s

A REVISION

OF THE

Atomic Weights of Sodium and

Chlorine

BY

THEODORE WILLIAM RICHARDS

AND

ROGER CLARK WELLS

OF HARVARD UNIVERSITY

WASHINGTON, D. C:

Published by the Carnegie Institution of Washington

April, 1905

Carnegie Institution of Washington, Publication No. 28.

PRESS OF
HENRY E. WILKENS PRINTING CO.

WASHINGTON, D. C.

CONTENTS.

Introduction 5

Balance and weighing 10

Preparation of pure materials 12

Water 12

Hydrochloric acid 12

Sodic chloride 13

Silver 16

Lime 24

Utensils 25

The solubility of argentic chloride 25

The occlusion of dissolved substances by argentic chloride 31

The ratio of sodic to argentic chloride 35

The ratio of sodic chloride to silver 44

The fusion of sodic chloride in vacuum 55

The synthesis of argentic chloride 57

The new atomic weights of sodium and chlorine and their effect on other

atomic weights 67

Summary 69

3

a revision of the atomic weights of sodium

and Chlorine.

By Theodore William Richards and Roger Clark Wells.

INTRODUCTION.

The investigation of the atomic weights of the extremely common
elements sodium and chlorine was undertaken because of an unac-
countable discrepancy which appeared in the composition of a sample
of very pure sodic bromide, as compared with the results of Stas.
This sodic bromide had been prepared for the purpose of determining
the transition temperature of its hydrated crystals ; and a preparation
which yielded a constant transition point, therefore giving evidence of
great purity, nevertheless possessed a perceptibly lower combining
weight than that indicated by Stas's results.

Such a discrepancy as this was not to be passed over lightly. It
indicated either an unknown constant impurity in our sodic bromide
— and hence a possible error in our transition temperature — or else a
flaw in the classical work of Stas. When the first cause of disagree-
ment had been carefully sought in vain, the second was pursued.

Thus a physico-chemical investigation demanding great purity
of materials led to a quantitative research of unexpected magnitude ;
and in turn this quantitative investigation depended continually upon
physico-chemical methods and considerations. In confirmation of
previous work at Harvard, it was found that the physico-chemical
conditions of experiment were of as serious import as the purity of the
materials, and of far greater significance than an increase in the scale
of operations. The essential circumstances here defined must receive
consideration in any other research of a similar kind.

It is needless to say that an investigation seeking to test the ac-
curacy of Stas's work is a serious undertaking, for undoubtedly much
of his work deserves a rank among the most accurate of quantitative
chemical results ; certainly he distanced all those who preceded him.

5

•

6 ATOMIC WEIGHTS OF SODIUM AND CHLORINE.

Nevertheless, he was by no means infallible, as his long oversight of
the solubility of argentic chloride, his difficulty in obtaining colorless
amnionic bromide, the uncertainty concerning the oxygen in his silver,
and the recent proof of an important error in his work on iodine,*
all testify. Moreover, Stas's work is not to be considered as accurate
in proportion to the scale of his experiments, as many have been
inclined to suppose. Mere increase in quantity of material can
avail nothing if the same physico-chemical error or the same chemi-
cal impurity contaminates the whole. Indeed, it is doubtful if beyond
a moderate scale of operations increase of quantity decreases even
the probability of accidental error. For example, the extreme varia-
tions in his work on the synthesis of argentic chloride, in which
on the average nearly 200 grams of the compound were weighed
in each analysis, amounted to 0.006 per cent, or 12 milligrams of
material.! An accuracy of 0.6 milligram with 10 grams of sub-
stance, very easy to attain as far as the collection and weighing of
the material is concerned, would have brought him as good agreement.
But large quantities have no advantage to offer besides the increased
accuracy in weighing and the diminished percentage effect of a given
accidental loss of material in transference. For this reason the use
of extremely large quantities must rather be considered as evidence
of a lack of sound judgment than an index of unusual accuracy. Stas
himself acknowledged this by using only about 10 grams of material
in each of his last experiments.!

Regretfully one must admit that a critical study of the work of
Stas upon sodium in particular reveals several possibilities of error, as
well as at least one inconsistency. He analyzed both the chloride and
the bromide ; of these salts the former received the greater attention,
and will be discussed here in greater detail.

In reviewing his results upon sodic chloride there are two series
to be considered, separated by an interval of time of over twenty years.
Although at first sight these two series seem to have yielded almost
identical results, asi a matter of fact they are essentially inconsistent
with one another. Both results were obtained by the method of Gay-
L-ussac, titrating weighed amounts of silver in solution with salt ; but
at the earlier time the latter was added until all precipitation ceased,
while at the later time equivalence was assumed when the supernatant
liquid gave equal opalescence with excess of either chloride or silver.

*The work of Ladenburg and Scott has been confirmed by Dr. G. P. Baxter,
of Harvard University, whose recent analyses show that the atomic weight of
iodine can hardly be lower than 126.97. Proc. Acad., 40, 419 (1904).

tj. S. Stas, Oeuvres completes 1, 341 (Brussels, 1894). JOeuvres, 3, 503.

POSSIBLE ERRORS OF STAS.

Anyone with experience will recognize that a much higher
apparent atomic weight of sodium was to be calculated by the first
procedure than by the second, other conditions being the same, and
yet the results were almost identical. The immediate inference is, of
course, that some other condition or conditions must have varied in
order to compensate for the effect of the altered method of titration .
This inconsistency, which appears also in his work on potassium and
ammonium, has never been explained, and one of the functions of the
present work was to account for it.

Among the possible sources of error in Stas's work, the most serious
seems to have been the practice of dropping solid salt into the solution
of argentic nitrate.* Of course this practice caused the greater part
of the argentic chloride to be precipitated in the presence of a
concentrated solution of salt, immediately around the solid — a circum-
stance tending greatly to promote the occlusion or inclusion of sodic
chloride in the precipitate. Stas, in his early work especially, was
fully awake to the danger of the occlusion of argentic nitrate in this
precipitate ; f but he seems to have had no fear of occlusion of other
salts. In the sequel we shall show that the chloride of silver tends to
carry down with it many salts from their concentrated solution, and
that the details of treatment determine whether or not these impurities
may be removed by washing.

Another possible cause of the discrepancy is to be found in the
methods used for preparing the materials. In Stas's earlier investiga-
tion, where the analytical end-point was erroneous, greater care was
taken in preparing the sodic chloride than in the later experiments.
In general, when purifying this salt, he employed violent treatment,
and very rarely recrystallized or in any way fractioned his material
afterwards. His favorite method was fusion with ammonic chloride
and ammonic chlorplatinate. Presumably the ammonic chloride was
used to expel other halogens, and the deposited platinum, Stas believed,
carried down silica and alumina with it. Except for these possible
advantages, this admixture of foreign materials seems to be a doubtful
expedient. In the ten experiments of i86o,| six different preparations
were made, from the following sources:

i. Sodic carbonate neutralized with hydrochloric acid.

2. Rock salt recrystallized.

3. Sodic sulphate repeatedly fused with ammonic chloride.

4. Sodic tartrate recrystallized and fused with ammonic chloride.

5. Sodic nitrate, similarly treated.

6. Sodic chlorplatinate recrystallized and fused.

tOeuvres, 1,337

*Oeuvres, 1, 759; also 3, 479.

JOeuvres, 1, 365.

8 ATOMIC WEIGHTS OF SODIUM AND CHLORINE.

The final step in every case consisted, as usual, in mixing the
salt with amnionic chloride and ammonic chlorplatinate, fusing until
colorless, and decanting the fused mass from the residue. The ten
results agreed almost exactly, indicating that the preparations were,
at least, identical ; but, of course, the results are all in error, because
of the erroneous end-point.

Only two preparations were used in the four experiments of
1881.* For one of these, impure sodic bicarbonate was neutralized
with hydrochloric acid. The salt, after three recrystallizations, was
repeatedly fused with ammonic chloride and chlorplatinate. f By
evaporating a solution of the resulting salt with chlorplatinic acid, it
was converted into sodic chlorplatinate without fractionation of any
kind. About nine-tenths of this material was dissolved in water.
Stas assumed that any potassic salt would not then dissolve ; but this
possibility is by no means excluded. The first lot of crystals from
the solution was decomposed and fused, and the resulting sodic chloride
used in the first two experiments. A second and third lot were also
obtained from the mother liquor of the first lot. The salt made from
the third lot was employed in the fourth experiment of 1881. Thus
there were only four crystallizations between the starting point (which
was admittedly impure) and the final product; and the first three of
these crystallizations were of common salt, isomorphous with potassic
chloride. In the third experiment of 1881 , the salt was made by neu-
tralizing pure acid sodic carbonate and repeatedly fusing with ammonic
chloride in platinum , with no fractionation at all . These specimens were
probably less pure than the earlier ones.

Thus it appears that every preparation of Stas was fused in con-
tact with platinum, which was usually in a finely divided "nascent "
condition, or at least in the presence of decomposing ammonic chlo-
ride. The question is therefore important whether or not platinum can
dissolve in melted salt at the temperature of 8oo°. We have found
that fused salt when pure has but a very slight dissolving effect upon
platinum, but in the presence of ammonic chloride or hydrochloric
acid at a red heat in the presence of air a noticeable quantity of plati-
num is dissolved, as is indicated by the loss of weight of the contain-
ing crucible. It seems likely, then, that Stas's material contained
traces of platinum, either in a very finely divided state or possibly as
a platinum salt ; but the amount could hardly have been large. Stas
invariably found non-volatile residues in his preparations, consisting
of silicates of sodium and calcium, which he discovered by vaporizing
the salt. These residues usually amounted to about 0.004 Per cent.

*Oeuvres, 1,755. tOeuvres, 1,685.

STAS'S SILVER.

It is not impossible, however, that yet other impurities might
have volatilized with the sodic chloride. In volatilizing his salts, Stas
used a boat of pure platinum.* It is possible that pure platinum
would itself have volatilized appreciably at the temperature required
to drive off 10 grams of salt in half an hour.f For either of the two
reasons just given, Stas's correction for impurity in his salt would
not have been large enough.

These impurities must, however, have been almost equally pres-
ent in the early and later samples, and could not explain the serious
discrepancy between them. At most, 'however, as we shall show,
they could hardly have exceeded o.oi per cent, an amount much
smaller than the discrepancy in question. Our own experience shows
that common salt is a substance very easy to prepare in a pure state.
Clearly the cause of the discrepancy must be sought elsewhere.

The other essential substance, to be prepared in a pure state, was
silver. This Stas prepared in a variety of ways, as is well known ;
but for all this work the silver was either cast in ingots from under an
oxidizing flux, or else ' ' granulated ' ' by dropping into water. His cri-
terion of pure silver was the melting of the surface of a button without
any apparent irregular expansion as it liquefied, or any flame-color,
and the absence of all specks and spots from the completely melted
globule. In all experiments it was reddened in a silver crucible before
weighing.

After Dumas had recommended the fusion of silver in a vacuum,
Stas made many experiments upon the occlusion of gases by silver.
He carefully investigated the samples of silver used in his atomic
weight researches, and also new preparations. He found that fusion
in a flux of sodic nitrate introduced a quantity of oxygen, and that
bars and blocks were slightly purer than ' ' granulated ' ' silver. In no
case was the amount of retained gas found by Stas enough to affect
seriously even a very accurate analysis ; but, as will be shown later,
it is probable that he did not find all the oxygen present. An error
in Stas's silver would not account for the discrepancy between his results
on sodium and ours ; for an impurity in the silver would cause his
atomic weight of sodium to appear too low, and not too high. On
the other hand, the impurity in Stas's silver is undoubtedly the cause of
the difference between his atomic weight of chlorine and ours.

*Oeuvres, 1 , 681.

tSee Hall, J. Am. Chem. Soc, 22, 494 (1900) ; Richards and Archibald,
Proc. Am. Acad., 38, 460 (1903); Hulet and Berger, J. Am. Chem. Soc, 26,
1512 (1904). There seems to be no doubt that platinum is slightly volatile at
1,000° in the presence of air. Probably it is not in pure nitrogen.

IO ATOMIC WEIGHTS OF SODIUM AND CHLORINE.

From these considerations it was clear that the reason for the
incompatibility of the results on sodium was probably to be sought
chiefly in the method of analysis rather than in the impurity of the
materials. The long-continued search for the causes of discrepancy
led finally to a satisfactory explanation of the whole anomaly. In
order to make this clear, the details of our own investigation must
be recounted.

BALANCE AND WEIGHING.

The Troemner balance which has served in many similar researches
was used during most of the present work.* Successive weighings on
it of the same object rarely differed as much as 0.03 milligram. For
a few of the conclusive final experiments a yet more sensitive and
perfect balance, especially made for this kind of work by the same
maker, was used.

The Sartorius weights were standardized from time to time by the
method devised by one of us.f It was found as usual that the larger
gold-plated and platinum-plated brass weights very slightly changed
from time to time, although the smaller platinum weights did not.
All weighings were made by substitution. Furthermore, the major
portion of the substituted tare consisted of a counterpoise exactly sim-
ilar to the crucible, tube, or boat which was being weighed. Thus the
weights required were never large in amount and errors due to changes
in meteorological conditions were avoided. The method of recording
and checking the weights is given in another place .$

While these obvious particulars are given here for the sake of com-
pleteness, it must be remembered that accuracy of weighing is easily
attained. The great errors in chemical quantitative work are usually
not in the weighing, but are due rather to impurities in the substances
weighed or incompleteness or irregularity of reaction. Only in a few
cases, with large vessels, was it necessary to use a telescope for reading
the vibrations, the observer being in another room and looking through
an intervening glass door.

The balance-room devoted to these most accurate determinations
was kept at a very constant temperature, being wholly inside a private
laboratory whose air temperature was regulated by a delicate thermo-
static attachment to its heating apparatus. The balance-room itself
had neither heating apparatus nor outside windows. This question
of constancy of temperature is of the most vital importance in the weigh-
ing of large vessels.

*Proc. Am. Acad., 26, 242 (1891).

tRichards, J. Am. Cbem. Soc, 22,144 (1900).

{Richards, Proc. Am. Acad., 31, 175 (1895); Z. Anorg. Chem.,10, 19 (1895).

DETAILS OF WEIGHING.

II

In reducing weights in air to weights in vacuum, the densities at
about 2o° were assumed to be as given below. The value for silver
proceeds from Stas, and was checked by one determination by us of
the purest silver (sample W, described later). The value for the
brass weights refers to the gold-plated Sartorius weights used in all
but a very few of the experiments ; the other set had a slightly higher
density, of which account was taken. The correction for 1.000 gram
of substance weighed in dry air at 200 and 760 mm. is stated for each
substance.

Substance.

Density at 200.

Correction for one
gram of substance,
20°, 760 mm.

Silver

IO49
2.14
5-56*
8.40

— O.OOOO29
+ .OOO418
+ .OOOO73

Sodic chloride. . . .
Silver chloride. . . .
Weights

Especially when sodic chloride or silver chloride was weighed the
temperature and barometric pressure were recorded at the time of
weighing. In case of significant variations of temperature or pressure
the corrections were changed accordingly. The correction is com-
puted as follows :

Correction = w (volume of i gram of substance — volume i gram
of weights), where w is the weight of i cubic centimeter of air at the
given temperature and pressure. The following small table contains
a sufficiently accurate statement of this value in convenient form for
use with a logarithmic slide rule.

Value of w, in grains.

Temperature! (°C) .

750 mm.

760 mm.

670 mm.

140

O.OOI214
.OOI205
.OOII97
.001189
.OoilSl
.001173

O.OOI230
.OOI22I
.OOIII3
.001204
.OOII96
.OOI188

O.OOI246
.001237
.001229
.OOI220
.OOI2I2
.001204

160

180

20°

22°

240

If the air is completely saturated with water vapor the above
values should be decreased by 0.000008 at 140, 0.000010 at 200, and
0.000013 at 240. But since the air of our balance cases was certainly
less than a quarter saturated with moisture, any correction on this
account would have been supererogatory.

*Richards and Stull, in an investigation as yet unpublished.

12 ATOMIC WEIGHTS OF SODIUM AND CHLORINE.

PREPARATION OF PURE MATERIALS.

WATER.

