,305
Acoustical Observations. I. Art. 46. 1877 .... „ 314
Perception of the Direction of a Source of Sound ... »
The Head as an Obstacle to Sound »
Reflection of Sound ......•• » 316
Audibility of Consonants .....•• »
Interference of Sounds from two unisouant Tuning-forks . . „
Symmetrical Bell ......•• »>
Octave from Tuning-forks ....••• »
Influence of a Flange on the Correction for the Open End of a Pipe „
The Pitch of Organ-pipes ,,320
37
578 CONTENTS OF VOLUMES I. — IV.
PAGE
On Progressive Waves. Art. 47. 1877. .... . . . « Vol. I. 322
On the Amplitude of Sound- Waves. Art. 48. 1877 . . „ 328
Absolute Pitch. Art. 49. 1877 . . . . . . „ 331
Note on Acoustic Repulsion. Art. 52. 1878 . . . „ 342
The Explanation of certain Acoustical Phenomena. [Singing
Flames, &c.] Art. 55.' 1878 . . -. . . . ,,348
On the Determination of Absolute Pitch by the Common
Harmonium. Art. 57. 1879 . . . . . . „ 357
On the Instability of Jets. Art. 58. 1879 . . . . „ 361
On the Capillary Phenomena of Jets. Art. 60. 1879 . . „ 377
Appendix I. [Vibrations about a Cylindrical Figure] ;. . „ 396
Appendix II. [Vibrations about a Spherical Figure] . . „ 400
AcousticalObservations.il. Art. 61. 1879. . ' . ••••-.••: „ 402
Pure Tones from Sounding Flames ..... „ 402
Points of Silence near a Wall from which a Pure Tone is
reflected .......... „ 403
Sensitive Flames . • „ 406
Aerial Vibrations of very Low Pitch maintained by Flames . „ 407
Rijke's Notes on a large scale ...... „ 408
Mutual Influence of Organ-pipes nearly in Unison . ... „ 409
Kettledrums ,,411
The .Eolian Harp ,,413
On Reflection of Vibrations at the Confines of Two Media
between which the Transition is Gradual. Art. 63. 1880 „ 460
Acoustical Observations. III. Art. 65. 1880 . . . „ 468
Intermittent Sounds ........ „ 468
A New Form of Siren . . . . . . ,,471
The Acoustical Shadow of a Circular Disk .... „ 472
On the Stability, or Instability, of certain Fluid Motions.
Art. 66. 1880 ,,474
On the Resultant of a large number of Vibrations of the same
Pitch and of arbitrary Phase. Art. 68. 1880 . ; . „ 491
On a New Arrangement for Sensitive Flames. Art. 70.
1880. . . . . . . . . „ 500
The Photophone. Art. 71. 1881 . ..... '. . . „ 501
On the Infinitesimal Bending of Surfaces of Revolution.
Art. 78. 1881 . 551
CONTENTS OF VOLUMES I. — IV. 579
PAGE
Acoustical Observations. IV. Art. 84. 1882 Vol. II. 95
On the Pitch of Organ- Pipes . . /. . ; . „ 95
Slow versus Quick Beats for comparison of Frequencies of
Vibration .*....... ,,97
Estimation of the Direction of Sounds with one Ear . ,,98
A Telephone-Experiment . * . . . ". . )? 99
Very High Notes. Rapid Fatigue of the Ear . . . . „ 99
Sensitive Flames ........ „ 100
Further Observations upon Liquid Jets, in Continuation of
those recorded in the Royal Society's ' Proceedings ' for
March and May, 1879. Art. 85. 1882 . . . . „ 103
On some of the Circumstances which influence the Scattering of
a nearly Vertical Jet of Liquid . . . . . . ,,103
Influence of Regular Vibrations of Low Pitch .... ,,106
The Length of the Continuous Part ,,110
Collision of Two Resolved Streams . . . . . ,,112
Collision of Streams before Resolution . . . . . ,,115
On an Instrument capable of Measuring the Intensity of Aerial
Vibrations. Art. 91. 1882 . . . ,,132
On Maintained Vibrations. Art. 97. 1883 .... „ 188
On the Vibrations of a Cylindrical Vessel containing Liquid.
Art. 101. 1883 ,,208
On the Crispations of Fluid resting upon a Vibrating Support.
Art. 102. 1883 ,,212
On Porous Bodies in Relation to Sound. Art. 103. 1883 . „ 220
On the Circulation of Air observed in Kundt's Tubes, and on
some Allied Acoustical Problems. Art. 108. 1883 . „ 239
Acoustical Observations. V. Art. 110. 1884 ... „ 268
Smoke-jets by Intermittent Vision ..... „ 268
Srnoke-jets and Resonators ...... ,, 269
Jets of Coloured Liquid ,,270
Fish-tail Burners ,,272
Influence of Viscosity ....... „ 273
On Telephoning through a Cable. Art. 115. 1884 . „ 356
On Waves propagated along the Plane Surface of an Elastic
Solid. [With reference to Earthquakes] Art. 130. 1885 „ 441
The Reaction upon the Driving-Point of a System executing
Forced Harmonic Oscillations of Various Periods, with
Applications to Electricity. Art. 134. 1886 . . ' . „ 475
37—2
580 CONTENTS OF VOLUMES I. — IV.
PAGE
On the Maintenance of Vibrations by Forces of Double
Frequency, and on the Propagation of Waves through a
Medium endowed with a Periodic Structure. Art. 142.
1887 Vol. III. 1
Diffraction of Sound. Art. 145. 1888 . ... . . „ 24
On Point-, Line-, and Plane-Sources of Sound. Art. 147.
1888 . . . . . „ 44
On the Bending and Vibration of Thin Elastic Shells, es-
pecially of Cylindrical Form. Art. 152. 1888 . . „ 217
Note on the Free Vibrations of an Infinitely Long Cylindrical
Shell. Art. 155. 1889 . . ' „ 244
On the Free Vibrations of an Infinite Plate of Homogeneous
Isotropic Elastic Matter. Art. 156. 1889 . . . „ 249
On the Uniform Deformation in Two Dimensions of a Cylind-
rical Shell of Finite Thickness, with Application to the
General Theory of Deformation of Thin Shells. Art. 162.
1889 . . . . .... „ 280
On Bells. Art. 164. 1890 ,,318
Appendix : On the Bending of a Hjperboloid of Revolution . „ 330
On the Sensitiveness of the Bridge Method in its Application
to Periodic Electric Currents. Art. 180. 1891 . . „ 452
On the Instability of a Cylinder of Viscous Liquid under
Capillary Force. Art. 195. 1892 , 585
On the Instability of Cylindrical Fluid Surfaces. Art. 196.
1892 594
On the Influence of Obstacles arranged in Rectangular Order
upon the Properties of a Medium. Art. 200. 1892 . Vol. IV. 19
On the Reflection of Sound or Light from a Corrugated
Surface. Art. 206. 1893 . . . . , „ 75
On the Minimum Current audible in the Telephone. Art. 211.
1894 ,,109
An Attempt at a Quantitative Theory of the Telephone.
Art. 212. 1894 . „ 119
On the Amplitude of Aerial Waves which are but just
Audible. Art. 213. 1894 125
CONTENTS OF VOLUMES 1. IV. 581
On the Passage of Waves through Apertures in Plane Screens,
and Allied Problems. Art. 227. 1897 . ' Vol IV 283
Perforated Screen.— Boundary Condition d(j)/dn=0 . . 284
Boundary Condition 0=0 . . . 2gfi
Reflecting Plate.— d^/dn = 0 .... 28g
Reflecting Plate. — 0 = 0 .... 289
Two-dimensional Vibrations .... 290
Narrow Slit. —Boundary Condition d^/dn = 0 ... 291
Narrow Slit. — Boundary Condition 0 = 0 . . . 293
Reflecting Blade. — Boundary Condition d<j>/dn = 0 . . . 294
Reflecting Blade. — Boundary Condition 0 = 0 . . . . 295
Various Applications ........ 295
The Limits of Audition. Art. 228. 1897 . . . . „ 297
On the Incidence of Aerial and Electric Waves upon Small
Obstacles in the Form of Ellipsoids or Elliptic Cylinders,
and on the Passage of Electric Waves through a Circular
Aperture in a Conducting Screen. Art. 230. 1897 . „ 305
Obstacle in a Uniform Field ....... „ 306
In Two Dimensions ........ „ 309
Aerial Waves ......... 310
Waves in Two Dimensions . . . . . . . ,,314
Electrical Applications . . . . . . . ,,317
Electric Waves in Three Dimensions ..... „ 318
Obstacle in the Form of an Ellipsoid ,,323
Circular Aperture in Conducting Screen ..... ,, 324
On the Propagation of Waves along Connected Systems of
Similar Bodies. Art. 235. 1897 „ 340
Some Experiments with the Telephone. Art. 239. 1898 . „ 357
On Iso-periodic Systems. Art. 242. 1898 .... ,,367
On the Cooling of Air by Radiation and Conduction, and on
the Propagation of Sound. Art. 244. 1899 ... „ 376
On Approximately Simple Waves. Art. 261. 1900 . . „ 486
On a Problem relating to the Propagation of Sound between
Parallel Walls. Art. 267. 1901 ,,532
Acoustical Notes. VI. Art. 270. 1901 .... „ 550
Forced Vibrations ,,550
Vibrations of Strings ........ „ 551
Beats of Sounds led to the Two Ears separately ... „ 553
Loudness of Double Sounds ....... „ 554
582 CONTENTS OF VOLUMES L— IV.
THERMODYNAMICS.