The traces of organic impurity in the distilled water of the labor-
atory were eliminated by redistilling from a weakly alkaline solu-
tion of potassic permanganate, using at first a glass condenser and
rejecting the first quarter of the distillate. Water thus prepared was
pure enough for preliminary nephelometric work, when the presence
of a trace of alkali could do no harm ; it was usually kept in Jena
flasks. In cases where the greatest purity was desired it was distilled,
of course, once further, and for the nephelometric experiments it was
kept in a closed bottle provided with a siphon and protected from pos-
sible traces of hydrochloric acid by a tube containing sodic hydroxide.
For final preparations and accurate analyses, where silica and alkali
must be excluded, the water was condensed and collected during the
third distillation wholly in platinum. Care was taken to exclude dust
in all stages of the proceedings; for, although visible dust might
weigh little, its presence in silver halogen salt would produce decom-
position and loss of weight in the final fusion. Only one who has
attempted to exclude all dust from solutions can appreciate the diffi-
culty of doing so. Thorpe and Rodger,* working upon viscosities,
found that sealed-in condensers were very helpful in excluding dust
from their distillates. We did not go as far as this, but the danger was
borne in mind at every stage of the work and guarded against.

HYDROCHLORIC ACID.

For the final work the purest acid of commerce was treated with
a few crystals of potassic permanganate, diluted, and boiled. This
should have expelled bromine and iodine, and must have oxidized
any trace of organic matter which might have been present. The
oxidation and boiling were repeated ; the solution was fractionally
distilled; and for the succeeding treatment the middle fraction was
selected and subjected to another distillation, in which a platinum con-
denser and receiver were employed. As a matter of fact, most of these
precautions were found to be unnecessary, since salt made, as described
on page 16, from the best "chemically" pure acid of commerce only
once redistilled in platinum, without any further precautions, gave the
same results as the purest material.

♦Phil. Trans., 185 A, 432 (1894).

PURIFICATION OF SALT. 13

SODIC CHLORIDE.

In most cases a perfectly pure salt can not be made by precipi-
tating the impurities, as many are prone to assume, but must itself be
precipitated or crystallized from the less pure solution. Sodic chlo-
ride is nearly insoluble in a concentrated solution of hydrochloric acid ;
therefore the usual method of precipitation by means of hydrochloric
gas seemed the best possible method of obtaining pure sodic chlo-
ride, since any included hydrochloric acid could be completely expelled
by fusion. This expectation was confirmed by trial, and when it had
been found that this means of purification was unusually suitable, most
of the specimens of salt were treated in this way. This and other
purifications were greatly assisted by centrifugal draining of the crys-
tals. For this purpose salt was collected in a platinum funnel, and
this was covered and supported on a stout frame capable of being
whirled by means of strong cord about a radius of a meter. A test-tube,
firmly fixed in the frame, collected the separated mother liquor. Cen-
trifugal draining, so important in technical work, has hardly received
the attention which it deserves in the scientific laboratory. Experiments
have shown that a good centrifugal apparatus is capable of separating
nine-tenths or more , according to the habit of the crystals , of the mother
liquor which would otherwise contaminate the solid, hence making the
purification at least ten times as effective as it would otherwise be.*
The whole treatment of precipitation and draining'^was carried on in
platinum vessels which had been effectively freed from superficial iron.

Besides this chief mode of purification, others were used in special
cases, and the original material came from a variety of sources, as
recorded below. One of these was Merck's purest sodic chloride,
which this firm stated had been prepared from German rock salt.
Another was the very pure sodic sulphate prepared for the experiments
on the transition temperature of this substance, and known to be very
pure, because of the constancy and accuracy of its transition point
(32.383°).! Another was a fine specimen of Stassfurt halite, perfectly
clear and colorless. A fourth specimen was made from the purest
sodic bicarbonate manufactured by the Solvay Process Company, of
Syracuse, New York, and very kindly furnished in large quantity by
that company. A fifth sample was made from pure sodic carbonate
from Merck. These preparations, differing widely in the steps of
manufacture, and in geographical source, all yielded essentially the
same atomic weight.

*Richards, Journ. Am. Chem. Soc.,27, 104 (1905).

fRiohards and Wells, Proc, Am. Acad., 38, 431 (1902) ; also Zeitschr. phys.
Chem., 43, 465 (1903).

14 ATOMIC WEIGHTS OF SODIUM AND CHLORINE.

The source of the materials having been given above, it remains
to record the details of their preparation.

In the first place, a large quantity of Merck's sodic chloride, which
had been purified from German rock salt, was thrice reprecipitated by-
hydrochloric acid. This salt is designated "sample A" below, and
was used in preliminary experiments. It was further crystallized from
water (B) ; some of this was fused with ammonic chloride and am-
nionic chlorplatinate (C) ; a portion was then precipitated with hydro-
chloric acid (D) ; and finally some was again fractionated from water
(E). None of these operations seemed to have any essential effect
upon the combining weight of the salt ; hence, even Sample A must
have been practically pure.

The very pure sodic sulphate was easily converted into chloride
by successive precipitation from solution by means of gaseous hydro-
chloric acid. The sulphate forms an especially fortunate source of
pure sodium, since none of the other alkaline sulphates yield crystals of
the same degree of hydration or the same crystal form. Hence sodic
sulphate which has been many times recrystallized must be extremely
pure. The ease with which a constant transition temperature may be
obtained is evidence of this.

A specimen of sodic sulphate, which had been shown to be pure
by its transition temperature, was twice further recrystallized before
adding hydrochloric acid. Four precipitations by this gas, with wash-
ings and drainings, removed practically all of the sulphuric acid.
Nevertheless, when the mother liquor was concentrated by evapora-
tion and tested, an extremely small trace of sulphate was found by
excess of baric salt to persist until the tenth precipitation. There
was no trace of visible baric sulphate on testing the mother liquor,
even in the nephelometer, after the eleventh precipitation, but the
salt was precipitated once more for certainty. This salt, thus twelve
times separated by hydrochloric acid, was designated F. It seems
highly probable that since so large an amount of sulphuric acid was
eliminated from it by this treatment, all other impurities must have
been separated.

For another specimen of salt, a number of fine, large, colorless
transparent crystals of halite, from near Stassfurt, Germany, were
dissolved in pure water. After the addition of a little sodic hydroxide,
boiling, and standing, the solution was filtered and precipitated with
hydrochloric gas. A solution of the latter salt in water stood several
days. The clear decanted portion was precipitated twice more, frac-
tionally crystallized from water, and dried over sodic hydroxide in a
vacuum desiccator. This specimen was called sample G.

PURIFICATION OF SALT. 1 5

The bicarbonate from Syracuse, New York, was of the following
composition, according to the report of the Solvay Process Company:

Per cent.

Silica 0.006

Ferric oxide and alumina 002

Calcic carbonate 030

Magnesic carbonate 005

Sodic chloride 070

Sodic carbonate 522

Sodic bicarbonate 99-300

This material, already very pure for a commercial substance, was
easily further purified by repeated washing with cold water, which was
sufficient to remove most of the chloride.

A portion of the Syracuse bicarbonate, thus washed four times with
cold water and thoroughly drained each time with a reverse platinum
filter, was nearly all dissolved in water. After settling, the result-
ing clear solution was decanted, neutralized with hydrochloric acid,
and then evaporated to dryness. The sodic chloride was fused in
platinum, dissolved, and considerably diluted. After long standing
the clear upper portion was decanted from very finely divided silica.
This method of fusion is a very effective means of separating silica,
but the precipitate is so fine as to require a long time to settle, and as
to run through most filtering septa. Hence cautious decanting, with
rejection of the considerable lower portion of the solution, is neces-
sary. The unimpeachable solution of salt thus obtained was concen-
trated and twice precipitated by hydrochloric gas (H), twice more
precipitated (I), and yet twice more precipitated (J). Analysis showed
these three to be alike, as will be recorded.

The Merck bicarbonate was thoroughly washed and once recrys-
tallized from water. It was then gently ignited, and the normal car-
bonate was three times crystallized from water. From this carbonate,
by the usual process of precipitation by hydrochloric gas and crystal-
lization from water, sample K was prepared. This sample gave
essentially the same combining number as the others.

With such conclusive evidence that no foreign base was present,
we turned our attention to the acid. The preceding preparations had
all been precipitated with hydrochloric acid gas, obtained simply by
warming the highly concentrated purest acid of commerce, because
we deemed it more convenient to use this constant material, even if
slightly impure, than to prepare so large a quantity of the purest acid.
A special set of comparative experiments, made with the purest acid,
proved that the other material had, as a matter of fact, been pure
enough for the most exacting requirements.

l6 ATOMIC WEIGHTS OF SODIUM AND CHLORINE.

For this purpose the very pure hydrochloric acid was used whose
preparation is described above (p. 12). This acid was neutralized
by an especially prepared sample of soda, made from a second por-
tion of the Syracuse bicarbonate. The purification was conducted
wholly in platinum vessels, of course. After being washed as before,
the bicarbonate was dried and heated, to convert it into the normal
carbonate. The solution of the latter in water was decanted daily as
long as any precipitated residue (calcic carbonate) appeared visible.
The sodic carbonate was then recrystallized four times, as the deka- or
hepta-hydrate, and once as the anhydrous salt. A concentrated
solution of the latter salt was then treated for a considerable time with
carbon dioxide under a bell jar until the bicarbonate had formed.
The carbon dioxide was made by gently heating sodic bicarbonate,
and was well washed with pure water. The pure bicarbonate thus
formed was washed with cold water, thoroughly drained, and dissolved
in an excess of the pure hydrochloric acid. The resulting salt was
twice fractionally recrystallized from water on the steam bath ; and the
crystals were washed and drained each time, in order to free them from
any possible calcium or magnesium chlorides, which are not easily
eliminated from carbonate. Since the atomic weight from this prepa-
ration (I,) was identical with that from the preceding preparations, the
previous salts, made by precipitation with hydrochloric gas, could have
contained no other halogen than chlorine.

It appears from the identity of all these specimens that sodic
chloride is among the substances whose preparation in a pure state is
an easy problem. It is further true that the expulsion of water from
it upon fusion without loss of halogen is a very simple matter ; more-
over, since we succeeded in proving that salt fused in a vacuum
possesses the same combining weight as that fused in air, the solid
contains neither occluded oxygen nor nitrogen in weighable amounts.
This matter is discussed further on page 55.

Considering these experiments, it is fairly safe to infer that all the
various preparations of sodic chloride were quite free from significant
quantities of any foreign substance.

SILVER.

Recrystallized silver nitrate served as the starting point for the
first specimens, and the already purified residues from the early experi-
ments were again converted into pure silver for later use. Many
samples were prepared from different sources.

The silver used in all the experiments on sodic chloride was at
first purified by precipitation as chloride and then recovered by reduc-

PURIFICATION OF SILVER. 1 7

tion of the chloride with invert sugar* and sodic hydroxide. While
it is true that this process is not the best for the final treatment of the
silver, the precipitation as chloride is an excellent step in the prelimi-
nary purification. The invert sugar was made from a filtered solution
of "rock sugar." No metals were found in it by electrolytic, treat-
ment. The sodic hydroxide used, after settling and decantation, had
been electrolyzed until iron was completely removed. This slow but
effective way of removing foreign metals was hastened by using a
rotating cathode, while a fixed anode stirred the solution. The
argentic chloride was precipitated from a somewhat dilute solution, to
avoid the occlusion of impurities, and must be thoroughly washed. It
is best to wash at first with cold water, in order to avoid the contrac-
tion of the precipitate. The reduction was carried out in a silver
dish to avoid the introduction of silicates. This precaution is not
needed when electrolytic purification is to follow. In some cases,
where the original material was not wholly free from suspicion, the
precipitation as chloride and reduction were repeated. After reduc-
tion, the washed silver was fused upon either sugar charcoal or pure
lime, before the blast lamp, whose nozzle was scrupulously cleaned.
If the large buttons of metal thus formed are not kept hot too long
and are cooled in a reducing flame, they are already extremely pure.f
For ordinary work, or even for work on atomic weights of usual
accuracy, its purity suffices. Nevertheless, it can not be considered as
perfectly pure, for it must contain, perhaps, o.ooi per cent of sulphur
and variable traces of carbon from the illuminating gas (or else oxy-
gen from the air), as well as an occasional trace of argentic chloride,
arising from incomplete reduction. When thus cooled it is likely to
contain numerous minute cells inclosing gas.

As one stage in the purification the electrolytic method used by
J. L,. Hoskyns Abrahall has many advantages. | The almost pure
material just described is made the anode of a cell containing a con-
centrated solution of argentic nitrate prepared from the same silver.
The cathode is a pure silver wire, upon which is deposited by a
properly regulated current beautiful crystals of electrolytic silver.
The finely crystalline powder which falls from the anode § may be easily

*According to Stas, saccharose at ioo° would have answered as well as invert
sugar at 6o°. The saccharose is, of course, inverted by the strong hot alkali.
Oeuvres, 3, 13.

t Richards, Proc. Am. Acad.,29, 65 (1893; Z. Anorg. Chem., 6, 98 (1893).
See also Proc. Am. Aoad.,38,450 (1903). This work was confirmed in the latter
part of the present investigation.

J J. Oh. Soc. Trans., 61, 660 (1892).

^Richards and Heimrod, Proc. Am. Acad., 37, 415 (1902).

l8 ATOMIC WEIGHTS OF SODIUM AND CHLORINE.

collected separately by placing the anode button in a watch glass or
low dish, wholly submerged in the electrolyte. In order to exclude
every chance of contamination, no metal but silver may be allowed
to come in contact with the electrolyte, although the danger from an
immersed platinum wire, either at cathode or anode, is obviously
slight. Three samples of silver were thus prepared, called respectively
M, N, and P. This silver must have been free from every conceivable
impurity, except about 0.02 to 0.05 per cent of the mother liquor from
which it was deposited.* This important impurity, consisting wholly
of water and argentic nitrate, can be eliminated only by fusing the
silver, a matter which will soon be discussed.

Besides this very pure material used in the work on sodium,
several new samples of at least equal purity were made for the even
more critical work on chlorine. Three of these new preparations
were made from thrice crystallized pure argentic nitrate. A portion
of this nitrate was converted into argentic chloride and reduced by
pure invert sugar and purified sodic hydroxide. A second portion
was reduced by the invert sugar and sodic hydroxide. A third por-
tion was reduced by ammonic formiate and amnionic acetate as
recommended by Stas. f Some of each of these lots was deposited by
electrolysis in a concentrated silver nitrate bath in the usual way
(S, T, U, respectively).

Not content with these preparations — although all appeared to
be equally pure — and anxious to leave no stone unturned whose lift-
ing might disclose some new source of error, we made yet two more
preparations with new precautions. For one of these, through the
great kindness of Mr. Richard Pearce, of the Boston and Colorado
Smelting Works at Argo, Colorado, a specimen of metal was secured
whose source and history were known as far back as possible. Below
is given Mr. Pearce's statement concerning it :

This specimen is the product of the treatment of ores mainly from Colorado,
in which it occurs associated with quartz, barite, calcite, pyrite, sphalerite, with
more or less copper in the form of chalcopyrite, together with small quantities
of arsenic, antimony, lead, bismuth, and tellurium. The silver minerals are
chiefly argentite, polybasite, pyrargyrite, proustite, tetrahedrite, stephanite, and
occasionally cerargyrite. A small percentage of the silver comes in copper mattes
purchased in other districts, more particularly Montana, where the silver ores
are much the same in character, and are smelted with copper ores, forming a
matte containing about 200 ounces of silver per ton.

The roasted ore is mixed with others, principally siliceous, and the mixture
is so arranged that, when smelted, it shall yield a slag containing as near as

*Riohards and Heimrod, Proc. Am. Acad., 37,415 (1902); Richards, Proc.
Am. Philos. Soc, 42, 28 (1903); Zeitsch. phys. Chem., 46, 189 (1903).
t Stas, Oeuvres, 3, 40.

PURIFICATION OF SILVER. IO,

possible 40 per cent of silica and a first -matte or fusible sulphide which assays
40 per cent of copper, 400 ounces of silver, and 6 ounces of gold per ton.

The first matte always contains a certain amount of lead, but the quantity
rarely exceeds 10 per cent. The next stage in the process includes the roasting
and concentration of the ore-metal, or first matte, to " white metal " containing
60 per cent of copper.

This white metal is now ready for the extraction of the silver, which com-
prises the following operations : Rough roasting; fine grinding; fine roasting for
sulphate of silver by Ziervogel's process ; leaching, and the precipitation of the
silver on plates of copper.

In the precipitation of the silver a certain amount of copper is found mixed
with the silver in the form of cuprous oxide and of small scales and scraps of
metallic copper, and a process of refining is necessary previous to melting. This
copper is removed by prolonged boiling with water containing a small quantity
of sulphuric acid, into which air is injected by means of a small jet of steam.
Sulphate of copper is formed, which is carefully washed out of the silver. The
silver is then dried and melted into bars of an average fineness of 99.9 per cent.

The silver came to us in the granulated form obtained by dropping
into water, and having been prepared with unusual care, was found to
be even purer than Mr. Pearce claimed. After solution in nitric acid,
it was crystallized fifteen times from acid solutions with centrifugal
draining. The solubility in cold nitric acid is so slight that this train
of purification could be effected without serious loss. Even after the
second crystallization the salt was free from any observable trace of
copper, and all other metallic impurities must have been eliminated
before the last crystals were obtained. The pure argentic nitrate was
dissolved in pure water and precipitated by hot ammonic formiate,
prepared from pure freshly distilled formic acid and ammonia which
had been collected in platinum. It is better to use pure ammonic
formiate instead of a mixture of ammonic formiate and acetate as Stas
did, because in the former case less carbon is included in the cells of the
crystals, and, moreover, that which it included occurs in a less danger-
ous form. Since the reaction takes place according to the equation

2AgN03 + 2HCOONH4 = Ag2 + 2NH4N03 + C02 + HCOOH

every 170 grams of argentic nitrate need 46 grams of pure formic acid
(or about 42 cubic centimeters of 90 per cent acid, with specific gravity
1.2) converted into ammonic formiate. It is necessary to take this
double amount in order that the ionized hydrogen (nitric acid) may
be removed by the salt of the weaker acid. The precipitation was
conducted in good glass. The beautiful white precipitate was washed
with the purest water, until the washings gave no Nessler test. This
silver, V,* although it had not been electrolyzed, was as pure as any

*The preparation work and analyses of this silver were conducted wholly by
T. W. Richards.

20 ATOMIC WEIGHTS OF SODIUM AND CHLORINE.

which we made, judging from the analysis. The method of purifica-
tion just described is the most convenient and safest of all. The only
improvement which might be suggested is the use of a silver dish for
conducting the final precipitation. This improvement was tried in the
next preparation.

Part of the silver, not electrolyzed, proceeding from the earlier
reduction with ammonic formiate and ammonic acetate (p. 18) was
dissolved in redistilled nitric acid. The silver nitrate, once more
crystallized, was dissolved in pure water, in a large silver dish.
Ammonic formiate was freshly prepared by passing pure ammonia gas
into redistilled formic acid. With it the silver nitrate was reduced in
a silver dish, but otherwise as described above, washed free from
ammonia and dried in the same silver dish. This sample was called W.

There is no reason to believe that any of these samples of silver
contained any appreciable impurity, except the traces of included
mother liquor already alluded to. It is probable that Stas's silver
also was equally pure at this stage in the proceedings. The subse-
quent treatment of the silver, in the effort to eliminate the mother
liquor and prepare the metal for weighing, seemed to be in our experi-
ments the only cause of difference in the quantitative behavior of our
samples, and there is every reason to believe that Stas introduced
during this treatment the impurities which have rendered a revision
of his work imperative.

Because of this importance of the subsequent treatment, it is
necessary to record a very detailed statement of the precise operations
of the present research in this particular, and to point out the differ-
ences between these operations and the methods of Stas.

Everyone must agree that fusion alone is the only safe method of
eliminating wholly the liquid included in the cells of the crystals,
whether these crystals be produced by electrolytic or chemical reduc-
tion. The difficulty is to find an environment for the silver during its
fusion from which it will not absorb or retain foreign substances.

The question presents a two-fold difficulty, because both the solid
support of the silver and the atmosphere which surrounds it must be con-
sidered. The former of these difficulties may be discussed first.

In what vessel may the silver be fused, in order to avoid con-
tamination ? Stas, in his posthumous work, recommends fusion in
a cupel of basic calcic phosphate, in the flame of a blast lamp, and
subsequent ignition in hydrogen. This process, while possibly as
good as he contends, may nevertheless be a dangerous one. Everyone
knows the risk to a platinum crucible of igniting in it a phosphate
with reducing material, and yet Stas took no steps to prove the absence

PURIFICATION OF SILVER. 21

of traces of argentic phosphide from the metal prepared in this way.
His own observation, indeed, confirms this suspicion, for he states
that silver thus fused in an oxidizing flame became covered with a
yellow crystalline film, which instantly disappeared in a reducing
flame. The great instability of the oxide of silver affords reason for
doubting his assumption that this film was an oxide ; it seems more
probable that it was a phosphate, produced by oxidation of silver
phosphide, and decomposed by reduction to this compound again.
Stas, relying entirely upon the superficial appearance of the silver
and its flame test as a guide to its purity, would not have discovered
the presence of phosphide.

The possibility of this danger is enough to incline the careful
chemist to avoid the piesence of phosphorus. As a matter of fact,
Stas never used silver prepared in this way for his atomic weight
researches,* and no one else seems to have used this procedure except
Scott, f who fused 5 grams of silver in this way for a single analysis.
Of course, however, a single analysis is too slender a basis upon
which to form any adequate judgment concerning its purity. Actu-
ated by the preceding considerations, we did not take the trouble to
test the method, preferring first to test others which gave more promise
of satisfactory results.

Another possible material, which has been partially tested in
previous work, is carbon. Sugar-charcoal may be prepared in a state
very free from metallic impurities, and would serve excellently if it
were not that carbon is probably soluble to a slight extent in the
molten metal. We gave this possibility a fair test, and found that
when the silver is fused on a cupel of sugar-charcoal with a clean
blast lamp and moderately pure illuminating gas, it is as a matter of
fact very pure. (Experiments 67, 68, 71, on page 61.) Probably the
oxidizing flame used to melt the metal prevents the introduction of
much carbon at first, and during the brief reducing period in which
the metal must be cooled to eliminate an excess of oxygen there is not
time to dissolve much carbon. Evidently, however, a process which
depends upon the mutual elimination of opposing errors is an uncer-
tain one, and hence not suitable for the most accurate work. That
carbon is really dissolved was well shown by another experiment, in
which a carbon boat in a porcelain tube filled with pure hydrogen
was used to contain the fusing silver. The, experiment yielded silver
slightly less pure than the previous average ; it must have contained

*Stas, Ouvres Com pi. , 3, 75.

tA. Scott, J. Chem. Soc. Trans., 79, 147 (1901).

22 ATOMIC WEIGHTS OF SODIUM AND CHLORINE.

0.005 per cent of dissolved carbon. (Experiment 7 9, p. 61.) Of course,
however, a single experiment has but little exact weight in such a
case, except to show that the process is probably undesirable.

Returning from these unsatisfactory methods to the one used very
early in his career by Stas, we tested once more the use of lime as a
support for the fusing silver. This seemed to be an especially appro-
priate substance for the containing vessel, because it is so easily
obtained pure , and because it is so hard to reduce that no metallic calcium
could be dissolved by silver. But, of course, the silver must be satu-
rated with dissolved lime, and it remained to be proved that a saturated
solution of lime in fused silver is so dilute as to introduce only a
wholly insignificant amount of impurity. This was easily shown by
actual experiment. Ten grams of silver fused on lime were dissolved
in pure dilute nitric acid in a platinum dish and precipitated as sulphide
by pure hydrogen sulphide gas, led in through a platinum tube. After
filtration on a platinum funnel, the remaining liquid was evaporated
to dryness in platinum. The minute trace of residue, probably due
to traces of dust, on moistening with hydrochloric acid gave no trace
of the calcium bands in a spectrometer which showed them plainly
from a solution containing 0.01 milligram of lime in 1 cubic centimeter.
Hence it is evident that less than this amount of lime was dissolved
by the silver, i. e., less than 0.0001 per cent, if any.

This experiment settled the question concerning the supporting
substance, but of course the lime must be perfectly pure. If the lime
is partially made from the nitrate, it must be very thoroughly ignited
before it is used, otherwise the silver may dissolve some of the oxygen
caused by the decomposition of the salt. This was shown by the
careful testing of a quantity of silver fused on a new boat, which had
not been thoroughly ignited. (Exps. 72 to 76, p. 61.) A boat in
which pure silver has already been fused has an advantage in this
respect, as well as because every accessible trace of reducible material
must have been dissolved out of it. Of course the lime used in the
present experiments was especially purified, as will be related.

It is true that even when carefully prepared a boat of lime is likely
to be so friable as occasionally to cause a partial coating of the button
of silver with the powdered lime, but the great surface tension of
molten silver keeps this all on the surface, whence it is easily dis-
solved by means of the dilute nitric acid always used by us to clean
the metal. Scott, who has objected to the use of lime, again on the
basis of a single experiment, does not mention having taken this
necessary precaution.

PURIFICATION OF SILVER. 27,

All the silver used in the three final series of experiments which
follow was thus ignited in pure, well-seasoned lime boats, capable of
holding from 30 to 50 grams at each fusion. The unglazed porcelain
boats in which the lime was molded were made on purpose for this
object by the Royal Berlin Porcelain Manufactory.

The environing atmosphere around the silver during fusion next
claims attention, since that, too, may vitiate molten metal. Obviously
air is impossible. A vacuum seemed a priori to be the best means of
preventing the access of impurity, but, to our surprise, we found that
silver fused in a vacuum was sometimes less pure than that fused in
the open flame of the blast lamp. (Exps. 72 to 76, p. 61.) This
puzzling circumstance was ascribed finally to the supersaturation of
silver with oxygen , derived either from included argentic nitrate or
from the calcic nitrate in the boat.* Silver precipitated by formiate
ignited on a pure boat does not show this impurity, because free from
oxidizing substances; and a few minutes' ignition in pure hydrogen
quickly removes in chemical fashion this supersaturation when it
exists, thus purifying the silver. The amount of hydrogen which
remains must be on the limit of the weighable ; at least, we were
unable to detect it quantitatively, because silver fused in hydrogen
yielded exactly the same amount of argentic chloride as that contain-
ing no oxygen which had been fused in a vacuum. This is shown by
the comparison of Exps. 77, 78, 80, 81, 83, and R. 2, with 84, 85,
and R. 3 (p. 65). Hence Stas's conclusion that silver does not dissolve
over 0.0004 per cent of hydrogen is supported, although indeed his
method of preparing the silver was somewhat different from ours, and
his method of distinguishing the hydrogen was somewhat doubtful.

The discovery of the residual trace of oxygen in the fused elec-
trolytic metal, which had contained occluded nitrate, was not made
until the work on sodium was finished, but fortunately three pieces of
the silver used in that work still remained, so that it could be com-
pared with the purest silver which we were able to make. The
comparison was made by weighing the chloride obtainable from known
weights of each sample, and the details will be recorded in full in the
chapter concerning the synthesis of this salt. For the present it is
enough to say that the silver used in the work on sodium yielded
0.004 per cent less chloride than the purest silver, f and therefore was

*A trace of chlorine, which, however, could hardly have heen present, would
have the same effect.

tThe average of analyses 65, 66, 70 was compared with the average of the
final series (see pp. 61 and 65).

24 ATOMIC WEIGHTS OF SODIUM AND CHLORINE.

assumed to contain that percentage of impurity. A change of only
0.002 in the atomic weight of sodium was caused by this difference.

All the silver used in the final series of syntheses of silver chloride
was free from dissolved oxygen, and three of the specimens must
have been free from hydrogen also. The method of treating each is
sufficiently specified in the table of results and need not be further
discussed here. The silver used in this part of the work was the
purest we were able to prepare, and we know of no impurities which
it could have contained, except in some cases the insignificant trace
of hydrogen already mentioned.

The boat used for the fusion was made from a mixture of pure
lime and calcic nitrate, and was contained in a porcelain tube closed
with Hempel stoppers. The tube was heated to bright redness in a
Fletcher furnace.* The resulting bars of silver were freed from the
adhering lime by scrubbing with clean, hard sand and successive
treatments with dilute nitric acid. The argentic nitrate was dissolved
away by ammonia, and many washings with water were followed by
drying the bar in a hard glass tube at about 4000 in a vacuum. f
Water might otherwise become sealed into the microscopic cavities,
often present in fused silver, during the next operation of cutting into
blocks. The cutting is best carried out upon a silver plaque by means
of a clean steel coldchisel. The blocks thus cut were superficially
etched once more with nitric acid, so dilute that a possible trace of
iron could not assume the passive state, and the cleaning and drying
operations were repeated. The pieces then preserved in a desiccator
over potash are ready for weighing and analysis.

After the details of the analysis have been discussed, the relation
of the purity of this silver to that of Stas will be made clear.

UME.
The preparation of pure lime upon which to fuse the silver was a
matter of great importance, because of the danger that any impurity
in it might contaminate the silver. Its preparation was begun by dis-
solving marble in nitric acid, boiling with an excess of calcic hydroxide,
filtering, neutralizing, and twice recrystallizing the calcic nitrate. %
Calcic carbonate was then precipitated with ammonic carbonate,
repeatedly washed, and finally dissolved in redistilled nitric acid.

*Proc. Amer. Acad., 30, 379 (1895).

tThe silver used in Experiments R 1, R 2, and R 3 (p. 65) was wrapped in
pure silver foil during this ignition to prevent contact with hot glass.

$ Calcic nitrate should be recrystallized 'below its transition temperature, 44°,
a proceeding much facilitated by inoculation with the respective hydrate.

THE SOLUBILITY OF ARGENTIC CHLORIDE. 25

The resulting nitrate, made alkaline with ammonia, was electro-
lyzed to remove a suspected trace of iron and filtered through a layer
of calcic carbonate upon a Gooch crucible. It was found that
amnionic carbonate, of variable composition, could be conveniently
distilled with steam from its solution. With such a distillate, pre-
pared wholly in platinum, calcic carbonate was again precipitated
and well washed. A small portion of this was converted into nitrate.

Ignition of the calcic carbonate to oxide was carried out in por-
celain, since platinum is slightly volatile at high temperatures, and
might have contaminated the lime. A dry mixture of calcic oxide,
with about a fifth its weight of powdered anhydrous nitrate, was
packed into specially made large unglazed boats of Berlin porcelain.
The molded mixture when heated to redness sintered firmly together
and made a chemically pure receptacle, large enough to contain about
50 grams of silver at a time. In order to prevent adhesion of the
boat to the glaze of the porcelain tube, the boat should not fit the
tube very closely, and it is well to cover the glaze under the boat with
a layer of pure powdered lime.

UTENSILS.

Of course all the containing pieces of apparatus — dishes, funnels,
crucibles, retorts, condensers, and so forth — were of platinum wherever
silica was to be feared and platinum was not itself harmful. In this
latter case, according to circumstances, either silver vessels or vessels
of the best insoluble glass (Jena, or " nonsol ") or porcelain, or in some
cases fused quartz, were used. In this way the greatest possible purity
of the preparations was assured.

THE SOLUBILITY OF ARGENTIC CHLORIDE.

Mulder pointed out in 1857* that the solubility of argentic chlo-
ride is sufficient to affect all quantitative results into which the pre-
cipitation of this substance enters. Stas overlooked this solubility in
his early determinations, and hence in 1876 repeated some of them.f
Other more recent experimenters J have in general confirmed Stas's
later conclusions about this solubility, so that it is not important here
to give a lengthy statement of our further confirmatory results.

There can be no doubt that freshly precipitated argentic chloride
is soluble in cold water to the extent of several milligrams per liter, and

*Mulder, Die Silber-Probirmethode. Leipzig, 1859.
tStas, Ouvres, 1, 751.

jFor references see Bofctger, Zeitsoh. phys. Chem., 46, 189 (1903); Rich-
ards, Proc. Am. Acad., 29, 69 (1893).

26 ATOMIC WEIGHTS OF SODIUM AND CHLORINE.

that this solubility steadily diminishes, without a very sharp change
at any time, as the precipitate remains in contact with the liquid.

A variable solubility of this nature may be due to one of two
causes. Kither the precipitate is capable of assuming two or more allo-
tropic forms, of which the least stable appears at first,* or else it
separates in a finely divided state, and gradually becomes aggregated
into larger masses. The work of Ostwaldf and of Hulett % has shown
how greatly a difference in size of the grains may affect the solubility,
especially of a salt only slightly soluble.

The latter explanation seems the more probable for several reasons.
In the first place, no sharp change in the solubility has ever been
detected, such as would be expected from the appearance of a new,
more stable solid phase. Again, the difference between the specific
gravity of the flocculent precipitate and fused horn-silver is very
slight, and their heats of solution in cyanide are identical, indicating
little or no difference of molecular constitution between the two.||
Moreover, as will be shown later, the curdy precipitate may be success-
fully washed in a measure unattainable with crystalline powders, show-
ing that the innermost recesses of the solid are more than usually
accessible, and therefore that the structure is i very finely divided and
loosely knit. The more compact, less soluble precipitate, obtained
either by heating or by long standing, has largely lost this possibility
of being thoroughly washed, and has become also much less soluble —
facts which are quite in line with the above hypothesis. However this
may be, the uncertainty of the solubility is a fact of important signifi-
cance to the analyst, for it obliges him to determine the amount of
solution in every experiment — the solubility can never be assumed as
known.