PAGE
On the Dissipation of Energy. Art. 35. 1875 . \ . . Vol. I. 238
On the Work that may be gained during the Mixing of Gases.
Art. 36. 1875 ....... '.. „ 242
On a Question in the Theory of Lighting. Art. 76. 1881 . „ 541
On the Tension [Pressure] of Mercury Vapour at Common
Temperatures. Art. 87. 1882 . . ... Vol. II. 125
On the Theory of Illumination in a Fog. Art. 121. 1885 . „ 417
On the Thermodynamic Efficiency of the Thermopile.
Art. 129. 1885 . . ... ... . „ 438
Notes, chiefly Historical, on some Fundamental Propositions
in Optics. Art. 137. 1886 ... ... „ 513
The History of the Doctrine of Radiant Energy. Art. 154.
1889 . . . . ... . . . Vol. III. 238
On the Character of the Complete Radiation at a Given
Temperature. Art. 160. 1889 . .- . . . „ 268
Superheated Steam. Art. 188. 1892 . . . . . „ 538
Heat Engines and Saline Solutions . . . . . „ 539
Heat Engines and Saline Solutions . . . . „ 540
Remarks on Maxwell's Investigation respecting Boltzmann's
Theorem. Art. 190. 1892 554
The Theory of Solutions. Art. 224. 1897 ... . Vol. IV. 267
Liquid Air at One Operation. Art. 240. 1898 . . . „ 360
The Law of Partition of Kinetic Energy. Art. 253. 1900 ,,,.,." „ 433
Remarks upon the Law of Complete Radiation. Art. 260.
1900 . . . . . . .... . .. „ 483
On Balfour Stewart's Theory of the Connexion between
Radiation and Absorption. Art. 263. 1901 . ...» .. „ 494
Does Chemical Transformation influence Weight? Art. 269.
1901 . . . . . " . *. . . . „ 549
On the Magnetic Rotation of Light and the Second Law of
Thermodynamics. Art. 271. 1901 . . . . . ,,557
CONTENTS OF VOLUMES I.— IV. 583
DYNAMICAL THEORY OF GASES.
Note on a Natural Limit to the Sharpness of Spectral Lines
Art. 23. 1873 . . . . Vol. I. 183
On the Work that may be gained during the Mixing of Gases.
Art. 36. 1875 ' ,,242
On the Dark Plane which is formed over a Heated Wire in
Dusty Air. Art. 93. 1882 Vol. II. 151
On the Limit to Interference when Light is Radiated from
Moving Molecules. Art. 157. 1889 . . . . Vol. III. 258
On Van der Waals' Treatment of Laplace's Pressure in the
Virial Equation: Letters to Professor Tait. Art. 181.
1891 . m 465
On the Virial of a System of Hard Colliding Bodies.
Art. 182. 1891 ,,469
Dynamical Problems in Illustration of the Theory of Gases.
Art. 183. 1891 ,,473
Introduction ......... „ 473
Collision Formulae ........ „ 473
Permanent State of Free Masses under Bombardment . . „ 474
Another Method of Investigation ...... „ 479
Progress towards the Stationary State ..... „ 480
Pendulums in place of Free Masses ..... „ 485
Remarks on Maxwell's Investigation respecting Boltzmann's
Theorem. Art. 190. 1892 ,,554
On the Physics of Media that are composed of Free and
Perfectly Elastic Molecules in a State of Motion.
[Introduction to Waterston's Memoir.] Art. 191. 1892. „ 558
The Law of Partition of Kinetic Energy. Art. 253. 1900 . Vol. IV. 433
On the Viscosity of Gases as affected by Temperature.
Art. 259. 1900 ,,481
PROPEKTIES OF GASES.
On the Relative Densities of Hydrogen and Oxygen. (Pre-
liminary Notice.) Art. 146. 1888 .... Vol. III. 37
On the Composition of Water. Art. 153. 1889 „ 233
On the Relative Densities of Hydrogen and Oxygen. II.
Art. 187. 1892
Density of Nitrogen. Art. 197. 1892 .... Vol. IV. 1
584 CONTENTS OF VOLUMES I. — IV.
PAGE
On the Densities of the Principal Gases. Art. 201. 1893 . Vol. IV. 39
The Manometer . . . „ 40
Connexions with Pump and Manometer ..... „ 43
The Weights . . . ... . . . „ 44
The Water Contents of the Globe ... . . . „ 45
Air . . ...... . . . . „ 46
Oxygen . . . . . . . . . „ 47
Nitrogen . . . . . -. ...;.. , ...• .....•; „ 48
Reduction to Standard Pressure . . -,. . , „ 50
Note A. On the Establishment of Equilibrium of Pressure
in Two Vessels connected by a Constricted Channel . . „ 53
On an Anomaly encountered in Determinations of the
Density of Nitrogen Gas. Art. 210. 1894 ... „ 104
Argon, a New Constituent of the Atmosphere. By Lord
Rayleigh and Prof. William Ramsay. Art. 214. 1895 . ,,130
Density of Nitrogen from Various Sources . • . . „ 130
Reasons for Suspecting a hitherto Undiscovered Constituent in Air „ 135
Methods of Causing Free Nitrogen to Combine . ... „ 138
Early Experiments on sparking Nitrogen with Oxygen in presence
of Alkali ,,141
Early Experiments ori Withdrawal of Nitrogen from Air by
means of Red-hot Magnesium . . . . . . ,,144
Proof of the Presence of Argon in Air, by means of Atruolysis „ 150
Negative Experiments to prove that Argon is not derived from
Nitrogen or from Chemical Sources . . . ... ,,153
Separation of Argon on a large scale . . . . . ,,155
Density of Argon prepared by means of Oxygen ... „ 165
Density of Argon prepared by means of Magnesium .. . „ 167
Spectrum of Argon ........ „ 168
Solubility of Argon in Water „ 170
Behaviour at Low Temperatures . . . . . . ,,173
The Ratio of the Specific Heats of Argon . . . ,,174
Attempts to induce Chemical Combination . . . . ,,176
General Conclusions „ 180
Addendum, March 20 (by Prof. W. Ramsay) . . . , ,,184
Addendum, April 9 ........ ,,187
Argon. Art. 215. 1895 ,,188
On some Physical Properties of Argon and Helium. Art. 218.
1896 . . V . . 'i . . . . „ 215
Density of Argon . . . . . . . „ 215
The Refractivity of Argon and Helium . . .. . . „ 218
Viscosity of Argon and Helium . .' ' . . '. . „ 222
Gas from the Bath Springs . .... . . „ 223
Buxton Gas . . .... . . . . ,,223
Is Helium contained in the Atmosphere ? . . . „ 224
On the Amount of Argon and Helium contained in the Gas
from the Bath Springs. Art. 219. 189(i . . ;\ „ 225
CONTENTS OF VOLUMES I. — IV. 585
PAGE
Theoretical Considerations respecting "the Separation of Gases
by Diffusion and Similar Processes. Art. 223. 1896 . Vol. IV. 201
The Theory of Solutions. Art. 224. 1897 . . ". 267
Observations on the Oxidation of Nitrogen Gas. Art. 225. 1897 270
On the Viscosity of Hydrogen as affected by Moisture. Art. 234
1897 „ 888
On the Densities of Carbonic Oxide, Carbonic Anhydride, and
Nitrous Oxide. Art. 236. 1897 „ 347
Carbonic Oxide .......... 347
Carbonic Anhydride ......... 349
Nitrous Oxide ... ...... 350
Liquid Air at One Operation. Art. 240. 1898 . 360
On the Character of the Impurity found in Nitrogen Gas
derived from Urea [with an Appendix containing Details
of Kefractometer]. Art. 241. 1898 . . . . „ 361
Details of Refractometer . . . . . . . „ - 364
On the Cooling of Air by Radiation and Conduction, and on
the Propagation of Sound. Art. 244. 1899 . . . „ 376
On the Conduction of Heat in a Spherical Mass of Air confined
by Walls at a Constant Temperature. Art. 245. 1899 . „ 382
On the Viscosity of Argon as affected by Temperature. Art. 254.