The matter is yet further complicated by the well-known fact
that the presence of excess of either ionized silver or ionized chlorine
causes a depression of the solubility, in the manner demanded by the
law of concentration effect. § A large excess of either precipitant
reduces the solubility to so small a value as to be practically negligible

*Ostwald, Z. phys. Ghem., 22, 307, (1897).

fOstwald, Z. phys. Chem.,34, 495 (1900).

JHulett, ibid,37, 385 (1901) ; 47,357 (1904).

|| Laevinsohn determined the specific gravities. His results were verified and
the ithermo-chemi'Cal argument added to them by Dr. W. N. Stull and one of us,
in some work not yet published.

JHoitsema, Z. phys. Chem., 20,272 (1896).

THE SOLUBILITY OF ARGENTIC CHLORIDE. 2"J

in most cases.* This may indeed be the reason why the precipitate
appears in a form so loosely knit, for at the moment of precipitation
it is almost impossible that the two precipitating liquids should for more
than an instant be exactly equivalent. Hence each first particle of
precipitate would quickly be surrounded by a liquid unfavorable to
its union with another in compact crystalline fashion.

Mulder thought that an excess of either silver or chloride equiva-
lent to four times the weight of the dissolved argentic chloride was
enough to complete the precipitation, and Stas thought that exactly
three times the equivalent amount sufficed. It is clear now that this
limit is determined merely by the efficiency of the means used to
detect the last faint cloud of precipitate, and then only approximately.

As has been said, the changing solubility demands that every
filtrate be tested quantitatively. The methods which may be used
are all open to some criticism. The determination of electrolytic con-
ductivity is in vain when any other electrolyte is present, and its
interpretation is complicated even in pure aqueous solution by possible
hydrolytic complications. The determination of electromotive force
depends upon the rigorous application of the law of concentration
effect to electrolytes — a somewhat doubtful question. The actual
evaporation of the solution and weighing of the precipitate leads to
many errors when silver is in excess, f and the determination of the
intensity of the opalescence upon precipitating the dissolved chloride
depends greatly on the details of the production of the opalescence.

From among all these methods we chose the last, after a very care-
ful study of their probable errors. But although fairly confident of its
results, for reasons shortly to be given, we were careful in every case
to allow in each experiment as little dissolved argentic chloride as
possible, so as to reduce the possible error to a minimum.

The method of comparing the intensity of the cloudiness of opal-
escent solutions and therefore the quantities of suspended precipitate
has already been discussed at some length in a recent paper on the
nephelometer4 This instrument consists of a device for the compari"
son of the tint or brightness of two solutions containing precipitate,
viewed against a dark background. By making one of these opales-
cent mixtures from a known amount of material, an unknown amount
may be estimated by comparison with it. It is easy to show that

^Compare page 30.

t Richards, Proc. Am. Acad., 29, 71 (1893).

% Richards and Wells, Am. Chem. Jour., 31,235 (1904).

28 ATOMIC WEIGHTS OF SODIUM AND CHLORINE.

with this instrument the average of many results is within i or 2 per
cent of the truth, so that the error with 2 milligrams of dissolved
argentic chloride is probably less than 0.04 milligram. This degree
of accuracy is as great as can be obtained in the collection of a pre-
cipitate for weighing, hence the method is to be ranked among those
suitable for exact work.

With the perfected nephelometer it was possibly to study the pro-
gressing precipitation of very dilute argentic chloride in a way pre-
viously impossible, and several unexpected irregularities were found
during its course. Since these are important in the quantitative sense,
they must all be briefly discussed.

The most striking of these is the variation in! the time of complete
precipitation caused by the presence of electrolytes. For example, a
solution of argentic chloride, on adding a slight excess of soluble argen-
tic salt, becomes opalescent far more quickly in the presence of much
sodic nitrate or nitric acid than in the absence of such an electrolyte.
At the end of five minutes the former solution appears about three times
as cloudy as the latter ; and several hours are needed for the two solu-
tions to attain equality of opalescence at a maximum value, which there-
after remains constant. It appears from this fact that the precipitate
probably at first forms in a colloidal s'tate, since electrolytes are known
to hasten the precipitation of colloids. This inference is confirmed by
the occasional appearance of pale blue or pink tints in the incipient
precipitate, which disappears as the cloudiness becomes more intense.

Oddly enough, a solution of pure argentic chloride (made by shak-
ing the curd)' precipitate with water and filtering) attains its maximum
cloudiness after subsequent addition of argentic nitrate more rapidly
than a solution of similar dilution made by mixing equivalent amounts
of very dilute argentic nitrate and sodic chloride solutions. The trace
of sodic nitrate left in the latter case seems to be unable to counteract
some other tendency holding the precipitate in solution. At first it
seemed possible that minute invisible ' ' crystal germs ' ' might be
present in the filtered solution, which accelerated the appearance of
the opalescence, but this surmise was soon overthrown'experimentally.
A solution made from curdy silver chloride and found nephelomet-
rically to contain 1.28 milligrams per liter, was diluted with an equal
bulk of pure water, and its concentration after agitation for over one
hour was determined by the nephelometer. The average concentra-
tion found by five trials, 0.61 milligram per liter, agreed with that
expected within a reasonable limit of error, and moreover the speed
of precipitation was no less than before. There could hardly have

THE SOLUBILITY OF ARGENTIC CHLORIDE. 20,

been any " crystal germs " existing in the more dilute solution, hence
one must assume that the undissociated part of the argentic chloride
is really in a different state, when it is dissolved out of salt long formed
from that in which it is when in the act of precipitation. The
observation at least adds one more anomalous fact, perhaps giving a
clue to some of the other anomalous facts, concerning the solubility of
argentic chloride.

The application of these observations to the nephelometric analysis
of a very dilute silver chloride solution is obvious. It is a difficult
matter to prepare a standard which will behave in exactly the same
way as the solution to be estimated. By adding appreciable quantities
of nitric acid to each, it is nevertheless possible to hasten the precipita-
tion of each so greatly as to cause them to proceed at a speed almost
equal ; and the maximum cloudiness thus quickly attained persists
unchanged in relative intensity for hours. It seems highly probable
that the result thus reached really represents the amount of dissolved
argentic chloride ; especially because the same result is attained after
a much longer time even when no electrolyte is present.

The actual extent of the solubility of argentic chloride thus esti-
mated varied through wide limits, as has already been stated. For the
solution freshly filtered from the curdy precipitate in the usual condi-
tion at 200 the amount dissolved was ordinarily about 1.5 milligrams
per liter, a result about equal to that of Kohlrausch and Rose and others.
On long standing such a solution was found to decrease in concentra-
tion, on one occasion to as little as 1.1 milligrams per liter, although
no visible precipitate was seen on the glass. The trace of material
represented by this difference was probably adsorbed by the glass.
This observation points out the necessity of determining the strength
of an unknown solution immediately after its filtration, a precaution
which was carefully heeded in the quantitative part of the work.

Although sodic nitrate and nitric acid hasten the precipitation
of the curdy opaque form of argentic chloride, they augment slightly
the amount which finally remains in solution. By means of the
nephelometer it was found that a saturated solution of curdy argentic
chloride in the presence of about tenth normal sodic nitrate and a
small amount of nitric acid usually contained as much as 3 milli-
grams of the slighly soluble halide per liter. A mixture of this type
was present at the conclusion of each nephelometric experiment
recorded in the experiment on the ratio between sodic chloride and
silver, and the bearing of the actual value on the investigation will
be there discussed.

f

•%-:

30 ATOMIC WEIGHTS OF SODIUM AND CHLORINE.

The solubility of silver chloride in a liquid containing a large
excess of one of its ions has been stated to be " practically negligible
in most cases" (p. 27). In determining the ratio between silver and
silver chloride, however, it seemed advisable to determine even this
solubility, which may partly represent the solubility of the undissociated
silver chloride, as distinguished from that ionized, and also includes
the slight correction for the imperfect retaining power of even the best
filters. The fact that a single part of silver chloride may be detected
in 30,000,000 parts of water by the nephelometer in the presence of
an excess of silver nitrate shows the approximate order of this solu-
bility.*

In five experiments, in which silver chloride was precipitated by
an excess of hydrochloric acid, the combined mother liquor and wash
waters remaining after filtration were evaporated to dryness. The
residue was taken up with a very small quantity of ammonia and
then neutralized with nitric acid. After the addition of a slight
amount of sodic chloride, the silver chloride was estimated by the
nephelometer, by comparison with tubes made up with known amounts
of silver in the same way.

Silver chloride found.

In 3 liters 0.09 <mg. AgCl.

In 3 liters 13

In 3 liters 05

In 2 liters 05

In 2 liters 05

Mean 03 mg. per liter

This value represents the solubility of argentic chloride in water
containing sodic nitrate and an excess of hydrochloric acid, together
with those traces of the substance which could not be collected upon a
filter. It is not certain that the same value would be obtained if it
were possible to determine this infinitesimal solubility in the presence
of an excess of argentic nitrate ; but the fact that the limiting value
previously found by one of us is about of this order seems to render
this probable. Hence, for the sake of completeness, the correction
was applied to the results in the latter case as well as in the former,
although it was so small as not even to affect the third decimal place
of the atomic weight of sodium.

Further consideration of the solubility of argentic chloride will be
given in other appropriate places.

♦Richards, Proc. Amer. Acad., 29, 74 (1893).

OCCLUSION OF DISSOLVED SUBSTANCES. 3 1

THE OCCLUSION OF DISSOLVED SUBSTANCES BY ARGENTIC

CHLORIDE.

It is well known that all precipitates have a tendency to carry
down with them other substances from the solution in which they are
formed.* According to the nature of the precipitate, the mechanism
of this contamination varies. A basic precipitate such as ferric hydrox-
ide is likely to carry down traces of acid in the form of basic salt ; an
acid precipitate such as silica is likely to carry down traces of base ;
many substances are contaminated in the act of precipitation by the
admixture of isomorphous substances in solid solution ; all or nearly
all crystals contain within themselves minute cells inclosing mother
liquor ; and very finely divided precipitates, because of their large sur-
face, pertinaciously adsorb in varying amount impurities from the super-
natant solution. Obviously contamination from any of these sources
must be scrupulously avoided in accurate work, and the peculiar qual-
ities of each precipitate under investigation must be studied.

As has been said, argentic chloride was known by Stas to occlude
argentic nitrate, which is easily detected by its blackening effect upon
fusing the contaminated chloride. f This test is a delicate one. Stas
found that upon working with solutions of argentic nitrate weaker
than decinormal, the occlusion was so slight as to be negligible ; there-
fore in his early experiments he took great pains never to use solutions
stronger than this. Later he seems to have overlooked this precaution,
since he used much more concentrated solutions. Although realizing
that argentic nitrate is occluded by the precipitate, Stas seemed never
to have guessed that other salts might also be taken up in this way.
Evidently the question was one which needed further investigation,
and the outcome of the investigation is recorded below.

It was convenient at first to determine whether or not the occlusion
of sodic nitrate or sodic chloride would also affect the appearance of
the silver salt on fusion, in order to discover an easy qualitative test
for the presence of either. As a matter of fact, the former salt was

*Jannasch & Richards, J. Prkt. Chem., 39, 321 (1889) Schneider, Z. phys.
Chem., 10, 425 (1892) ; Richards, Proc. Am. Acad., 35, 2>77 (1900) ; Z. phys.
Chem., 46, 189 (1903).

tRichards, Proc. Am. Acad., 29, 76 (1893). The blackening occurs in full
measure only when the argentic nitrate is occluded from aqueous solution. Per-
fectly dry argentic nitrate fused with dry chloride is but slightly decomposed, and
the fused mass is merely gray, not black. The probable explanation is obvious.

32 ATOMIC WEIGHTS OF SODIUM AND CHLORINE.

found by actual admixture to color the fused silver chloride a pale buff,
rendering it opaque ; while the latter had no perceptible effect upon its
appearance. Thus Stas could not have detected the presence of sodic
chloride in his precipitate by the appearance, whether fused or unfused.
An impurity of argentic nitrate being thus the most easily detected
among the possible impurities, the next step was to use the contamina-
tion with this salt as a means of testing the properties of the precipitate.

It was early found that upon shaking pure argentic chloride with
a solution of argentic nitrate, the latter is adsorbed upon the surface of
the precipitate. If a solution containing argentic nitrate and the
curdy precipitate is divided into two equal volumes, one containing the
precipitate and one clear, and if the precipitate is then carefully washed,
more argentic nitrate will be found in the portion which contained
the precipitate than in the other portion . A further evidence of this
adsorption is found in the fact that the theoretical amount of washing is
especially inadequate to remove the last traces of adhering electrolyte.

These experiments show not only that the nitrate is adsorbed, but
also that some at least of the adsorbed salt may finally be washed off.

In view of this result, it seemed probable that the reason why
silver chloride occludes silver nitrate in the act of precipitation is
because the former adsorbs the latter during this process, and the
adsorbed material is then covered up by newly-formed precipitate.
The more dilute the precipitating solutions, the less considerable would
be the adsorption.

The peculiar curdy, flocculent form of this precipitate suggested
that since the material is so flexible, long-standing and gentle agita-
tion with pure water might allow the occluded material to escape,
although from a more rigid precipitate this escape is known to be
impossible. With this possibility in view, a number of experiments
were made in the following manner: Argentic chloride was precipi-
tated from several portions of a concentrated solution of the nitrate,
thus occluding much (in some cases as much as o.i per cent) of this
salt. The effort was then made to wash these precipitates ; and the
possibility of thorough washing was found to vary greatly with the
treatment of the precipitate. In those cases where the precipitate
had remained for two or three days in contact with a mother liquor
rich in silver, it was found much more difficult to remove the nitrate
than in cases where the washing was more prompt. It seems as if
the innermost cells of the precipitate are closed upon standing during
the well-known contraction which then occurs. In one case where
normal solutions had been used in precipitation, six prompt, gently-
agitated washings with pure water were enough to remove practically

OCCLUSION OF DISSOLVED SUBSTANCES. 33

all the occluded material and yield a chloride which fused without
blemish ; but this praiseworthy behavior was the exception rather
than the rule. In this case the innermost cells must have been cleaned
before they closed during the compacter aggregation of the precipi-
tate ; but usually traces of silver nitrate were inclosed, and when they
were once inclosed were beyond the reach of even a dozen washings.
Therefore more dilute solutions of argentic nitrate were used in
precipitating, in order to find the limit beyond which a satisfactory
precipitate was certain. Third normal solutions were found to be
still slightly too concentrated, but fifth normal solutions always
3'ielded a satisfactory result, if speedily but gently washed six times,
and then occasionally agitated in pure water in order to dissolve out
the last trace of nitrate. In previous determinations in this laboratory
still less concentrated solutions (tenth normal) were used, so that no
correction need be applied to those determinations on this account.*
This further dilution is necessary unless the curdy precipitate is
speedily washed or agitated with a solution containing no excess of
silver.

Having thus obtained light on the mechanism of the occlusion of
argentic nitrate, we turned our attention to the possibility of occluding
sodic chloride. Here, as had been said, no qualitative experiment can
conveniently prove the presence of the impurity. Hence recourse was
had to a quantitative experiment. An amount of pure salt known by
previous experiment with very dilute solutions to be exactly equivalent
to a given weight of pure silver was carefully weighed out, and the
solid was dropped into a fairly concentrated solution of the nitrate,
according to Stas's later method of procedure. After two hours'
thorough shaking and a day's standing, which Stas had always given,
it was found that 0.014 Per cent more of salt had to be added in order
to attain exact equivalence, showing that this amount of salt remained
occluded by the precipitate even after the shaking and standing. Agi-
tation for another day leached out about one-half of the occluded salt,
but seven days' soaking and agitation were required before the full
amount (0.014 Per cent) of extra silver nitrate was needed to attain
equivalence with the salt dissolved out of the precipitate. Subse-
quently in several days no more salt appeared, hence the final condi-
tion must have been reached. Since probably some silver nitrate
also was occluded, it is clear that these figures represent a minimum
value, and the certain error introduced by Stas's method of procedure

*See for example, Richards, Proc. Am. Acad., 29, 76 (1893).

34 ATOMIC WEIGHTS OF SODIUM AND CHLORINE.

was demonstrated. The reason why he obtained consistent results
was simply because he used exactly the same method in each analysis.

Obviously, then, a solid salt is not to be dropped into a precipi-
tating solution — indeed, such a proceeding is so repugnant to the
instinct of a precise quantitative analyst that it is surprising Stas should
have been guilty of it. But how dilute must the precipitating sodic
chloride be in the present case in order to avoid the danger ? Careful
experiments showed that with fifth normal solutions, as in the case of
silver, such slight occlusion as existed was quickly removed by gentle
agitation, and after a short time the mixture showed a constant equi-
librium point. Hence in our final experiments the solution was always
as dilute as this. In previous similar work in this laboratory, the
solutions of halide were rarely stronger than fifth normal, and never
much stronger, so that the error from this source must have been
very small.

It might be supposed that when the sodic chloride and argentic
nitrate are too dilute to be carried down with the precipitate, sodic
nitrate is not likely to be carried down in important amounts. For if
both silver and chloride are fifth normal, the resulting sodic nitrate
must be tenth normal, a dilution at which even silver nitrate is not
sensibly occluded. Nevertheless, as a matter of fact, the final experi-
ments yielded quantitative evidence of the occlusion of a trace of
this salt, amounting to 0.005 per cent of the weight of the precipitate.
This is probably because the precipitate was obliged to stand for some
time in the solution, and it is doubtful if this source of error could be
wholly avoided.

From these experiments it appears that the degree of dilution at
which occlusion ceases to be seriously troublesome is about the point
at which most physical chemists agree that the realm of dilute solu-
tions begins. Hence it would seem that in these cases of adsorptive
occlusion the undissociated salt is that which is occluded, in confir-
mation of other work.*

Obviously the occlusion of argentic nitrate, or of sodic chloride,
would affect the results of the analysis, no matter whether silver or
argentic chloride served as the weighed standard wherewith to com-
pare the weighed sodic chloride, although more seriously in the former
case than in the latter. On the other hand, the occlusion of sodic
nitrate could only affect the result when the argentic chloride was
weighed ; its occlusion would have no effect on the amount of silver
required to precipitate the chlorine. The possible effect of these dis-
turbing influences will be alluded to again during the recital of the
quantitative operations.

*Richards, MacCaffery, and Bisbee, Proc. Am. Acad., 36, 377 (iQOi).

THE RATIO OF SODIC TO ARGENTIC CHLORIDE. 35

It is needless to say that these experiments were carefully carried
on in complete darkness, in order to avoid the known disturbing effect
of photochemical action. The temperature was usually 200 C. Un-
doubtedly this condition is important, since the aggregation of argentic
chloride is greatly affected by change of temperature.

THE RATIO OF SODIC TO ARGENTIC CHLORIDE.

Two obvious means are available for determining the combining
weight of sodic chloride — one, by weighing the amount of argentic
chloride which may be made from a given weight of this salt ; the
other, by discovering the weight of silver needed to precipitate all its
chlorine. Both these methods involve a consideration of all the
special points mentioned in the preceding sections, and both are essen-
tially gravimetric, although both involve the use of the nephelometer
in order to estimate the last traces of silver or chlorine in the super-
natant liquid.

The actual investigation of these two methods proceeded simul-
taneously in order to economize the time during the long delays to
which each was frequently subject. Indeed, in some of the previous
work in this laboratory, where the pure materials were scarce, the two
methods were actually combined in the same analysis, by first deter-
mining the weight of silver needed to precipitate the chloride, and
then weighing the chloride produced. This last procedure is never-
theless somewhat objectionable on account of its complication. It is
moreover subject to slight opposing errors ; the avoidance of one of
these involves the admission of the other, as has already been pointed
out in a preceding communication.* The magnitude of the uncer-
tainty thus caused is not great, and the final error involved is insig-
nificant even in atomic weight investigation of ordinary caliber, usually
not exceeding o. 01 per cent; but in the present case extraordinary
accuracy was sought. Hence, although the methods were studied
simultaneously, they were never combined in the same analysis. For
this reason the work is easily divided into two sections, each treating
one of the two ratios named above. Of these it is convenient to dis-
cuss first the ratio of the chlorides to one another ; and this discussion
is taken up in the present chapter.

After the preliminary experiments had indicated the nature of the
problems involved, a single specimen of pure sodic chloride was used
for analysis until all the details of analysis were perfected. Then the

*Richards and Archibald on the atomic weight of caesium. Proc. Am. Acad.,
38, 443 (1903).

36 ATOMIC WEIGHTS OF SODIUM AND CHLORINE.

various samples of salt were analyzed by methods of known precision.
The temptation to vary two conditions in a single new experiment
is thus easily avoided. The detailed description of every step of the
analytical process follows.

It was soon found that salt when free from excess of acid can be
fused in a platinum crucible in the air without either attacking the
crucible to an important extent, or becoming appreciably alkaline.
Quantitatively, it was later proved, as will be shown, that salt thus
prepared is identical to that fused in a platinum boat in a vacuum and
transferred and weighed in the familiar ' ' bottling apparatus ' ' which
has served so conveniently for hygroscopic substances. It was found,
moreover, that a crucible full of salt thus prepared did not gain in
weight during the operation of weighing, on account of the dry winter
atmosphere of the steam-heated weighing room. Probably this would
not have been the case in a moist climate; but being the case, it facil-
itated greatly the present work.

The crucible generally used for fusing the salt was, of course, pre-
viously freed from iron, and during twenty-three fusions it lost only
0.47 milligram, or 0.02 milligram each time. It is possible that a por-
tion of the weight was lost by volatilization during the preliminary igni-
tions of the crucible, when it was being prepared for weighing alone ;
but probably most of this trace of platinum went into solution in the
salt. The weight of the crucible before fusion was always taken as
the true one, however ; therefore even this slight possibility of loss of
course could not affect appreciably the quantitative result.

During the fusion of the salt the crucible was inclosed in a larger
covered Berlin porcelain crucible, especially made to fit it, and heated
by a blast lamp in a small clay furnace. The products of combustion
were suitably deflected, so that only pure air should be in contact with
the salt in the crucible. The fused salt was cooled in a desiccator in
the balance room, and the weighing was conducted by the substitution
of a suitable tare-crucible, as usual.

In order to avoid all risk of accidental loss, the salt was dissolved
from the crucible by placing it in a large covered beaker, and after
covering it carefully with water, agitating it gently from time to time
until solution was complete. This solution thus carefully prepared
was transferred Ito a 2 -liter Erlenmeyer flask with a well -ground
glass stopper, where it was diluted to suitable volume (over 5 liters per
mol). Here it was precipitated by the at least equally dilute argentic
nitrate in excess. The precipitate was freed from the supernatant
mother liquor by filtration as soon as it had cleared enough to make
this separation possible. Promptness is needed in order to avoid the

THE RATIO OF SODIC TO ARGENTIC CHLORIDE. 37

permanent occlusion of sodic and argentic nitrate in the manner already
stated, but too great promptness prevents complete precipitation of the
silver halide. About fifteen hours was usually allowed for the settling
of the solution. Agitation is not necessary.

The clear supernatant liquid was decanted through a large, care-
fully prepared Gooch crucible. The mat of this Gooch filter, through
which all the mother liquor and wash waters were passed, was prepared
of selected fibrous asbestos shreds, which had been boiled with nitric
acid and well washed. The mat was a trifle thicker than a filter
paper — so that the holes of the crucible were indistinctly visible, when
viewed in front of a strong light — and was not allowed to extend up
the sides of the crucible. A preliminary ignition of the mat was found
to result in a slight disintegration and brown coloration, and caused
loss during washing ; therefore before weighing the mat was dried at a
lower temperature, namely, about 1500. This same temperature was
used later in drying the argentic chloride. During the operation of
washing at first all but a trace of the precipitate was left in the flask,
and it was then rewashed by decantation. After each washing, the
precipitate was allowed to drain several minutes, for silver chloride is
known to hold water like a sponge. At the second or third washing
a disintegration of the precipitate often took place, resulting in an
emulsionlike mixture, but in most cases this could be prevented by the
presence of a little nitric acid. The precipitation had been made with
silver nitrate in excess. The first three washings of 0.2 liter each
were made with very dilute silver nitrate, and this was displaced in the
fourth washing by about 0.05 liter of pure water. In such wash
waters, which, with the mother liquor, usually amounted to about 2
liters, no trace of chloride could be found by the nephelometer. The
main weight of silver chloride was corrected by + 0.06 milligram,
however, in every case, for the otherwise determined solubility in such
wash waters (p. 30). From five to eight subsequent treatments
with pure water, very slightly acid, in 0.1 liter portions, completed
the washing. These later wash waters were treated in the manner
described below, in order to determine the chloride which they always
contained.

The precipitate was now transferred to the filtering crucible wholly
by means of a stream of water. It was found convenient first to
transfer and pack down by air pressure a small portion of the precipi-
tate on the asbestos mat, in order that the greater part of the argentic
chloride might be easily separated from this mat for fusion. By resting
the lip of the flask on the Gooch crucible it is possible to direct a
stream of water from a wash bottle into the flask and finally wash out

38 ATOMIC WEIGHTS OF SODIUM AND CHLORINE.

every particle of precipitate. In the most careful experiments the wash
bottle was provided with protecting bulbs to preserve the purity of the
water, or was dispensed with altogether, being replaced by a jet driven
by hydrostatic pressure. At this stage the cloth covering was neces-
sarily removed from the flask. The flask was finally rinsed with
ammonia, which was carefully tested for a possible trace of silver
chloride.

On drying the precipitate the temperature was made to rise very
slowly above ioo° during two or three hours. This deliberate drying
was found to be more effective than the immediate application of high
temperature, presumably because a hard coating was not then formed
about the mass of the precipitate. Finally the temperature was main-
tained at about 1500 for two or three hours longer. In this way fairly
constant weight is easily attained, which fifteen hours' further heating
at 1500 changes very slightly. The last traces of water were always
expelled by fusion, as is described below.

In many of the experiments the silver chloride was dried in an
electric drying oven heated by hot platinum wires conveying a current
of an ampere or two.* When properly constructed this oven gives
very satisfactory results. The working of the oven is somewhat more
satisfactory if the platinum wires are not heated to redness. In some
of the later experiments the precipitate was dried in a clean copper
oven heated by a gas flame with no apparent difference in result.

The precipitate thus collected and dried was carefully weighed
and then freed from adhering asbestos and transferred to a weighed
porcelain or quartz crucible for fusion. The crucible and contents
were tared against a similar crucible and contents, and the argentic
chloride under examination was then fused in a clean i oven constructed
from a large porcelain crucible, from the interior of which the products
of combustion of the gas were carefully deflected. The fusion is neces-
sary, because minute drops of water are always held by the hardened
argentic chloride in sealed cells. It is true that the loss on fusion is
not great, never amounting to 0.01 per cent of the weight of the pre-
cipitate, but it was enough to receive consideration. In every case it
was determined and the appropriate individual correction applied to
each weight. The maximum value of the correction was 0.008 per
cent and the minimum 0.003 Per cent. That the loss was not due to
volatilization of silver chloride is easily shown by the invariability of
the weight upon fusing a second time. That it was not due to accidental
reduction was shown in the later experiments by fusing again in a

*Richards, Am. Chem. Journal, 22, 45, 1899.

COLLECTING THE ARGENTIC CHLORIDE.

39

stream of chlorine, when the mass gained on the average only o.ooi per
cent. The fused argentic chloride, even before its treatment with
chlorine, was usually pearly and colorless.

In any case, it is clear that this loss on fusion, which has been
rejected as of doubtful significance by some authors,* really represents
included water, and must be heeded.

The determination of the weight of the main mass of the pre-
cipitate having been thus determined, attention must be directed to
several other precautions necessary to insure exact work. In the first
place, all the liquid which has passed through the Gooch crucible was
passed through a small washed filter, in order to collect the minute
asbestos shreds which even the best fibrous asbestos seems to lose under
pressure. This was done as soon as possible after filtration and before
proceeding to solubility determination. A siphon filtering arrange-
ment, or a large inverted flask fitted with a carefully cleansed rubber
stopper with two tubes arranged so as to maintain constant level,
obviated the tedious process of filtering by hand. The little filter was
finally washed free from salts, and the weight of the ignited residue
(minus the filter ash) added to the weight of Gooch crucible and pre-
cipitate. This residue usually amounted to only one or two tenths
of a milligram. Observation in the microscope easily showed that the
residue was really asbestos.

Because it was the object of the research to test every step of the
processes involved, the loss and recovery of asbestos shreds was tested
by special blank experiments. Large volumes of water were passed
through empty Gooch crucibles provided with the usual mats ; and
the loss of each crucible after drying was found, as well as the weight
of shreds recovered from the filtered water. Below are given the results
of these blank experiments :

IyOSS Of

crucible.

Asbestos
found.

I.

2.

3-

Average . .

O.OOO34
.OOOl6
.OOOI9

O.OOO34
.00045
.OOOl6

O.OOO23

O.OOO32

In two of these experiments the loss very nearly equaled the
amount of asbestos found ; the other one must have been contaminated
by an accidental outside impurity ; but the results of the blank experi-
ments are enough to show that the correction is a real one and not to
be omitted.

* Scott, J. Ohem. Soc. Trans., 79, 147 (1901).

40 ATOMIC WEIGHTS OF SODIUM AND CHLORINE.

In most experiments the residue of asbestos shreds was found,
after ignition, to contain a trace of silver. This was undoubtedly pro-
duced from the decomposition of silver chloride during ignition, the
latter salt having passed through the Gooch as exceedingly fine parti-
cles or having precipitated, in time, upon the asbestos shreds. The
amount of this silver, as well as the amount of asbestos shreds, varied
according to the efficiency of the Gooch mat and the fineness of the
holes of the Gooch filter — the smaller those holes, the better. The
loss of chlorine upon ignition was determined in sixteen experiments,
chiefly in those upon the synthesis of silver chloride. This was done
by determining the trace of reduced silver by Volhard's process, using
j^ normal solutions. The average of determinations, which varied
from o.oo to 0.07 mg., amounted to only 0.02 mg. This correction
was applied. Although negligible, its magnitude had to be deter-
mined in order that one might be assured of safety in neglecting it.

With the asbestos at our disposal, the prepared Gooch crucible was
not at all hygroscopic. An hour's standing in the balance case had no
effect upon its weight. Even upon ignition, the mat, weighing less than
0.03 gram, lost only a negligible amount — and it was never heated
above 1500 in the actual analyses.

As has already been said, no chloride could be found by the neph-
elometer in any of the wash waters containing any considerable amount
of argentic nitrate. Hence, after filtration through the small filters,
these were cast aside and the small solubility correction of 0.03 mg.
per liter applied to them.

The next two or three wash waters containing only small amounts
of nitrate were collected separately, because they contained traces of
chloride also. The subsequent wash waters, free from nitrate but
containing much chloride, were all united. The chloride in both
separate portions was carefully determined by means of the nephelo-
meter — that is, by comparison of opalescence. The two portions, one
containing traces of argentic nitrate and one free from salt, must be
estimated separately, because if all the wash waters are mixed together
the opalescence usually occurs prematurely, a circumstance which
interferes with the accuracy of the nephelometric comparison.

The amount of silver chloride thus found in the wash waters was
added to the weight of that collected on the Gooch crucible.

Any accidentally included foreign substance — for example, silica
from 'the glass-stoppered flasks, dust from the air, or occluded sodic
nitrate — which would make the obtained weight of silver chloride too
great, would make the computed atomic weight of sodium too low.
Since our results were lower than Stas's, particular attention was paid

A TYPICAL DETERMINATION. 41

to these and similar possibilities. One question especially seemed
to demand an answer, namely, that concerning the possibility of the
abrasion of powdered glass from the ground-glass stopper of the pre-
cipitation flask.

A newly ground stopper sheds glass noticeably. It must always
be polished and rubbed a long time until the loose particles are thor-
oughly dislodged. A flask of soft glass never ceases to grind away
slightly every time the stopper is inserted. The large 2 -liter Erlen-
meyer flasks used in the final experiments were of very hard insoluble
glass, with rather wide necks, very suitable for precipitation and for
washing. In order to test the extent of their abrasion, one of the
stoppers was removed and replaced a hundred times. The resulting
powdered glass was collected on a small fine filter and found to weigh
0.2 mg. Since, during a single determination, the stopper was not
replaced over a dozen times, it is safe to infer that the error from this
cause could not have exceeded 0.02 mg. — a negligible quantity.

Other sources of contamination were so carefully guarded against
that they can hardly have been present.

The foregoing general description may be profitably illustrated
by recounting the details of a single experiment, before the results of
all the experiments are given.

Experiment 63.

April 5, 1904. Salt J.

Grams.