1900 ,,452
On the Passage of Argon through Thin Films of Indiarubber.
Art, 255. 1900 ,,459
On the Weight of Hydrogen desiccated by Liquid Air. Art. 256.
1900 ,,461
On the Viscosity of Gases as affected by Temperature. Art. 259.
1900 ,,481
Spectroscopic Notes concerning the Gases of the Atmosphere.
Art. 264. 1901 ; ,,496
On the Visibility of Hydrogen in Air ..... », 496
Demonstration at Atmospheric Pressure of Argon from very small
quantities of Air ... ..... » 499
Concentration of Helium from the Atmosphere . . . „ 500
On a New Manometer, and on the Law of the Pressure of
Gases between T5 and O'Ol Millimetres of Mercury.
Art. 266. 1901 ,,511
Introduction ... . . • • • • » &H
Improved Apparatus for Measuring very small Pressures . „ 514
Experiments to determine the Relation of Pressure and Volume
at given Temperature . • • ' • • > 519
586 CONTENTS OF VOLUMES I. — IV.
ELECTEICITY AND MAGNETISM.
I'AOK
On some Electromagnetic Phenomena considered in connexion
with the Dynamical Theory. Art. 1. 1869 . . . Vol. I. 1
On an Electromagnetic Experiment. Art. 2. 1870 . . ,,14
On the Theory of Resonance. Art. 5. 1870 . . . „ 33
Introduction . : . . . . '. . „ 33
Part I ~if ..--.:.• „ 37
Several Openings ........ ,,39
Double Resonance ........ ,,41
Open Organ-pipes ........ ,,45
Long Tube in connexion with a Reservoir .... ,,48
Lateral Openings ........ ,,50
Part II ,,51
Long Tubes . „ 51
Simple Apertures . . . . *. . . „ 52
Cylindrical Necks ........ ,,53
Potential on itself of a uniform Circular Disk ... „ 55
Nearly Cylindrical Tubes of Revolution . . . „ 62
Upper Limit ......... ,,62
Application to straight Tube of Revolution whose end lies on
two infinite Planes ....... ,,64
Tubes nearly Straight and Cylindrical but not necessarily of
Revolution .*.-.. „ 64
Tubes not nearly Straight ...... ,,66
Part III ,,67
Experimental ......... „ 67
An Experiment to illustrate the Induction on itself of an
Electric Current. Art. 20. 1872 . . . . 167
Some General Theorems relating to Vibrations. Art. 21. 1873 „ 170
Section I. The natural periods of a conservative system, vibrating
freely about a configuration of stable equilibrium, fulfil the
stationary condition . . . . . . . ,,170
Section II. The Dissipation Function . . . . . ,,176
Section III. [The Reciprocity Theorem] . -. . . . „ 179
On the Approximate Solution of Certain Problems relating to
the Potential. Art. 39. 1876. .. . . . .'.'.. . „ 272
Questions from Mathematical Tripos Examination for 1876.
Art. 41. 1876. . . ... ., , , . . ..'.-.. „ 280
On a Permanent Deflection of the Galvanometer-Needle under
the influence of a rapid series of equal and opposite In-
duced Currents. Art. 45. 1877 . . . . '. . „ 310
Uniformity of Rotation. [Phonic Wheel] Art. 56. 1878 . „ 355
CONTENTS OF VOLUMES I. — IV. 587
PAGE
The Influence of Electricity on Colliding Water Drops. Art. 59.
1879 . . . . . Vol. I. 372
Note on the Theory of the Induction Balance. Art. 69. 1880 „ 497
On the Electromagnetic Theory of Light. Art. 74. 1881 . „ 518
On the Determination of the Ohm [B.A. Unit] in Absolute
Measure. By Lord Rayleigh and Arthur Schuster. Art. 79.
1881 Vol. II. 1
Part I. By Lord Rayleigh „ 1
Part II. By Arthur Schuster ,,20
Adjustment of the Instruments and Determination of Constants „ 20
The Observations ........ ,,24
Air Currents ,,28
Reduction of Observations ...... ,,30
Results ,,34
Experiments to Determine the Value of the British Association
Unit of Resistance in Absolute Measure. Art. 80. 1882 „ 38
Measurements of Coil . . . . . . . . „ 51
Calculation of GK „ 53
Calculation of L ,,53
Theory of the Ring Currents ...... ,,54
L by Direct Experiment ....... ,,55
Correction for Level ,,63
Correction for Torsion ....... ,,64
Value of GK corrected for Level and Torsion . . . . „ 64
Calculation of U ........ ,,64
Measurement of tan p ....... ,,64
Measurement of D ........ ,,65
Reduction of Results ........ ,,66
Comparison with the Standard B. A. Units .... ,,75
On the Specific Resistance of Mercury. By Lord Rayleigh
and Mrs H. Sidgwick. Art. 81. 1882 . . . . „ 78
On a New Form of Gas Battery. Art. 83. 1882 . . ' „ 94
On the Absolute Measurement of Electric Currents. Art. 88.
1882 ,,126
On the Duration of Free Electric Currents in an Infinite
Conducting Cylinder. Art. 89. 1882 . . . . ,,128
On the Equilibrium of Liquid conducting Masses charged with
Electricity. Art. 90. 1882 ,,130
Comparison of Methods for the Determination of Resistances
in Absolute Measure. Art. 92. 1882 . . . . ,,134
Kirchhoff's Method, Maxwell's Electricity and Magnetism, § 759 „ 135
Weber's Method by Transient Currents, Maxwell, § 760 . . -„ 137
588 CONTENTS OF VOLUMES 1. — IV.
PAGE
Comparison of Methods for the Determination of Resistances
in Absolute Measure (continued) ..... Vol. II. 134
Method of Revolving Coil . . .' . . . ,,139
Method of Foster and Lippmann . . . . . „ 143
Weber's Method by Damping ,,145
Lorenz's Method ........ „ 145
Experiments, by the Method of Lorenz, for the Further Deter-
mination of the Absolute Value of the British Association
Unit of Resistance, with an Appendix on the Determi-
nation of the Pitch of a Standard Tuning-Fork. By Lord
Rayleigh and Mrs H. Sidgwick. Art. 94. 1883 . . „ 155
Details of Measurements :
Diameter of Disc . . . . . . . . ,,167
The Inductioii-Coils ,,168
The Distance- Pieces ........ „ 169
The Induction-Coefficients . . . . . . ,,170
The Resistance-Coils ,,171
Appendix : Frequency of Vibration of Standard Fork . . „ 177
Second Appendix : On the Effect of the Imperfect Insulation of
Coils ,,182
On the Mean Radius of Coils of Insulated Wire. Art. 95.