Corrected excess weight of crucible over counterpoise 0.19994

Corrected excess crucible plus fused salt over counterpoise.. . 576754

Weight salt in air (240, 764 mm.) 5.56760

Vacuum correction -f- 0.00231

Weight of sodic chloride in vacuum 556991

As a suitable excess of silver nitrate, 16.40 grams were taken for
precipitation. The concentration of the silver salt being about fifth
normal and that of the sodic salt not much greater, the total mother
liquor was 1 .4 liters. Five washings of the precipitate were made after
it had stood over night, with 0.2-liter portions of water containing
about 0.05 gram of silver nitrate per liter. Washings Nos. 6 and 7
were made with pure water and collected separately. Washings 8 to
12 were made with slightly acidified water and collected separately.
The whole precipitate was finally washed upon the Gooch crucible.
After drying it appeared perfectly white.

42

ATOMIC WEIGHTS OF SODIUM AND CHLORINE.

Corrected excess weight Gooch crucible (dried at 1500) over Grams.

counterpoise 0.21709

Corrected excess weight crucible and AgCl (dried at 1500)

over counterpoise 13.87382

Weight of silver chloride in air 1365673

"Vacuum correction _|_ 0.00100

Weight of unfused AgCl in vacuum 13-65773

The mother liquors and wash water were passed through a small
filter, which was ignited in a small porcelain crucible.

Grams.

Corrected excess porcelain crucible over counterpoise 0.42525

Corrected excess, plus residue and ash 0.42535

Residue and ash.
Ash

0.000 10
0.0000 1

Residue, asbestos shreds 0.00009

Loss of chlorine during ignition (see page 39) +0.00001

Total residue.

0.000 10

A large portion of the silver chloride, 10.50 grams, was fused in a
small approximately counterpoised crucible weighing about 8 grams.

Grams.

Correoted excess crucible, plus AgCl, over counterpoise* 10.64846

After fusion, corrected excess over counterpoise 10.64769

Loss of 10.50 gm. AgCl on fusion

13.66

.00077

Hence loss of total AgCl

10.50

X 0.00077=0.00100

The nephelometer revealed no chloride in the mother liquor and
first five washings ; the correction of +0.07 milligram was applied for
this solubility even in the presence of the excess of silver nitrate.

Washings No. 6 and 7, volume 0.220 liter in all, were found
to have a trace of chloride, and its amount was estimated by the
nephelometer, making comparison with two different standard tubes.

The factor for converting 1 milliliter of standard solution in a
nephelometer tube into silver chloride per liter was 0.047.

Standard tube

containing
NaCl solution.

Time of

forming

opalescence.

Heights of opalescences
appearing equal.

Ratio of
heights.

Soluble silver
chloride.

Standard.

Wash waters.

0.02 cc.

15 min.

30

IOO

O.30

0.30X0.02X0.047=
0.00028 gram per
liter.

O.OO5

7 min.

IOO

80

....

i.i8X°-oo5Xoo.47=

9 hrs.

100

85

I.I8

0.00028 gram per
liter.

*In this case the counterpoise contained no argentic chloride, but in the
most careful experiments the counterpoise contained as much of this salt as the
test crucible.

THE FIRST SERIES OF FINAL RESULTS. 43

For 220 cubic centimeters the weight of dissolved AgCl is 22° x

1000
0.0002S = 0.00006 gram.

Similarly for washings Nos. 8 to 12 (0.94 liter) was found on the

average 0.00146 gram of argentic chloride per liter (the individual

determination being 1.47, 1.41, i-54, and 1.44 milligrams). For 940

cubic centimeters we have — ' X 0.00146 = 0.001^7 gram, and hence

1000 °

the total dissolved silver chloride = 0.00007 + 0.00006 + 0.00137 =

0.00150.

Altogether we now have for the total weight of argentic chloride :

Grams.

Weight of argentic chloride (vaccum) in crucible 13.65773

Asbestos, etc., washed through Gooch crucible -f- .00010

Dissolved -)- .00150

Loss of fusion — .00100

Corrected weight of argentic chloride 13.65833

Ratio, AgCl : NaCl = 13.65833 : 5.56991 = 100.000 : 40.7803.

Molecular weight NaCl (if Ag = 107.93 and CI = 35.455) =

40.7803 X I43-385 58.473

Assumed atomic weight CI 35-455

Hence atomic weight Na 23.018

Most of the following determinations were essentially similar to
this, but a few small differences should be noted. In Experiment 27
the eight washings were extended over five days. In Experiment 64,
which must be considered as the best single experiment of all, the
concentrations were lowered to tenth normal, and twelve washings
were made, the first within three hours, the last after a day's lapse.
Considerable nitric acid was added to the first seven wash waters.
The slightly higher result of this experiment probably indicates
diminished occlusion of sodic nitrate.

Three experiments were rejected, two because blackening of the
argentic chloride showed that the argentic nitrate in the first wash
waters had not been eliminated by the subsequent washing with pure
water, and one because of known loss of argentic chloride. These
oppositely erring results would have no effect on the final average if
they had been included ; but in such a table as this it seems advisable
to include only results not vitiated by known sources of error. On the
following page are given the final data and results of this comparison
of argentic and sodic chlorides.

44 ATOMIC WEIGHTS OF SODIUM AND CHLORINE.

The ratio of sodic and argentic chlorides.
FINAL SERIES.

Parts NaCl equal to

Exp.

Prep.

Wt. NaCl. (in vac.)

Wt. AgCl (ill vac.)

100.000 parts AgCl.

Grams.

Grams.

13

K.

3-27527

8.03143

40.781

19

B.

5-56875

13.65609

40.779

20

A.

4.18052

IO.25176

40.779

22

C.

4-54319

II. 14095

40.779

23

D.

1-97447

4.84196

4O.778

27

B.

3-97442

9-74547

4O.782

6l

H.

6.69495

16.41725

40 . 780

62

I.

2.88692

7-07955

40.778

63

J-

5-56991

13-65833

40 . 780

64

J-

5-85900

14.36693

4O.781

44.52740

109.18972

Av. 40.780

If the atomic weight of silver is taken as 107.930 and that of
chlorine as 35.455, the atomic weight of sodium, calculated from the
above average, will be 23.017, a result 0.14 per cent lower than Stas's.

In spite of every effort to detect some error tending to diminish
the atomic weight of sodium and thus to cast doubt on this low value,
and in spite of increasing precautions and increasing skill in the details
of the experimentation, the last results are no higher than the very
first one. Hence further prosecution of this method seemed to be of
no avail. The reason for the difference between this value 23.017 for
the atomic weight of sodium and Stas's 23.048 is best discussed after
the results of the comparison of sodic chloride with metallic silver
have been detailed.

THE RATIO OF SODIC CHLORIDE TO SILVER.

Since the time of Gay-L,ussac, the method of titration has been a
favorite way of determining ratios between silver and soluble halogen
salts. If the end-point were definite, this method would be capable
of extreme accuracy. It involves no transference, collection, or
weighing of a precipitate ; consequently the introduction of silica or
alkali from a glass-stoppered flask, occlusion of sodic nitrate, loss of
asbestos, and other sources of error, are altogether avoided. After
weighing the factors in the proper proportions, dissolving and pre-'
cipitating, one has only to withdraw a portion of the mother liquor
and determine whether or not the true end-point of the reaction has

THE METHOD OF GAY-LUSSAC. 45

been reached. It is not strange, therefore, that many eminent experi-
menters adopted this method in order to determine many atomic
weights. The solubility of silver chloride, however, was a fact which
Gay-Lussac, Pelouze, Dumas, and many others, as well as Stas in his
earlier experiments, did not consider; in consequence, all used the
wrong end -point. In his last experiments Stas, recognizing the
solubility of argentic chloride, attained the end-point in two ways.
The first method consisted in titrating to the extreme limit with
sodic chloride and then running back to the opposite extreme limit
with silver nitrate ; the true end-point was assumed to be half-way
between the two limits.* The second method consisted in bringing
the mother liquor to such a point that two portions of it gave equal
opalescent precipitates with silver nitrate and sodic chloride respec-
tively ; this was the end-point established by Mulder. Before accepting
this latter method, Stas attempted to prove for himself that a solution
of silver chloride in water gave an equally intense opalescent effect
with an excess of argentic nitrate and sodic chloride. He concluded
that the opalescence in the two cases was identical, but at the same
time admitted that the opalescent precipitates were extremely unstable
and varied continuously in appearance from the moment of precipitation
to the time they had settled out completely .f At first, not hoping to
improve essentially in this respect upon the methods of Stas, we used
this end-point, and after a few preliminary trials made eight careful
determinations, using in all 33.1035 grams of pure salt and 61.0898
grams of pure silver. The atomic weight of sodium calculated for the
average of these experiments was 23.032, with maximum and minimum
of 23.040 and 23.023, respectively.

These experiments were of value as giving practice in the manip-
ulation, but the extreme variation was so much greater than it ought
to have been that clearly some undiscovered cause of variable error
was present in the results.

A study of all the possible causes of irregularity eliminated all
probabilities except the uncertainty of the end-point. The purity of
the salt and the silver left nothing to be desired, for the same samples
gave constant results when the argentic chloride was weighed. Occlu-
sion had been precluded by the use of solutions of not much over
deci-normal strength, and the mechanical operations were so extremely
simple as to leave little opportunity for error. Hence much time was

^Oeuv.,1 755. This is unsatisfactory for many reasons. See Richards, Proc.
Am. Acad. ,29, 82 (1893).
tOeuv.. 1, 158.

46 ATOMIC WEIGHTS OF SODIUM AND CHLORINE.

spent in investigating the opalescent phenomena attending the pre-
cipitation of traces of argentic chloride. The improved and precise
nephelometer* served excellently in this tiresome research. We found
that its readings could always be depended upon, and that if any
irregularity occurred the fault was with the precipitate within the liquid
and not in the means of observing this precipitate.

The first step in these tests was to prepare a solution of pure
argentic chloride, free from excess either of silver or of chlorine. For
this purpose pure floccy silver chloride was very thoroughly washed
in a flask by decantation. From time to time two nephelometer tubes
were filled with the clear decanted wash water ; to one was added a
milliliter of argentic nitrate solution, which contained a milligram of
silver, and to the other a milliliter of an equivalent solution of sodic
chloride. After stirring, these tubes were examined in the nephelome-
ter within ten or fifteen minutes, and the intensities of the two opal-
escences were compared. No regularity could be detected in theresults,
successive trials varying over 50 per cent from one another. For
example, in one trial the tube containing an excess of silver was twice
as intense in opalescence as the other, while in the following trial, with
equal proportions of the same solution, it was only two-thirds as
intense as the other. The average of a dozen comparisons confirmed
approximately Stas's and Mulder's statement that argentic chloride
gives equal opalescence with the two precipitates, but the wide varia-
tions between the extremes emphasizes Stas's conclusions as to the
uncertainty of the results.

Subsequent experiments showed that the presence of an electro-
lyte greatly accelerated the formation of the precipitate, indicating a
colloidal transition stage. The preceding experiments had been
almost free from electrolyte, and in view of this inference, it is easy
to see why the formation of the precipitate had been so irregular, for
slight differences in the extent and speed of mixing might cause impor-
tant differences in the coagulation of the colloid. Clearly, in making
nephelometric experiments, an excess of electrolyte must be present.
Other possible causes of error, such as the effect of light, and the pres-
ence of absorbed argentic nitrate or common salt in the precipitated
argentic chloride from which the solution was made, were carefully
guarded against in the further experiments. In order to avoid pos-
sible hydrolysis, f nitric acid was always added.

♦Richards and Wells, Am. Chem. Jour, 31, 235 (1904).

f Mulder, Die Silber-Probirmethode (1859, p. 72). Some of Mulder's observa-
tions dn this direction may ihave been due ito occluded material, but the danger of
hydrolysis is not wholly to be despised.

EQUALITY OF OPALESCENCE.

47

Even when all these precautions had been taken, however, incom-
prehensible irregularities were still present in the results. It was then
found that the time allowed for the formation of the opalescence had
not been adequate ; that in some cases one tube had clouded much
faster than the other, but that in all cases neither tube had attained
maximum intensity. Often as much time as eight hours was needed
for the attainment of this maximum ; but the time must not be too
greatly extended, or some of the precipitate will be deposited on the
bottom of the tube. The performance of the precipitation when
chloride is in excess is thus not wholly parallel with that when silver
is in excess, already discussed on p. 28.

In view of these facts, a new comparison was made. A quantity
of silver chloride was precipitated in red light from cold decinormal
solutions of silver nitrate and sodic chloride, washed twelve times with
water weakly acidified with nitric acid and eight times further with pure
water, the last washings being made in absolute darkness. Pure water
containing such quantities of nitric acid and sodic nitrate as would
occur in an actual mother liquor from a quantitative precipitation,
was then added, and the mixture was allowed to stand with occasional
agitations for a day. In making the nephelometer observations on
the decanted liquid, sufficient time was allowed for the precipitates to
become constant. Three completely separate trials were made in the
nephelometer ; and of these are recorded below the depths of liquids
which seemed to be equally opalescent.

Observation.

Time.

Tube+AgN03-

Tube+NaCl.

65
66
67

After 8 hours. . .
After 8 hours. . .
After 8 hours . . .

98 mm.
100

103

100 mm

100
100

100.3

100

These observations, made with great care, settled every doubt
about the correctness of the equal-opalescence end-point. Thus it is
clear that Stas encountered difficulties with this method only because
he, like all the previous experimenters, did not wait long enough to
allow the opalescence really to establish itself in the absence of elec-
trolyte to hasten precipitation.

In the course of these experiments several subordinate points
were noted which are worth recording. When the opalescent precipi-
tates were made directly in the tubes from sodic chloride and argentic

48

ATOMIC WEIGHTS OF SODIUM AND CHLORINE.

nitrate, having first one and then the other in excess, the speed of
formation was not like that observed when a solution of argentic
chloride was treated with the two electrolytes. In the former case, the
maximum opalescence was attained more slowly and the tube con-
taining an excess of chloride always became cloudy much more rapidly
than that containing an excess of silver, while in the latter case the
difference was much less marked. This adds to the evidence that the
dissolved argentic chloride is appreciably colloidal, and that the forma-
tion of this colloidal portion requires time.

The possible effect of light upon this end-point was another cir-
cumstance of interest, although not of immediate bearing on our
quantitative work, since actinic light had been scrupulously excluded.
In order to ascertain the effect of light the solution already tested in
observations 65, 66, and 67 was agitated in the flask for an hour at a
distance of about 30 cm. from a white incandescent electric light. Ob-
servations 68 and 69 were made with the mother liquor. It was then
agitated in diffused daylight for another hour, an exposure which made
the precipitate decidedly bluish, and observations 70 and 71 taken.

Observation.

Tube+AgN03.

Tube+NaCl.

68
69
70

7i

Average

100 mm.
100

97

100

100 mm.

100
100
100

99

100

The solution thus gave the same results in the nephelometer as
before. Bottger, in his study of the solubility by the conductivity
method, also observed that the conductivity did not sensibly alter until
the conductivity vessel was exposed to strong light for a considerable
length of time. A slight coloring of the precipitate made no dif-
ference.* The explanation must be that chlorine thus set free does not
ionize. According to Vogel.f silver subcloride is insoluble in very
dilute nitric acid. Hence neither an excess of ionized silver nor
chloride should appear in the mother liquor. L,ight, therefore, could
not have been the cause of any of the variations in preceding observa-
tions, and its complete exclusion had been supererogatory.

*Zeit. phys. Chem., 46, 521 (1904).

t Gilbert's Annalen, 72, 286.

DETAILS OF ANALYSIS.

49

Having thus, by many experiments, of which those given above
are merely samples, assured ourselves that the true end-point may be
decided by equality of opalescence when sufficient time is allowed to
complete the opalescent precipitation, it became an important matter
to make a conclusive series of analyses by this method.

In the first place, a given piece of the purest silver was carefully
weighed and was dissolved in a flask upon the steam bath. For
every gram of silver about 3 milliliters of nitric acid, of density
about 1.20, were used. This density was secured by mixing pure
distilled nitric acid of the usual concentration with an equal volume
of water. After the silver was dissolved, more water was added and
the nitrous fumes were expelled by a brief, very gentle ebullition. A
tower of bulbs, ground to fit the neck of the flask, prevented any loss
of silver nitrate during the operations. A special experiment proved
the efficiency of the bulbsas a means of retaining spray. In this man-
ner of treatment nitric acid always remained in slight excess.

Assuming a probable value for the atomic weight of sodium,
the exact weight in air of salt equivalent to the weight of the given
piece of silver was calculated and weighed out for fusion. There was
always a slight loss of weight when the salt was fused in the platinum
crucible, and if, after fusion, more than a milligram of salt was still
required, approximately the required amount of well-dried salt was
added at once by a platinum spatula. The amount so added never
exceeded a few milligrams and its introduction could have caused no
error, since in our laboratory salt was not hygroscopic. The exact
weight of salt corrected to vacuum was obtained. Any further still
slighter lack of equivalence was then made up by volumetric addition
of the proper quantity of the same solutions which were to bring the
mother liquor to equal opalescences after the precipitation. Of course
the recorded weights include all these small additions.

Both salt and argentic nitrate were diluted to a concentration
about fifth normal, and were then mixed in complete darkness with
all possible precautions, the argentic nitrate being very slowly poured
into the salt solution. When in a day or two the precipitate had set-
tled and the mother liquor had become clear, the latter was examined
nephelometrically .

It was found to be convenient, in determining the amount of a
deficiency of either silver or chloride, to calculate once for all the
extent of this deficiency in terms of the ratio of lengths of the two
nephelometer columns giving equal apparent opalescence. This was
in the first place calculated by means of the law of concentration effect ;
but practically a slightly different value, based upon experience, was

5o

ATOMIC WEIGHTS OF SODIUM AND CHLORINE.

found to be more accurate. A ratio of ioo millimeters to 80 millimeters
was found to signify an excess or deficiency of o. 15 milligram of silver
per liter, and other amounts in proportion. Since the average of many
readings was probably accurate within 2 per cent and the volume rarely
exceeded a liter, it would appear that the end-point was determined
within 0.02 milligram of silver. This does not, of course, prove that
the results are accurate to within this small margin, because other
experimental details may have led to larger error. The difference
between the maximum and minimum results corresponds, indeed, to
a much greater uncertainty, as will be seen. Hence the end-point is
determined precisely enough for our purpose.

The final series of experiments, given below, includes only those
experiments in which every known source of error was avoided. In
order to illustrate the details of the method, the full record of a single
experiment, in so far as it does not repeat what has already been described,
is here given.

Experiment 53.

Grains.

Weight of sodk chloride (in vacuum) (Sample I) 5.08685

Weight of silver (in vacuum) (Sample O) 9.38819

These were dissolved and mixed on February 2, 1904. After two

hours' agitation the mixture stood until next day, when an examination

was made with the nephelometer. Two test tubes were filled with the

mother liquor by means of a clean pipette, withdrawing 0.07 liter for

this purpose. The two solutions were very nearly equal in opalescence

With occasional shaking the mixture was therefore allowed to stand

yet another day, in order to insure the thorough solution of possibly

adsorbed material. The subsequent proceedings are summarized

below :

February 4 (0.07 liter withdrawn).

Time of standing in
nephelometer.

Nephelometer readings.

(a) Reading of tube
with excess AgNOg.

(6) Reading of tube
with excess NaCl.

a + b.

30 minutes

2 hours

18 hours

45
43
46

54
54
55

.83
.80

.83

This reading indicates a slight deficiency of silver. Hence 0.15
milligram of silver in solution was added to the residual mother liquor,
and the mixture was occasionall3r shaken.

DETAILS OF ANALYSIS.

51

February 8 (0.070 liter withdrawn).

Time.

Nephelometer readings.

a.

b.

a+b.

5 minutes. .
16 hours. . . .

48
60

3°

57

I.60
I.05

There was now a slight excess of silver. Hence there was added
0.03 milligram of sodic chloride, corresponding to 0.06 milligram of
silver.

February 16 (0.07 liter withdrawn).

Time.

Nephelometer readings.

a.

b.

a-f- b.

X hour

2 hours . . .

50
SO
50

39
44
46

I.28
1. 14
I.09

f
12 hours i

53
36
55

33
70

52

37
24

54
35
59
33
67
53
38
?3

O.98
I.03

o-93
1. 00
1.04
0.98
0.97
1.04

360

362

1. 00

Thus precise equality had been attained.

Volume liquor now remaining 435 cc.

Withdrawn during trials 280

Original volume 715

If the first addition of Ag had been made to the original total

7 1 S
mother liquor it would have been 0.15 X —^ = 0.19 mg. Similarly

715
0.06 X - — = 0.08 mg. Hence extra silver required to produce equal

opalescence in original mother liquor = 0.19 — 0.08 mg. = 0.0001 1 gm.

LjlLIBRAi

L-£A ***<& J 3*1

52

ATOMIC WEIGHTS OF SODIUM AND CHLORINE.

Total silver = 9.38819 + 0.00011 = 9.38830 gm. This silver
contained 0.0037 Per cen^ °f impurity. Weight of pure silver present
9.38795. Then Ag : NaCl = 9-33795 = 5-08685 = 100.000 : 54.185.

Molecular weight NaCl = 107.930 X 54- 185, if Ag = 107,930. 58.482
If atomic weight CI 35-455

Atomic weight of sodium 23.027

Ten such complete analyses are recorded in the table below.
The ratio of sodic chloride to silver.
FINAL SERIES.

Exp.

Preparations.

Wt. salt,
(in vac.)

Observed

wt. silver.

(in vac.)

Correct wt.

of purest

silver, (vac.)

Parts NaCl

equiv. to

100.000 parts

silver.

NaCl.

Silver.

Grams.

Grams.

Grams.

41

F.

N.

3.96051

7.30923

7.30896

54.187

44

K.

N.

2.32651

4-29371

4-29355

54-186

46

K.

N.

5.36802

9.90736

9.90699

54.184

45

G.

M.

4.00548

7-39237

7.392IO

54-186

47

G.

N.

4.69304

8.66133

8.66IOI

54-186

52

H.

P.

3.27189

6.03864

6.03842

54.i85

53

I.

P.

5-08685

9.38830

9-3S795

54-185

55

J-

P.

3-66793

6.76977

6.76952

54-183

56

I*.

P.

5-48890

10.13030

10.12993

54-185

57

h.

P.

3-55943

6.56933

6.56909

54.i85

41.42856

76-45752

54-i85

The agreement of these results is now as good as could reasonably
be expected. Whereas the earlier similar experiments, made before
time enough had been allowed for the full development of the opal-
escence, showed an extreme variation in the results of 0.017, this
extreme variation is now only 0.004. Since in these last experiments
seven different specimens of salt were used, and three different speci-
mens of silver, it seemed hardly probable that further work in the same
direction would add certainty to the average. The chemical identity
of the various preparations seems to have been proved.

The atomic weight of sodium calculated from the ratio 100.000 :
54.185, the average of these ten determinations, is 23.027, if silver and
chlorine are respectively assumed to be 107.930 and 35.455. This
is 0.010 higher than the value obtained by reference to argentic
chloride — a difference which seems small in comparison with much
quantitative chemical work, but which nevertheless is far greater than
the sum of the probable experimental errors of the two series of results.

DETAILS OF ANALYSIS. 53

The conflicting outcome of all these carefully conducted experi-
ments thus still remained not easily explicable. Their value for the
atomic weight of sodium differed by over o.i per cent from Stas's
result, and furthermore, as has just been said, the comparison of sodic
chloride with silver yielded a higher result, 23.027, than that found
by comparison with argentic chloride, 23.017. Both these discrepan-
cies required explanation before the results could be accepted without
reserve.

In the first place, much time was spent in the search for some
possible error in our results. For example, several of our samples of
salt, especially the preparation numbered K, were tested for lithium
with the greatest care by fractional treatment with alcohol. The
extreme mother liquors showed no trace of the lighter metal, although
in parallel experiments it was found that less than 0.01 per cent of
lithium chloride could be detected in this way, even by the simple
flame test. This proportion could not affect the atomic weight of
sodium by more than 0.002 ; but since no trace of lithium was found
in any of the specimens of our salt tested, it is clear that our low. value
of the atomic weight could not be due to lithium. The constant
results found with differently prepared samples from different sources
seemed to exclude the possibility of the presence of the very few other
metals which would yield too low a value for the atomic weight,
namely, beryllium, magnesium, aluminium, and calcium. Acid could
not have been present, since the material was usually prepared from
salt finally crystallized from pure neutral solution, and was always
fused in air. It gave, moreover, a wholly neutral reaction to methyl
orange.

Because no flaw could be found in the operations leading to the
new results, the conclusion was inevitable that the old ones were in
error. In order to prove this without question, experiments were
instituted in which each one in succession of the gravest faults already
pointed out in Stas's work were imitated, in order to detect the indi-
vidual errors introduced by each.

In the first place, two experiments were made in which the salt was
prepared approximately according to Stas's method, the substance
being fused with ammonic chlorplatinate. The salt thus prepared was
compared with pure silver by means of the nephelometer, using, as
Stas did, the opalescence which had been given only fifteen minutes
to form. In this way 2.67137 and 5.51270 grams of salt needed,
respectively, 4.92908 and 10.1717 grams of silver, corresponding to an
atomic weight of sodium of 23.039 in each case. Since the value
obtained from salt certainly free from platinum in the same way (p. 45)

54 ATOMIC WEIGHTS OF SODIUM AND CHLORINE.

was 23.032, the difference of 0.007 maY probably be ascribed to plati-
num, which had not been precipitated by thermal decomposition.
Presumably, therefore, traces of platinum were present also in Stas's
preparations.

In the next place, the effect of the occlusion of sodic chloride by
argentic chloride was quantitatively studied. It will be remembered
that Stas a/ways dropped dry sodic chloride crystals into his silver nitrate
solution instead of adding the salt in the dissolved state. Such a pro-
cedure offered the maximum opportunity for the occlusion of sodic
chloride. In the following experiments, made with the purest mate-
rial (salt F and silver P) and executed with all the precautions
outlined above, this error alone was purposely committed. After
dropping in the solid salt, the mixture, having been shaken two hours,
was allowed to stand for a day, all according to the manner of Stas. In
three experiments (Nos. 42, 51, 48,) 5-37853> 3-340i8, and 5-29539
grams of salt required, respectively, 9.92510, 6.16392, and 9.77122
grams of silver, thus giving three determinations of the atomic
weight of sodium, as follows: 23.033, 23.033, and 23,036, or, on the
average, 23,034.

In the last experiment the gradual dissolving of this occluded
salt was shown in an indubitable manner. On continued shaking
throughout a week, more and more silver had to be added, until finally
the mixture came to complete constancy at a point corresponding to
9.77264 grams of total silver. This quantity gives 23.028 as the
atomic weight of the sodium, showing that within a week nearly if
not quite all the occluded sodic chloride had been disengaged.

If, then, we sum up the effects which we have found for the three
chief errors probably imade by Stas in his later work, the following
result is obtained :

Our value of the atomic weight of sodium from the ratio

of sodic chloride to silver 23 .027

Effect of too prompt reading of opalescence -f--005

Presence of platinum in salt + .007

Occlusion of salt in silver chloride +-007

Result which we should have obtained by Stas's com-
plete method 23.046

This result is so near that found by Stas's assistants in 1881
(23.050) as to make it seem probable that these errors really existed in
Stas's work, and that the present investigation avoided them. The
acknowledged presence of traces of gas in Stas's silver and silica in
in his salt may be neglected in this comparison as being errors of

FUSION OF SODIC CHLORIDE IN VACUUM. 55

mutually opposing tendency and almost equal effect. The rectifying
of previous work by means of a later correction is always a doubtful
process, and the above table is given only to show that Stas's small
mistakes have all been traced.

It not so easy to give the numerical magnitude of the correction
to be applied to Stas's eaiiicr work, which, as has been said, is incon-
sistent with the later work, because it gives the same result, although
a wholly false end-point was used. It is probable, however, that in the
early work, carried out by Stas himself, the precipitated argentic
chloride was much more thoroughly shaken with its mother liquor
than the later work executed by MM. Nyst and Cabilliauw under Stas's
direction. Continued shaking would diminish the solubility of the
argentic chloride, as well as disengage the occluded salt. Thus the
decrease of the latter error might have balanced the effect of a greater
error in the end-point, although this in its turn would be less marked
than it would have been had the mixture been less thoroughly shaken.

All these considerations tend to support the present determination
as against the verdict of the older ones, but no explanation is yet
afforded as to the reason for the difference between the values 23.017
and 23.027 found by different processes in the present work. Long
consideration of the matter led finally to the conclusion that the
accepted value of the atomic weight of chlorine, 35.455, upon which
the calculation depends, must be in error. Accordingly this problem
also was attacked experimentally. But before these lengthy experi-
ments are recounted brief mention must be made of the successful
attack upon another doubtful point, which concerns the sodic chloride.

THE FUSION OF SODIC CHLORIDE IN VACUUM.

This matter has been given an especial heading because it is one
which has never before received adequate treatment in an investigation
of this kind. It is well known that many substances, when fused in
the air, dissolve either oxygen or nitrogen, or both, thereby increas-
ing sensibly their weight. While it was hardly to be expected that
this would be the case with sodic chloride, the point should not pass
unchallenged in a research striving to attain the highest possible
accuracy.

Hence two analyses were made of sodic chloride fused for some
time in a platinum boat inclosed in a completely evacuated porcelain
tube, and subsequently bottled in dry air in the often described " bot-
tling apparatus." So far as we know these are the only quantitative
experiments ever made in which all the weighed substances were fused

56

ATOMIC WEIGHTS OF SODIUM AND CHLORINE.

in a vacuum before weighing. It is true that Stas has shown that
amnionic chloride sublimed in a vacuum gave the same equivalent
referred to the silver as preparations sublimed in air ; but this choice
of ammonic chloride had the disadvantage that its two dissociation
products possess different rates of diffusion, and moreover his silver had
not been fused in a vacuum as ours had been. The data and results
are given below.

The ratio of sodic chloride to silver ; doth substances fused in a vacuum.

No.

Sample
salt.

Sample
Ag.

Weight.

NaCl.
(in vac.)

Weight.

Purest

corrected

Ag.
(in vac.)

Parts
NaCl eq.
to 100.000
parts Ag.

59
60

J.

J-

P.
P.

3.38684
4.68529

6.25046
8.64634

54-185

54-188

54-I87

These experiments show that probably not enough gas is absorbed
during fusion of the salt in air sensibly to alter the apparent combining
weight. It is true that gases easily remain in supersaturated solution
in liquids, and that long-continued relief from pressure is necessary to
remove them ; but fused sodic chloride is so mobile and possesses so low
a density as to make the retention of much gas very unlikely. That air
may be adsorbed by the cold salt is of course highly probable, but the
surface of a fused lump is so small that the weight of this adsorbed
air must be negligible. One would here expect a lower result from
the salt fused in a vacuum, if gas were dissolved in the other ex-
periments, but the results were in fact slightly higher, although not
beyond the probable limit of experimental error. It appeared from
tests with indicators that the salt remained strictly neutral after fusion
in a vacuum, as it did in the air.

Being unable to conceive of any other sensible inaccuracies which
could affect the analyses of the chloride of sodium, we turned our
attention next to the composition of the chloride of silver, because of
the before- mentioned suspicion that Stas was here also in error. An
additional reason for this revision is found in the fact that this atomic
weight of iodine, found at about the same time, has been found to be
about 0.1 per cent too low.*

*Ladenburg, Berichte d. d. chem. Gesell., 35, 2275. Scott, Proc. Chem. Soc,
18, 112. Kothner and Aeur, Ber. d. d. chem Gesell., 37, 2536 (1904). Also,
Liebig's Ann., 337, 123 and 362 (1904). Baxter, Pro. Am. Acad.,40, 419 (1904)-

THE SYNTHESIS OF ARGENTIC CHLORIDE. 57

THE SYNTHESIS OF ARGENTIC CHLORIDE.

Many accurate analyses were made by Stas which involve more
or less directly the atomic weight of chlorine, but of these we are
concerned in the present paper only with those which compare silver
with chlorine, because such alone could cause an error leading to the
discrepancy under discussion. The experimental solution of the
problem is thus a comparatively simple matter, to be accomplished by
combining silver with chlorine. Among all Stas's experiments there
seems to have been only seven which dealt directly with this ratio.*
These were carried out by two methods. Of these methods one, the
ignition of silver in a stream of chlorine, is acknowledged by Stas to
possess errors which he could not wholly correct. Hence the three
experiments made by this method are of doubtful value, and only four
analyses remain for the purpose of obtaining directly this very important
ratio. Even in these the large glass globes employed were more or less
attacked by the strong acids or the fused chloride, and, moreover, Stas
questioned the purity of the silver used in one case.f There is obviously
no doubt of the need of obtaining more light upon this matter.

The fundamental requirements of a satisfactory synthesis of argentic
chloride are as follows: First, that the silver must be pure; secondly,
that it must all be collected in the argentic chloride, and thirdly, that
this argentic chloride must be pure. Because all the error accumulates
upon the consequent atomic weight of chlorine and is magnified in
calculation, the requirements are usually rigorous.