1883 ,,184
On the Imperfection of the Galvanometer as a Test of the
Evanescence of a Transient Current. Art. 105. 1883 . „ 228
On the Measurement of Electric Currents. Art. 107. 1883 „ 237
On the Measurement of the Electrical Resistance between Two
Neighbouring Points on a Conductor. Art. 111. 1884 . „ 276
On the Electro-Chemical Equivalent of Silver, and on the
Absolute Electromotive Force of Clark Cells. By Lord
Rayleigh and Mrs H. Sidgwick. Art. 112 1884 . . „ 278
The Fixed Coils ......... „ 289
The Suspended Coil ,,290
Determination of Mean Radius of Suspended Coil . . . ,,291
Calculation of Attraction ....... „ 295
The Silver Voltameters ,,297
Appendix. [Mathematical Table] . . . . . . ,,327
Explanation of Figures ....... „ 328
Notes:
Note to § 25. [Effect of Temperature on Silver Deposits] . „ 329
Note to § 26. [Mascart's revised Calculation] ... „ 329
Note to § 27. [Copper and Silver] ..... „ 330
Note to § 30. [Clark Cells] ,,331
Note to § 32. [Post Office Daniells] ,,331
Note 1 to § 37. [Clark Cells] ,,331
Note 2 to §37. [Clark Cells] . . . ... „ 332
CONTENTS OF VOLUMES I. — IV. 589
PAGE
A Lecture Experiment on Induction. Art. 114. 1884 . Vol. II. 355
On Telephoning through a Cable. Art. 115. 1884 . . „ 356
On a Galvanometer with Twenty Wires. Art. 116. 1884 . „ 357
On Clark's Standard Cells. Art. 117. 1884 . . . „ 359
On the Constant of Magnetic Rotation of Light in Bisulphide
of Carbon. Art. 118. 1885 ,,360
The Helix ,,367
Correction for Finite Length ...... „ 368
Appendix: Notes on Polarimetry in general .... „ 378
Postscript. [Work of H. Becquerel] ,,383
Uber die Methode der Dampfung bei der Bestimmung des
Ohms. Art. 120. 1885 ,,415
Self-induction in Relation to Certain Experiments of Mr Wil-
loughby Smith and to the Determination of the Ohm.
Art. 123. 1885 ,,422
A Theorem relating to the Time-Moduli of Dissipative
Systems. Art. 125. 1885 ,,428
On the Thermodynamic Efficiency of the Thermopile. Art. 129.
1885 ,,438
On Professor Himstedt's Determination of the Ohm. Art. 131.
1886 ,,448
On the Clark Cell as a Standard of Electromotive Force.
Art. 132. 1886 ,,451
Testing Dynamos. Art. 133. 1886 ,,474
The Reaction upon the Driving-Point of a System executing
Forced Harmonic Oscillations of Various Periods, with
Applications to Electricity. Art. 134 1886 . . . „ 475
On the Self-Induction and Resistance of Straight Conductors.
Art. 135. 1886 ,,486
Notes on Electricity and Magnetism. I. On the Energy of
Magnetized Iron. Art. 139. 1886 „ 543
Notes on Electricity and Magnetism. II. The Self-Induction
and Resistance of Compound Conductors. Art. 140. 1886 „ 551
The Interrupters ........ „ 553
The Induction-Compensators ...... „ 555
Appendix. — The Induction-Compensators [p. 557] „ 577
Notes on Electricity and Magnetism. III. On the Behaviour
of Iron and Steel under the Operation of Feeble Magnetic
Forces. Art. 141. 1887 . 579
590 CONTENTS OF VOLUMES I. — IV.
PAGE
Is the Velocity of Light in an Electrolytic Liquid influenced
by an Electric Current in the Direction of Propagation ?
Art. 151. 1888 . Vol. III. 213
The Clark Standard Cell. Art. 165. 1890 . . . . „ 333
On Huygens's Gearing in Illustration of the Induction of
Electric Currents. Art. 171. 1890 . v"-- . . „ 376
On the Sensitiveness of the Bridge Method in its Application
to Periodic Electric Currents. Art. 180. 1891 . . . „ 452
On the Influence of Obstacles arranged in Rectangular Order
upon the Properties of a Medium. Art. 200. 1892 . Vol. IV. 19
On the Minimum Current audible in the Telephone. Art. 211.
1894. . . ., . . , 109
An Attempt at a Quantitative Theory of the Telephone.
Art. 212. 1894 . ". . . . .. . . „ 119
The Electrical Resistance of Alloys. Art. 221. 1896 . . „ 232
Observations on the Oxidation of Nitrogen Gas [by the
Electric Flame]. Art. 225. 1897 .... jtgfe „ 270
On the Passage of Electric Waves through Tubes, or the
Vibrations of Dielectric Cylinders. Art. 226. 1897 . „ 276
General Analytical Investigation ....... 276
Rectangular Section ......... 279
Circular Section .......... 280
On the Passage of Waves through Apertures in Plane Screens,
and Allied Problems. Art. 227. 1897 . 283
Perforated Screen.— Boundary Condition d<f>/dn = Q „ 284
Boundary Condition <£ = 0 . . . . . . . „ 286
Reflecting Plate.— d(f)/dn = 0 ........ 288
Reflecting Plate. — <£=0 ........ 289
Two-dimensional Vibrations . . . ; : . . . „ 290
Narrow Slit. — Boundary Condition d(f)/dn = 0 „ 291
Narrow Slit. — Boundary Condition $ = 0 . . . . „ 293
Reflecting Blade.— Boundary Condition d(f>/dn=.0 . . „ 294
Reflecting Blade. — Boundary Condition $ = 0 . . . . „ 295
Various Applications . . . ....'. . „ 295
On the Measurement of Alternate Currents by means of an
obliquely situated Galvanometer Needle, with a Method
of Determining the Angle of Lag. Art. 229. 1897 . „ 299
CONTENTS OF VOLUMES 1. — IV. 591
PAGE
On the Incidence of Aerial and Electric Waves upon Small
Obstacles in the Form of Ellipsoids or Elliptic Cylinders,
and on the Passage of Electric Waves through a Circular
Aperture in a Conducting Screen. Art. 230. 1897 . Vol. IV. 305
Obstacle in a Uniform Field ...... „ 306
In Two Dimensions ........ „ 309
Aerial Waves . . . . ... . . . „ 310
Waves in Two Dimensions ....... „ 314
Electrical Applications ........ „ 317
Electric Waves in Three Dimensions ...... 318
Obstacle in the Form of an Ellipsoid . . . . „ 323
Circular Aperture in Conducting Screen. „ 324
On the Propagation of Electric Waves along Cylindrical Con-
ductors of any Section. Art. 231. 1897 . 327
The Electro-Chemical Equivalent of Silver. Art. 232. 1897 . „ 332
Note on the Pressure of Radiation, showing an Apparent Failure
of the Usual Electromagnetic Equations. Art. 238. 1898 „ 354
Some Experiments with the Telephone. Art. 239. 1898 . „ 357
The Mutual Induction of Coaxial Helices. Art. 252. 1899 „ 431
On the Magnetic Rotation of Light and the Second Law of
Thermodynamics. Art. 271. 1901 „ 555
On the Induction-Coil. Art. 272. 1901 557
OPTICS.
Note on the Explanation of Coronas, as given in Verdet's
Legons d'Optique Physique, and other works. Art. 6.
1871 Vol. I. 76
Some Experiments on Colour. Art. 7. 1871 . . . „ 79
Yellow ... „ 85
On the Light from the Sky, its Polarization and Colour.
Art. 8. 1871 ,,87
Appendix .......... ,,96
On the Scattering of Light by Small Particles. Art. 9. 1871 „ 104
On Double Refraction. Art. 10. 1871 . . . . ,,111
On the Reflection of Light from Transparent Matter. Art. 11.
1871 120
592 CONTENTS OF VOLUMES I. — IV.
PAGE
On a Correction sometimes required in Curves professing to
represent the connexion between two Physical Magnitudes.
Art. 12. 1871. . .. . -; . :-... . . . Vol.1. 135
On the Reflection and Refraction of Light by Intensely Opaque
Matter. Art. 16. 1872 ,,141
Preliminary Note on the Reproduction of Diffraction-Gratings
by means of Photography. Art. 17. 1872 . . „ 157
On the Application of Photography to copy Diffraction -
Gratings. Art. 18. 1872 . . . . . . ,,160
On the Diffraction of Object-Glasses. Art. 19. 1872 . . „ 163
Note on a Natural Limit to the Sharpness of Spectral Lines.
Art. 23. 1873 , . ., 183
On the Manufacture and Theory of Diffraction-Gratings.
Art. 30. 1874 . . ,,199
Insects and the Colours of Flowers. Art. 31. 1874 . ... . „ 222
Investigations in Optics, with special reference to the Spectro-
scope. Art. 62. 1879, 1880 „ 415
Resolving, or Separating, Power of Optical Instruments . . „ 415
Rectangular Sections ,,418
Optical Power of Spectroscopes . . . . . . „ 423
Influence of Aberration ....... „ 428
On the Accuracy required in Optical Surfaces ... „ 436
The Aberration of Oblique Pencils ..... „ 440
Aberration of Lenses and Prisms ...... „ 444
The Design of Spectroscopes ...... „ 453
On Reflection of Vibrations at the Confines of two Media
between which the Transition is Gradual. Art. 63.
1880 '. . . . ,,460
On the Minimum Aberration of a Single Lens for Parallel
Rays. Art. 64. 1880 ,,466
On the Resolving-Power of Telescopes. Art. 67. 1880 . „ 488
On the Resultant of a large number of Vibrations of the same
Pitch and of arbitrary Phase. Art. 68. 1880 . . . „ 491
On Copying Diffraction-Gratings, and on some Phenomena
connected therewith. Art. 72. 1881 . . . „ 504
On Images formed without Reflection or Refraction. Art. 73.
1881 513
CONTENTS OF VOLUMES I. — IV.
593
PAGE
On the Electromagnetic Theory of Light. Art. 74. 1881 . Vol. I. 518
On the Velocity of Light. Art. 75. 1881 . . . . „ 537
On a Question in the Theory of Lighting. Art. 76. 1881 . „ 541
Experiments on Colour. Art. 77. 1881 . . 542
The Use of Telescopes on Dark Nights. Art. 82. 1882 . Vol. II. 92
On the Invisibility of Small Objects in a Bad Light. Art. 96.
1883 ,,187
Distribution of Energy in the Spectrum. Art. 99. 1883 . ,,198
On the Constant of Magnetic Rotation of Light in Bisulphide
of Carbon. Art. 118. 1885 ,,360
The Helix ,,367
Correction for Finite Length ...... „ 368
Appendix : Notes on Polarimetry in general .... „ 378
Postscript. [Work of H. Becquerel] . . . . . „ 383
Optics. Art. 119. 1884 ,,385
On the Theory of Illumination in a Fog. Art. 121. 1885 . ,,417
A Monochromatic Telescope, with Application to Photometry.
Art. 122. 1885 ,,420
On the Accuracy of Focus necessary for Sensibly Perfect
Definition. Art. 126. 1885 ,,430
On an Improved Apparatus for Christiansen's Experiment.
Art. 127. 1885 ,,433
Optical Comparison of Methods for Observing Small Rota-
tions. Art. 128. 1885 ,,436
On the Colours of Thin Plates. Art. 136. 1886 . . „ 498
Notes, chiefly Historical, on some Fundamental Propositions
in Optics. Art. 137. 1886 ...... ,,513
On the Intensity of Light Reflected from Certain Surfaces
at Nearly Perpendicular Incidence. Art. 138. 1886 . „ 522
Description of Apparatus ....... „ 525
Prism of Crown Glass (I) ,534
Prism of Crown Glass (II) ,537
Plate Glass Silvered Behind ,538
Silver-on-Glass Speculum ....... , 539
Mirror of Black Glass ,539
On the Maintenance of Vibrations by Forces of Double
Frequency, and on the Propagation of Waves through a
Medium endowed with a Periodic Structure. Art. 142.
1887
R. IV.
Vol. III. 1
38
594 CONTENTS OF VOLUMES I. — IV.
PAGE
On the Existence of Reflection when the Relative Refractive
Index is Unity. Art. 143. 1887 + .-. Vol. III. 15
Wave Theory of Light. Art. 148. 1888 .... ,,47
Plane Waves of Simple Type . ... . . . „ 49
Intensity ........*.. „ 51
Resultant of a Large Number of Vibrations of Arbitrary Phase „ 52
Propagation of Waves in General . . . . » ' . . „ 54
Waves Approximately Plane or Spherical . . . . „ 56
Interference Fringes . ... . , . . „ 59
Colours of Thin Plates . .; .'.".-. . . „ 63
Newton's Diffusion Rings . * » . • . . ,, 72
Huygens's Principle. Theory of Shadows . . . • „ 74
Fraunhofer's Diffraction Phenomena . . . . „ 79
Theory of Circular Aperture „ 87
Influence of Aberration. Optical Power of Instruments . . „ 100
Theory of Gratings . ...',. . .- . „ 106
Theory of Corrugated Waves . ... . . . _ „ 117
Talbot's Bands , " „ 123
Diffraction when the Source of Light is not Seen in Focus . „ 127
Diffraction Symmetrical about an Axis . . . . . „ 134
Polarization . . . . . . . .» „ 137
Interference of Polarized Light . „ 140
Double Refraction ,,148
Colours of Crystalline Plates . . . . . „ 156
Rotatory Polarization . . . • . , . . „ 159
Dynamical Theory of Diffraction ...... „ 163
The Diffraction of Light by Small Particles . . . . ,,170
Reflexion and Refraction ....... ,,176
Reflexion on the Elastic Solid Theory . . . . . ,,181
The Velocity of Light ,,187
On the Reflection of Light at a Twin Plane of a Crystal.
Art. 149. 1888 . . ..... . ,,190
Equations of a Dialectric Medium, of which the Magnetic Per-
meability is Unity throughout ...... „ 190
Isotropic Reflexion ........ ,,192
Propagation in a Crystal .... . . . . . „ 194
Reflexion at a Twin Plane ........ „ 194
Incidence in the Plane of Symmetry . . . . . „ 195
Plane of Incidence perpendicular to that of Symmetry . . „ 197
Doubly Refracting Power Small . . . . . . „ 200
Plate bounded by Surfaces parallel to Twin Plane . . , . „ 200
On the Remarkable Phenomenon of Crystalline Reflexion
described by Prof. Stokes. Art. 150. 1888 . . . „ 204
Is the Velocity of Light in an Electrolytic Liquid influenced
by an Electric Current in the Direction of Propagation ?
Art. 151. 1888 213
CONTENTS OF VOLUMES I. — IV. 595
PAOB
The History of the Doctrine of Radiant Energy. Art. 154.
1889 ..." Vol.111. 238
On the Limit to Interference when Light is Radiated from
Moving Molecules. Art. 157. 1889 , 258
Iridescent Crystals. Art. 158. 1889 . . . • . „ 264
On the Character of the Complete Radiation at a Given
Temperature. Art. 160. 1889 „ 268
On the Visibility of Faint Interference-Bands. Art. 161.
1889 ,,277
On Achromatic Interference-Bands. Art. 163. 1889 . . „ 288
Introduction ......... „ 288
Fresnel's Bands „ 289
Lloyd's Bands „ 292
Limit to Illumination ........ „ 294
Achromatic Interference-Bands ...... „ 296
Prism instead of Grating ....... „ 299
Airy's Theory of the White Centre „ 301
Thin Plates ,,303
Herschel's Bands . . . . . . . . „ 309
Effect of a Prism upon Newton's Rings . . . . . „ 311
Analytical Statement „ 314
Curved Interference-Bands „ 316
On Defective Colour Vision. Art. 173. 1890 . . . „ 380
Instantaneous Photographs of Water Jets. Art. 174. 1890 . „ 382
On Pin-Hole Photography. Art. 178. 1891 429
Some Applications of Photography. Art. 179. 1891 . . „ 441
On Reflexion from Liquid Surfaces in the Neighbourhood of
the Polarizing Angle. Art. 185. 1892 496
Postscript (October 11) „ 511
Aberration. Art. 189. 1892 ,,542
On the Intensity of Light reflected from Water and Mercury
at nearly Perpendicular Incidence. Art. 198. 1892 . Vol. IV. 3
Appendix. [Curvature due to Capillarity] „ 13
On the Interference Bands of Approximately Homogeneous
Light; in a Letter to Prof. A. Michelson. Art. 199.
1892 „ 15
On the Influence of Obstacles arranged in Rectangular Order
upon the Properties of a Medium. Art. 200. 1892 . „ 19
Interference Bands and their Applications. Art. 202. 1893 „ 54
596 CONTENTS OF VOLUMES I. — IV.
-
PAGE
On the Theory of Stellar Scintillation. Art. 203. 1893 . Vol. IV. 60
Astronomical Photography. Art. 204. 1893 .... „ 73
Grinding and Polishing of Glass Surfaces. Art. 205. 1893 . „ 74
On the Reflection of Sound or Light from a Corrugated
Surface. Art. 206. 1893 . . . • . . . . „ 75
On a Simple Interference Arrangement. Art. 207. 1893 . „ 76
On some Physical Properties of Argon and Helium. Art. 218.
1896 „ 215
The Kefractivity of Argon and Helium . . •, •-,.. •'•.', . „ 218
The Reproduction of Diffraction Gratings. Art. 220. 1896 . „ 226
On the Theory of Optical Images, with special reference to the
Microscope. Art. 222. 1896 . . . . ' . . „ 235
On an Optical Device for the Intensification of Photographic
Pictures. Art. 233. 1897 . ... . . „ 333
Rontgen Rays and Ordinary Light. Art. 237. 1898 . . ; . „ 353
On the Character of the Impurity found in Nitrogen Gas
derived from Urea. Art. 241. 1898 . . . . „ 361
Details of Refractometer „ 364
Transparency and Opacity. Art. 246. 1899. . . . „ 392
On the Transmission of Light through an Atmosphere con-
taining Small Particles in Suspension, and on the Origin
of the Blue of the Sky. Art. 247. 1899 . 397
The Interferometer. Art. 248. 1899 . .''.'. . „ 406
The Theory of Anomalous Dispersion. Art. 250. 1899 . „ 413
On the Law of Reciprocity in Diffuse Reflexion. Art. 258.
1900 '••'"• V • • • ,, 480
Remarks upon the Law of Complete Radiation. Art. 260. 1900 „ 483
On Approximately Simple Waves. Art. 261. 1900 . , „ 486
On the Stresses in Solid Bodies due to Unequal Heating, and
on the Double Refraction resulting therefrom. Art. 265.
1901 ,,502
Polish. Art. 268. 1901. ..' . v i » > -i „ 542
On the Magnetic Rotation of Light and the Second Law of
Thermodynamics. Art. 271. 1901. • . • .- . •-• ,- „ 555
CONTENTS OF VOLUMES I. — IV. 597
MISCELLANEOUS.
PAGE
On a Correction sometimes required in Curves professing to
represent the connexion between two Physical Magnitudes.
Art. 12. 1871 Vol. I. 135
A History of the Mathematical Theories of Attraction and
the Figure of the Earth from the time of Newton to that
of Laplace. By I. Todhunter, M.A., F.R.S. Two Volumes.
(London, Macmillan & Co., 1873.) Art. 29. 1874 . . „ 196
Insects and the Colours of Flowers. Art. 31. 1874 . . „ 222
Questions from Mathematical Tripos Examination for 1876.
Art. 41. 1876 ,,280
On Mr Venn's Explanation of a Gambling Paradox. Art. 50.
1877 ,,336
Uniformity of Rotation. [Phonic Wheel] Art. 56. 1878 . „ 355
Address to the Mathematical and Physical Science Section of
the British Association. Art. 86. 1882. . . . Vol.11. 118
On the Dark Plane which is formed over a Heated Wire in
Dusty Air. Art. 93. 1882 ,,151
Suggestions for facilitating the Use of a Delicate Balance.
Art. 104. 1883 ,,226
Presidential Address. [Montreal] Art. 113. 1884 . . „ 333
Professor Tait's " Properties of Matter." Art. 124. 1885 . „ 424
The History of the Doctrine of Radiant Energy. Art. 154.
1889 Vol. III. 238
Clerk-Maxwell's Papers. Art. 177. 1890 . . . „ 426
Experiments in Aerodynamics. [Review of Langley's] Art. 184.
1891 ,,491
The Scientific Work of Tyndall. Art. 209. 1894 . . Vol. IV. 94
The Theory of Solutions. Art. 224. 1897 .... ,,267
Liquid Air at one Operation. Art. 240. 1898 ... „ 360
Polish. Art. 268. 1901 ........ ,,542
Does Chemical Transformation influence Weight ? Art. 269.
1901. ,,549
38—3
INDEX OF NAMES.
Abbe, II 412, 519, IV 236, 239
Abney, II 346, 421, III 173, 439, IV 73
Adams, .T. C. Ill 2
Agamennone, III 236
Airy, G. I 166, 253, 255, 256, 261, 416, 417,
428, II 122, 501, III 61, 87, 90, 123, 180,
292, 301, 544
Airy, H. Ill 267
Aitken, II 154, III 358, 365, 368
Amagat, IV oil
Ampere, III 151
Andre, C. I 418, III 95
Angstrom, I 160
Appun, I 331
Arago, II 498, III 34, 102, 139, 140, 156,
159, IV 60, 69
Armstrong, Lord, IV 562
Auerbach, II 210
Austen, Boberts-, IV 549
Ayrton, II 469, IV 117, 144
Baines, III 267
Balfour, F. M. I 547
Balfour, G. W. I 548
Balmer, IV 345
Baly, IV 169, 514
Barrett, II 101
Barry, I 500
Bartoli, IV 354
Barus, III 569
Basset, III 389, 578, 593
Beattie, IV 557
Becquerel, H. II 338, 346, 361, 365, 377,
382, 383
Beer, I 123, 131
Beetz, I 372
Bell, C. Ill 382
Bell, G. I 501, II 288, 349
Bell, L. Ill 116
Bellati, I 313
Berthelot, III 399, IV 197
Bertrand, I 232
Bessel, I 140, 338, IV 549
Bichat, II 377
Bidone, I 377
Bidwell, II 341
Billet, I 77
Biot, III 102
Bohr, IV 513, 529, 530
Boltzmann, I 329, II 346, IV 125, 128, 354,
433, 444, 483
Borda, I 299
Bosanquet, I 230
Bosscha, I 354, II 291
Bottomley, IV 378
Bourdon, III 379
Boussinesq, I 271
Boys, III 382
Brewer, I 188
Brewster, I 210, 455, II 122, 240, 348, 396,
III 123, 138, 148, 159, 212, 266
Bridge, III 81
Briegleb, IV 140
Brillouin, II 448
Brough, IV 109
Briicke, I 99, HI 170, IV 102
Bryan, III 554, IV 433, 438
Buff, I 379, 393
Bunsen, IV 170
Burbury, III 555, IV 434
Burnside, III 555
Carhart, II 332, 473, III 333
Caron, IV 139, 140
Cauchy, I 111, 115, 122, 131, 141, 145,
150, 460, 522
Cavaille-Coll, I 320
Cavendish, IV 96, 136
Cayley, I 194, IV 27, 28
Cazin, II 280
Chaulnes, HI 72
Chladni, I 174, 351, II 212, III 319
Christiansen, I 142, II 433, III 15, IV 392
Christie, I 454, 455
Chrystal, I 310, II 157, 166, 168, 449
600
INDEX OF NAMES.
Clark, L. II 287, 340, 359, 451, III 333
Clarke, II 33
Clausiua, I 99, II 346, 521, III 170, 561,
IV 181, 491
Clerk Maxwell. See Maxwell.
Coddington, I 466
Common, IV 56, 542
Conroy, J. II 523, 539, IV 3
Cooke, III 37, 43, 236, 524, IV 52
Cornu, 1 537, II 347, 348, in 62, 112, 132,
303, 552
Cotes, II 513, III 56
Cotterill, III 538
Cottrell, I 308
Crafts, IV 51
Crookes, II 125, 340, 345, III 266, IV 159,
168, 170, 193, 198, 336
Cross, IV 118
Culverwell, IV 450
Czapski, IV 243
Dallinger, IV 236
Darwin, C. I 222, III 243
Darwin, G. II 344, 441, 593
Darwin, H. II 4
Davy, H. IV 96, 270
Dawes, I 416, in 92
De Coppet, II 461
De la Provostaye, I 143, 149
De La Rive, I 1
De La Rue, II 320, 340, IV 109
De Pontigny, I 411
De Vries, I 263
Debray, I 241
Delisle, HI 78
Desains, I 143, 149, HI 179
Deville, IV 139, 140
Dewar, II 301, 347, HI 354, 448, IV 232,
359, 461, 481
Dittmar, III 525
Donkin, I 177
Dora, II 415
Draper, H. I 207
Draper, J. W. HI 238
Drude, IH 497
Duff, W. IV 377
Dunkin, I 166
Dupre, in 346, 359, 364, 402, 412, 421, 422,
448, IV 416
Duprez, HI 570
Dvorak, I 342
Earnshaw, I 257
Ebert, III 258
Eiseulohr, I 123, 141, 146, 150, 522
Ellis, I 331, 333
Encke, I 194
Ettingshausen, II 332
Evans, M. II 459, III 334
Everett, I 82, 229, IV 63, 242
Ewing, II 543, 547, 579, 585, 587, 589,
IV 121
Exner, K. IV 72
Fabry, III 66
Faraday, I 351, II 193, 212, 239, 360,
III 161, 384
Ferranti, II 339
Ferraris, IV 109, 117
Ferrers, I 121, 338
Fitzgerald, G. I 518, III 132, IV 342
Fitzgerald, M. IV 477
Fizeau, I 537, III 60, 543, IV 59
Fleming, II 24, 37, 88, 458, 463, 467, 472,
IV 232
Forbes, G. I 537, II 348, 458, III 188
Forel, II 344
Foster, C. II 24, 88, 143
Foucault, I 164, 417, 488, 538, II 406,
HI 60, 384
Frankland, E. H 152, IV 271
Franklin, III 357
Fraunhofer, I 160, 416, 488, U 411, 519,
HI 79, 99
Fresnel, 1 112, 117, 120, 125, 218, 460, II 498,
HI 33, 50, 59, 127, 139, 140, 149, 156, 177,
183, 288, 544, IV 3
Froude, W. I 261, 290, 299, 322, 324, II 343,
HI 492, 494, IV 472
Fuchs, HI 407
Galton, F. I 308, 473, H 98
Gauss, III 400
Gautier, A. IV 496
Gernez, II 460, IV 420
Gerresheim, IV 141
Gerstner, I 261
Geuther, IV 140
Gibbs, W. I 540, H 342, III 189, 190, 359,
364, IV 37
Gilbert, HI 129
Gill, I 538
Giltay, I 313
Glaisher, J. W. H 263, III 2
Glazebrook, I 489, II 49, 57, 120, 137, 157,
164, 168, 229, 288, 362, 365, 414, 542,
HI 114, 154, 190
Gordon, J. E. H. II 238, 339, 361, 383
Gore, H 279
Gouy, III 263, 270, IV 353
INDEX OF NAMES.
601
Graham, IV 262
Gray, A. IV 255, 258
Gray, M. Ill 539
Green, I 90, 97, 111, 113, 121, 124, 126, 145,
218, 255, 460, III 140, 183, 186, 251
Greenhill, I 346
Griffiths, IV 332
Gripon, I 411
Grubb, I 454V III 56
Guthrie, I 267, 269
Hagen, II 125
Haidinger, IV 59
Hall, II 340
Hamilton, B. I 231
Hamilton, W. I 443, II 517, III 89, 155
Hampson, III 539, IV 360
Hanlon, I 297
Hansen, I 166
Harcourt, A. V. IV 104, 188
Hastings, III 154
Haughton, I 110, 123, 133
Haweis, III 327
Heaviside, II 484, 551, 572, III 452, 457,
459, 464, IV 327
Heine, I 339
Helmholtz, I 28, 33, 68, 97, 181, 287, 291,
298, 305, 319, 364, 417, 518, II 100, 122,
323, 342, 351, 379, 396, 412, 459, 463, 513,
516, 517, 521, III 116, 190, 277, 327,
IV 78, 202, 235, 243
Henderson, III 525
Henry, I 13, 305, 306, IV 101, 298
Herapath, III 560
Herschel, J. I 85, II 121, 405, 499, 508, 520,
HI 73, 90, 111, 161, 170, 240, 271, IV 544
Herscliel, W. I 416, 417, II 92, 411,
III 242, 309
Hertz, III 537, IV 321
Heydweiller, IV 549
Hicks, W. M. II 240, 343
Higgs, IV 73
Hilger, I 457
Hill, III 2, 4, 6, 7, 12
Himstedt, II 448
Hockin, II 79, 85, 237, 276, 469, IV 243
Hodgkinson, III 208
Hbek, III 547
Holden, II 92
Holman, IV 454, 482
Holmgren, III 380
Hopkinson, J. I 427, II 393, 459, 474, 543,
548, III 306, IV 503
Huffaker, IV 463
Huggins, II 347, III 547
Hughes, I 499, II 339, 349, 486, 551
Hunt, A. B. II 344
Hunt, B. Ill 239
Huygens, III 74, 77, 148, 376
Ibbetson, III 281, 284
Jacobi, IV 27
Jamin, I 120, 129, 141, 144, 152, 522,
HI 180, 496, 503, 511, IV 3
Japp, III 537
Jellet, III 162, 224
Jevons, II 200, IV 69
Jolly, v. IV 44, 45. 51
Jones, V. IV 431
Joule, II 280, III 561, IV 96
Kaiser, IV 422
Kayser, IV 345, 494
Keiser, III 233, 525
Kelland, I 255, III 82
Kelvin, I 2, 6, 16, 170, 228, 232, 294, 323,
325, 338, 346, 365, 474, II 10, 120, 218,
258, 266, 343, 351, 361, 475, 517, III 16,
17, 155, 162, 185, 241, 256, 343, 383, 401,
402, 413, 522, 554, 556, 577, 580, 582,
IV 204, 209, 342, 433, 450, 495, 540, 559
Kirchhoff, I 288, 291, 295, 299, 300, II 135,
222, 251, 513, 516, III 268, -t92, IV 377,
483, 494, 532, 537
Klingelfuss, IV 557
Knockeuhauer, III 128
Knowles, I 290
Kohlrausch, F. II 1, 47, 120, 126, 145, 237,
279, 280, 310, 340, IV 332
Kohlrausch, W. IV 298
Konig, B. I 331
Koosen, II 582
Korteweg, I 263, IV 78
Kundt, I 142, 156, II 239, 251, 338, 345,
IV 176, 337
Kurlbaurn, IV 485
Kurtz, I 122, 123
La Cour, I 356, II 8, 179
Lagrange, II 513, 515, IV 243
Lamb, I 475, II 442, 446, 571, III 250,
280, IV 20(5, 287, 294, 408
Langley, I 293, II 194, 199, 346, III 188,
238, 269, 276, 491, IV 468
Laplace, I 338, HI 397, 402, 417, 466, 515
Larmor, I 146
Laurent, III 162
Lecher, IV 330
Leconte, IV 99
602
INDEX OF XAMES.
Leduc, IV 51, 352
Lenard, I 393, III 392
Leslie, II 121
Linde, IH 539, IV 360
Lippich, I 134, II 378, HI 110, 163, 258
Lippmann, II 143, III 13
Lipschitz, in 44
Lissajous, H 218
Liveing, I 455, II 347
Lloyd, H. I 123, III 61, 155, 241, 292, 295,
297
Lodge, A. Ill 591
Lodge, O. I 497, II 154, 424, 443, IV 298
Lommel, I 166, III 34, 88, 90, 132, 134,
432, IV 207
Lorentz, I 518, 522, III 190, 470, 551,
IV 19
Lorenz, I 120, 130, 131, II 50, 120, 145,
155, 276, IV 19
Love, I 536, III 218, 227, 244, 280, 285,
IV 503
Lummer, IV 59
Lupton, IV 104
Macaulay, I 540
MacCullagh, I 111, 125, III 150
Macdonald, IV 330
Macdougall, IV 198, 223
Mach, IV 335
Madan, UI 211, 265
Magnus, I 143, 344, 379, 391, 393
Mallet, IV 140
Mallock, I 304, II 44, 348
Malus, III 139
Maquenne, IV 140
Marangoni, III 341, 358, 361, 364, 412, 448,
562
Mascart, II 126, 237, 280, 298, 310, 329, 414,
450, III 112, 289, 309, 548, IV 59, 60,
65, 307, 332, 364
Masson, I 144
Mather, IV 117
Mathews, IV 255, 258
Matthiessen, A. I 65, II 78, 85, 237, 276
Matthiessen, L. II 193, 212, III 384
Maxim, IV 475
Maxwell, I 1, 13, 43, 60, 79, 156, 168, 226,
235, 237, 276, 297, 471, 498, 518, U 11,
80, 99, 128, 170, 185, 228, 280, 281, 288,
290, 345, 346, 350, 396, 420, 480, 486, 492,
498, 561, 572, IU 49, 68, 190, 376, 380,
398, 401, 426, 470, 476, 517, 540, 554,
IV 32, 112, 304, 307, 397, 402, 413, 433,
491, 558
Mayer, A. M. I 331, 342, 468
Mayer, J. R. IV 96
McKichan, II 280
McLeod, I 331, 360, II 33
Mehler, III 45
Melde, II 190, III 1, IV 551
Melloni, in 242
Mendeleef, IV 202, 511
Meyer, 0. E. I 156, IV 222, 454
Michell, I 299
Michelson, A. I 538, U 348, III 60, 6<
188, 189, 213, 543, 549, IV 15, 59, 406
Michelson, W. Ill 268, 275
Mizuno, IV 557, 565
Moissan, IV 139
Montigny, IV 61, 66
Morley, E. W. Ill 189, 525, IV 352
Moulton, II 340
Mnir, m 11
Muirhead, II 473
Munro, I 85
Necker, III 134
Neumann, I 111, IV 503
Newall, IV 422
Newcomb, III 188
Newton, I 94, II 414, 498, 509, III 24, 65,
68, 70, 170, 289, 303, 311, 491, IV 96
Nicol, W. II 461
Niven, W. D. U 13, 54, III 426
Nobert, I 157, 160
Noyes, IH 525
Oberbeck, U 571, III 365, 367, IV 557
Obermayer, IV 482
Olszewski, IV 174
Costing, IV 552
Ouvrard, IV 139
Page, IV 118
Parkinson, I 466, II 414
Paschen, IV 483
Pasteur, III 161
Peal, II 194, m 267
Peirce, III 111
Penaud, HI 494, IV 472
Perot, III 66
Petzval, I 517, III 432, 451
Pickering, E. C. I 453, II 532
Place, LEI 70, 311
Planck, IV 483
Plateau, I 373, 388, 391, 395, II 110, 269,
348, UI 341, 360, 363, 370, 374, 585
Pochhammer, IV 276
Pockels, A. (Miss), HI 375, 572, 573, IV 425
Poincare, IV 434
Poisson, I 460, 472, II 498, UI 33, 78
INDEX OF NAMES.
603
Poynting, II 363, III 162
Preece, II 331, IV 109, 124
Prehlinger, IV 141
Preston, II 414, III 305, IV 406
Preyer, I 331, IV 298
Priestley, IV 137
Provostaye, III 179
Quincke, I 152, 155, 215, 387, 504, II 231,
236, III 111, 350, 367, 371, 383, 392, 412,
497, 562
Ramsay, III 472, IV 1, 130, 184, 192, 195,
215, 217, 222, 223, 224, 260, 265, 272, 351,
361, 481, 514
Randall, IV 222
Rankine, I 113, 261, 324
Reade, III 239
Regnanlt, III 37, 43
Reinold, II 349, 511, III 349, 425
Respighi, IV 61
Reusch, III 212, 266
Reynolds, 0. I 316, 323, 324, II 273, 344,
III 365, 575, IV 101, 298
Richards, III 37, 236, 524, IV 348
Richardson, IV 225
Ridout, I 500
Riess, I 354
Rijke, I 353, 408
Riley, II 231, 233
Robinson, I 454
Robison, III 491
Roiti, III 213
Rontgen, II 338
Rood, II 522, 533, III 179, 442
Roscoe, I 110
Rowland, II 1, 16, 47, 76, 120, 135, 339, 341,
347, 587, III 110, 113, 116, IV 73, 87
Rubens, I 144, III 189, IV 485
Riicker, II 349, 511, III 349, 425, IV 202
Rudberg, III 155
Rumford, IV 96
Runge, IV 345
Russel (Major), I 161
Russell, Scott, I 256, 261, II 258
Russell. IV 363
Rutherfurd, I 207, 458, 505
Rydberg, IV 346
Schoeder van der Kolk, II 91
Schulze, IV 222
Schuster, I 310, II 1, 20, 40, 43, 63, 75,
98, 340, 398, III 276, IV 170, 199, 346,
353, 369, 500
Schutzenberger, IV 139
Schwendler, III 452
Schwerd, III 82
Scott, A. II 302, III 37, 43, 234, IV 348
Seebeck, I 403
Sellmeyer, I 143, 156, IV 413
Shaw, II 83, 446
Shields, IV 184
Sidgwick, Mrs, II 43, 47, 63, 78, 103, 115,
155, 271, 273, 278, 471, 540, IV 332
Siemens, C. W. II 334
Siemens, Werner, II 78, 90, 91
Siljerstrom, IV 511
Simpson, I 205
Smith, A. Ill 75, 152, 169
Smith, F. J. IV 113
Smith, M. Ill 384
Smith, R. II 409, 513, III 56
Smith, W. II 422
Sommerfeld, III 163
Sondhauss, I 26, 36, 69, 351, III 342
Soret, III 78
Spottiswoode, I 458, II 340, IV 160
Stas, IV 133
Stefan, IV 354
Stephanelli, III 448
Stewart, B. Ill 241, 268, 552, IV 378, 483,
494
Stokes, I 50, 89, 91, 96, 99, 101, 113, 117,
141, 191, 220, 255, 257, 263, 322, 531, II 121,
241, 273, 344, 403, 419, 479, 498, III 49,
62, 66, 68, 72, 86, 89, 92, 123, 146, 154, 163,
179, 181, 204, 240, 272, 340, 569, 575, 588,
595, IV 78, 101, 209, 298, 321, 353, 376,
409, 540
Stoletow, II 339, 587
Stoney, II 198, IV 237
Strutt, R. J. IV 223
Struve, H. Ill 63, 92, 127
Stuart, II 9
Sumpner, IV 117, 144
Sutherland, IV 482, 514
Swan, III 148
Swinton, A. C. IV 562
Sylvester, III 428
Salmon, III 150
Sande Bakhuyzen, II 365
Savart, F. I 373, 388, 390, II 239, 269
Savart, N. I 403, 404, 405
Scheibler, I 331
Tait, I 170, 228, 232, 338, II 361, 424, 475,
517, III 256, 383, 465, 556, 593, IV 109,
124
Talbot, F. I 507, II 362, III 51, 69, 123,
134, 289, IV 545
Tate, IV 415
604
INDEX OF NAMES.
Taylor, S. II 240
Taylor, II 469
Thiesen, I 291
Thollon, I 456, 544, II 347, 552
Thompson, S. P. IV 353, 553
Thomson, J. J. I 518, 547, II 44, 449,
488, IV 276, 318, 323, 327, 353, 354. 503
Thomson, James, II 154, III 516, IV 63
Thomson, W. See Kelvin.
Thorpe, IV 130
Threlfall, I 547, II 458, IV 514
Todhunter, I 22, 196, 338, 492, IV 370
Tomliuson, C. Ill 347, 566, IV 420
Topler, I 328, II 406, IV 125, 128
Tower, II 344
Travers, IV 481
Tyndall, I 87, 101, 109, 305, 307, 316, 394,
531, 541, II 100, 151, 190, 220, 269,
III 1, 25, 170, IV 94, 379, 551, 554
Van der Mensbrugghe, III 347, 353, 412, 565
Van t' Hoff, IV 267
Venn, I 336
Verdet, *I 76, 164, 200, 417, 491, II 414;
in 99, 143, 146, 161, 291
Vince, I 293
Waals, van der, III 398, 421, 423, 465, 470,
471, 516
Walker, J. Ill 291
Walker, I 530
Walter, IV 557, 559, 565
Warburg, II 345, 544, IV 176, 337
Waterston, III 477, 558, IV 433
Watson, IV 434
Weber, H. F. II 1, 49, 120, III 63, 268, 275
Weber, W. II 120, 553
Wenham, III 492, IV 467
Wertheim, I 29, 36, 320
West, IV 363
Wheatstone, I 182
Whewell, III 243
Whitaker, IV 225
Whitehead, IV 87
Wiedemann, G. II 134
Wien, M. IV 110, 125
Wien, W. IV 354, 483, 555
Wild, II 279, 415
Williams, P. IV 146
Wohler, IV 139
Worthington, I 387, III 392, 398. 412
Wright, A. II 314, 323, 342, 454, 458, 459,
462, 472
Wright, H. R. IV 480
Wright, L. IV 237
Wiillner, I 152, IV 12
Young, J. I 537, II 348, III 188
Young, S. Ill 472
Young, T. I 460, II 235, 425, 498, 516,
III 72, 100, 111, 139, 177, 238, 271, 397,
400, 404, 414, 417, 419, 423, 544, IV %,
550
Zamminer, III 329
Zech, I 127
Zecher, III 213
Zollner, II 138, 141
CAMBRIDGE : PRINTED BY J. AND C. F. CLAY, AT THE UNIVERSITY PRESS.
QC3
R21s
Rajleigh
LIBRARY
Los Angeles
This book is DUE on the last date stamped below.
MAR 2 6 195&
MAR 1 7 REd
v \ 8
WOV 2 1959
JAN 2 3 RBI
OEC 2 1 19B2 .
INTERLIBRARY
OCT 2 2 1964
FOUR WEEKS FROM DATE
NON-RENEWABLE
\j c
HOY 1 8 RHH)
Form L9-100m-9,'52(A3105)444
FEB ? 1965
FEB 3 tert
STACK
JUL071987
ANNEX
RECEIVED
RECEIVE
JoAH^"1987
STACK ANNfX
OF RECEIPT
Engineering
Library ,J£ SOUTHERN REGIONAL LIBRARY FACILITY
Piip
*.