The fulfilling of the first of these requirements has already been
treated in detail, under the discussion of the preparation of silver.
Our knowledge of the essentials of this easy but subtle process grew
as the quantitative work on the synthesis of the chloride progressed.
Whether in the future anyone may be able to discover an unsuspected
source of impurity in our purest silver, is of course impossible to say.
At least, there can be no doubt that it was of distinctly a higher grade
than the silver used by Stas in most of his work.

In order to convert this silver wholly into the chloride without
loss or illegitimate gain, two methods were used, for fear that in one
of these a constant error might lurk unsuspected. The first consisted
in the usual method of solution in nitric acid and precipitation in
dilute solutions, being essentially similar to that used in obtaining the
ratio between common salt and silver chloride. The weighed silver

*Stas,Untersuchungen (trans. Aronstein) 173, 175 (1867), Ex Oeuvr., 1,333-
tStas (Aronstein), pp. 117, Oeuvres, l,p. 330.

58 ATOMIC WEIGHTS OF SODIUM AND CHLORINE.

was dissolved with all the precautions used in previous work and pre-
cipitated as chloride by hydrochloric acid at concentrations of about
fifth normal. Hydrochloric acid was added in excess and the pre-
cipitate was washed three times with dilute hydrochloric acid, in order
to diminish as much as possible the solubility of the silver salt. After
collection on a Gooch crucible, drying at 1500, and washing, the bulk
of precipitate was fused in a porcelain or quartz crucible and the loss
in weight by fusion determined.

It would appear at first as if the result of this process could furnish
nothing not directly calculable from a comparison of the two sodium
series, in one of which common salt was compared with silver and in
the other with argentic chloride. This is not true, however, for in
this previous work the precipitate formed in the presence of sodic
nitrate and excess of argentic nitrate, while in the present case an
occluded alkaline salt could not be carried down, because none was
present, and all the silver could be converted into chloride. For this
reason the synthesis of argentic chloride in the presence of nothing
but volatile substances is an important step in the work.

As usual, in this case also, improvement was effected as the work
progressed, and here as before the less perfect tentative work is included
in the preliminary table. The synthesis of argentic chloride is a pro-
cess of greater certainty than a metathetical reaction, for the reasons
just stated, and hence finer corrections could be discovered and applied.

These corrections, in addition to the two already heeded, namely,
that for the asbestos shreds carried away during filtration and that for
the inclosed wash water set free on fusion of the chloride, were five in
number.

The first of these was a correction for the presence of a trace of
argentic chloride clinging to the collected asbestos shreds. This silver
salt lost chlorine on ignition — a quantity which was determined by

N

dissolving the residual silver and titrating with sulphocyanate.

The amount of chlorine thus lost on the average was 0.02 milligram
in each synthesis — an amount which was added to the weight of the
asbestos shreds. This correction might in most cases have been
omitted, but it was determined for the sake of certainty.

The second of these corrections was for the slight solubility of
argentic chloride even in dilute hydrochloric acid. This solubility may
have been only apparent, the trace found being due to the imperfec-
tion of even the double filtration through asbestos and filter paper —
but in any case it needed determination and correction. The aver-
age solubility, from five determinations (page 30), was one part in

THE SYNTHESIS OF ARGENTIC CHLORIDE. 59

30,000,000 of water, which is about the order of the possible solubility
of undissociated argentic chloride, and hence may have represented
that quantity. The determination was made by evaporating the hydro-
chloric acid solutions, rinsing out the dish with ammonia and testing
the resulting solution nephelometrically by comparison with similarly
prepared known solutions of silver chloride, as has been said.

The third correction was for an occasional loss of chlorine by the
argentic chloride on fusion, owing either to the presence of argentic
nitrate or to traces of dust. The darkening thus caused represents a
real although often trivial deficiency of weight, but, fortunately, this is
easily restored by fusion in a mixture of pure chlorine and hydro-
chloric gas. Of course this process must not be adopted when argen-
tic nitrate is occluded during the analysis of a metallic chloride, for it
would then lead to a large excess of argentic chloride ; but it is perfectly
safe in the present case. The only suspicion which attaches to it is
the possibility that an excess of one of these gases might be dissolved.
This is undoubtedly a real danger, unless the fused salt is exposed to
the air and agitated in contact with it while fused. If this is done,
we have assured ourselves by repeated experiments that no essential
excess of chlorine remains dissolved.* Two facts pointed to this con-
clusion— first, that those samples of argentic chloride, which after the
preliminary fusion merely in air were perfectly colorless, transparent,
and pearly, gained no essential weight on fusion in chlorine (Expts.
80, R. 2 and R. 3, final series, p. 65) ; and secondly, that argentic
chloride fused in chlorine, properly freed from it, and subsequently
again and again fused in pure air lost no essential weight.

We always secured elimination of the possible trace of dissolved
chlorine by continuing to fuse in air with occasional removal of the
cover, and by carefully rolling the inclined crucible around, while
held with forceps, during the solidification of the fused mass. Spread-
ing the solid in this way makes it easier to fuse the salt once more
without breaking the very fragile quartz crucibles.

The attacking of glass by chlorine in contact with silver chloride
and a consequent loss of weight gave Stas considerable difficulty.
Baxter has found that even porcelain is attacked in displacing iodine
with chlorine in the silver salts. We did not greatly fear a loss of
weight of the porcelain crucibles, since our chlorine was dilute, and
was applied only for a few minutes ; nevertheless we adopted the
use of quartz as an added precaution in the later experiments. No
difference in results could be detected by its use.

*Dr. Baxter has come to the same conclusion. Proc. Am. Acad., 40, 432
(1904). And so also have Kothner and Aeuer, Lieb. Ann., 337, 127 (1904).

60 ATOMIC WEIGHTS OF SODIUM AND CHLORINE.

A fourth correction which we sought to determine was for the pos-
sible volatility of the argentic chloride during fusion. The constancy
of weight on repeated fusion, already mentioned, seems to be sufficient
proof for considering this correction as negligible. The conclusion is
strengthened by the results of Biltz and Victor Meyer, who found
argentic chloride to be but slightly volatile at much higher tempera-
ture.* It is not, of course, contended that argentic chloride has no
vapor tension at 5000, but only that the loss occasioned by its evapo-
ration is too small to receive consideration.

During fusion, however, there is sometimes a slight spurting, or
projection upward of minute drops of fused material. Hence the
crucible must always be covered with an accurately weighed lid — a
precaution which was always taken in the final experiments.

The phenomenon of spurting suggested that weighable quantities
of air may be absorbed during fusion in air, and led to our effort to
discover a fifth correction. Carefully conducted experiments showed,
however, that there was no loss in weight of 3 grams of ordinarily
fused argentic chloride when fused again in a vacuum. The fusions
were conducted in a porcelain boat, covered with another inverted to
serve as a lid. Slight traces of inclosed gas were noticed by Stas in
many substances, but our quantitative experiments showed that they
do not amount to a weighable quantity either in common salt or silver
chloride. Thus, all the final experiments in this paper refer to sub-
stances fused in vacuo, and hence presumably as free as possible from
inclosed air.

There were thus seven possible corrections, besides the correction
of the weights to the vacuum standard, which might be applied to the
results. Two of these, that for the possible volatility of argentic
chloride and that for weight of dissolved air, were so small as to be
entirely negligible. The chlorine lost from the traces of argentic chlo-
ride clinging to the displaced asbestos shreds was added to the weight
of these shreds, and is included therein in the tables below. The
three other corrections are recorded in detail in the following tables,
which contain the original weight of the precipitate dried at 1500 as
well as the final corrected values. The almost negligible correction
for the solubility of argentic chloride in dilute hydrochloric acid was
computed, in the preliminary series, from the volume of the washings.

*Berichte der d. Chem. Gel., 22, 727 (1889). Kothner and Aeur found a
volatilization of only 0.0002 in a case where several liters of gas had heen passed
over argentic chloride during several hours. This agrees essentially with our
result, for the likelihood of distillation in our case was much less (loc. cit,
p. 128).

THE SYNTHESIS OF ARGENTIC CHLORIDE.

6l

A few inessential changes in the mode of precipitation were occa-
sionally introduced. In Experiments 75, 76, and 79 both solutions were
only about tenth normal, and the number of washings was diminished ;
in R. 1 the argentic nitrate was very dilute, but the hydrochloric acid
was somewhat concentrated.

As has been said, the tentative experiments were all classed
together in a preliminary series, given immediately below. The table
explains itself, for the most part, except perhaps the second column.
This defines in the first place the designation of the silver, next the
nature of the containing vessel on which it was fused, and last the
atmosphere surrounding the metal during fusion. The word " blast "
signifies that the sample thus qualified was fused in a blast-flame of
illuminating gas and air, delivered from a clean nozzle, and that the
resulting mass was cooled in the reducing flame.

Synthesis of argentic chloride.

PRELIMINARY SERIES.

Description

of silver :
Sample; sup-
port arid en-
vironment
during fusion.

Ob-
served
weight
of silver
(vac).

First ap-
proxi-
mate
weight
AgCl
(vac).

Corrections to AgCl.

Cor.

weight

AgCl

(vac).

Pts. of
AgCl
pro-
duced
from
100.000
pts.Ag.

No.

of

exp.

Asbes-
tos res-
idue
(add.).

Loss on
fusion

(subtr.).

Gain

in

Cl2

(add.).

Cor.
for so-
lubil-
ity
(add.).

65
66

P, lime, vac

Grams.
9.06483
8.39217
5-37429

Grams.
12.04416
II. 14987

7.14072

Mg.
O.IO

0.35

0.26

Mg.
0.79

o.57
0.51

Mg.
O.I2

0.14
O.03

Mg.
O.06
O.06
O.06

Grams.
12.04365
1 1. 14985

7.14056

Per ct.
132.861
132.860
132.865

70

132.862

P, carbon, blast

67
68

8.08222
7.08517
7-97715

10.73844

9.41382

IO.59854

0.46
0.20

0.45

0.49
0.50
0.82

0.22
O.04
O.14

0.06
0.06
0.06

IO.73869

9.41362

10.59837

132.868
132.864
132.859

7i

132.864

72

73
74
75
76

8. 1 1978
8.53452
6.73284
8.91366
9.72295

10.78751

II-33904

8.94497

II.84225

12.91736

0.36
0.28

0.42

0.48
0.56

0.38
0.40
0.51
0.48
0.41

O.09
O.09
0.14
O.09
0.09

O.09
O.06
O.09
0.06
O.09

10.78767

"•33907

8.945 1 1

II.84240

12.91769

132.857
132.861

132.858

132.857
132.858

132.858

U, carbon, H2..
V, lime, blast...

79

8.63961

II.47877

0.30

0.61

0.07

O.09

II.47862

132.860

R. 1

I I- 13795

14.79856

0.24

0.50

O.IO

0.09

14.79849

132.865

132.861

132.848

62 ATOMIC WEIGHTS OF SODIUM AND CHLORINE.

Although not giving the true amount of chloride to be obtained
from the purest silver, this table was highly instructive. It showed in
the first place that even silver of this grade of purity must be o.oi per
cent purer than that of Stas.* Again, it showed that at least a part of
the difference between the two new values for the atomic weight of
sodium is due to an error in the assumed atomic weight of chlorine
used in the calculations. It proved that the analytical method was
capable of yielding very constant results with a given sample of silver
(Exps. 72-76). It confirmed previous resultsf in showing that portions
fused in these various ways were not very different, but showed slight
variations according to the method of fusion employed. It furnished
a means of determining, by later comparison of one or the other of
these results with those from the purest silver, the exact grade of
purity of the various specimens of silver used in the previous work in
this laboratory, in particular that used in the earlier part of the present
research. Finally, it furnished sufficient clue to the precautions which
must be taken to secure yet purer silver.

A careful study of this table led to the conclusion that a reducing
environment is necessary to eliminate traces of oxygen during fusion,
and that lime is better than carbon as a 'support during fusion. The
details of this matter have already been described (pp. 22, 23) under
the heading " Preparation of Pure Materials," and need not be further
detailed here. Suffice it to say that the silver used in the experiments
recorded in the table on page 65 was free from the sources of error thus
detected.

The final series comprehends ten experiments, made with four
samples of silver prepared entirely independently by three different

*There can be no doubt that this difference is due to the .fact that Stas fused
all the silver used in this part of his work in a porcelain crucible under borax
and nitre, thereby introducing both alkali and oxygen (Oeuvres, 1, 328). His
own subsequent experiments (Oeuvres, 1, 466) showed that this metal was at
least 0.005 Per cent less pure than his distilled granulated silver, which must,
in its turn, have contained some oxygen. His still later elaborate search for
oxygen in the silver is not satisfactory, because just before determining the
oxygen in it he ignited it in a reducing flame, and because he sought to deter-
mine the oxygen only by heating to redness in a vacuum, a process which may
have set free only the superficial oxygen. Dumas found 0.008 per cent of oxygen
in similar silver by fusion in vacuo (Ann. de ohim. phys., 14, 289, 1878), and
Mallet found that silver fused like Stas's purest in the oxyhydrogen blast con-
tained 0.005 per cent of oxygen. (Phil. Trans., 1 7 1, 1020 [1880]).
fRichards, Proc. Am. Acad. 29, 65 (1893).

THE SYNTHESIS OF ARGENTIC CHLORIDE. 6$

experimenters (Dr. Baxter having very kindly given us a piece of his
purest silver used in the determination of the atomic weight of iodine,
and each of the authors having made separate samples) and two entirely
different methods of synthesis.

Of these methods of synthesis one has already been described,
namely, the usual method of solution of the silver in nitric acid, pre-
cipitation by hydrochloric acid, and filtration on a Gooch perforated
crucible. Of course all the precautions and corrections already dis-
cussed were applied with as much care as possible. All the vessels
used were afterwards washed with a little ammonia, and the washings
tested nephelometrically in order to be sure that no silver was lost by
adsorption on the vessels. By this method Experiments 69, 77, 78,
80, 81, R. 2, and R. 3 were made. Further unvaried repetition of this
process seemed to promise no further light on the question.

Although this method seemed to leave little or nothing to be de-
sired as regards accuracy, it was nevertheless deemed advisable to test
it by making a few syntheses by means of entirely a different process,
in which all the vessels coming in contact with the precipitate could
be weighed. Accordingly a new method was devised, a modification
of one of those used by Stas, employing quartz vessels instead of
glass. A quartz dish was accurately weighed by substitution of a
counterpoise of a similar composition, surface area, and weight. A
watch-glass to serve as its cover was weighed in a similar fashion sep-
arately. A weighed piece of silver was placed in the dish, which was
supported in a large empty "desiccator," whose bottom and sides
were well moistened with pure water. The silver was covered with
redistilled nitric acid of the usual strength and left covered both by
watch-glass and desiccator lid to dissolve slowly over night in a warm
place. On the next day the moisture on the glass cover was washed,
with great care to avoid the slightest spattering, into the quartz dish.
If any spattering occurred it was caught in the large desiccator, which,
having been tightly covered, served also to catchall spray proceeding
from the solution of the silver. In two experiments, the silver recovered
from the water in the desiccator amounted to only 0.02 milligram
when estimated by the nephelometer, and in one no silver at all was
found in the desiccator.

Hydrochloric acid gas was next allowed to play upon the surface
of the silver nitrate solution in the quartz dish, and the resulting
chloride was caused to sink to the bottom by means of a very small
glass stirrer, afterwards carefully rinsed. The remaining solution of

64 ATOMIC WEIGHTS OF SODIUM AND CHLORINE.

nitric and hydrochloric acids was then evaporated to dryness upon a
clean steam bath under an appropriate hooded cover (Victor Meyer
funnel). The atmosphere around was kept as pure as possible during
this evaporation, which was usually conducted during the night.
Finally the silver chloride was fused in the quartz dish supported inside
of a large porcelain crucible from which the products of combustion
of the gas-flame were suitably deflected. The weighed cover caught
the one or two minute drops projected upward during the act of
fusion. Then a mixture of chlorine and hydrochloric acid gas was
passed in, so as to convert any residual traces of nitrate of silver
wholly into argentic chloride. The precaution of cooling, mentioned
on p. 59, enabled us to preserve the quartz dish intact through the
analyses, notwithstanding the extreme fragility of thin fused quartz.
Its weight was conserved through three experiments within the limits
of the accuracy of the weighing. Experiments 83, 84, and 85 were
made by this method. One of these is given below in detail as an
example of the method.

Experiment 84.

Grams.

Corrected weight of silver in air 6.25336

Vacuum correction 0.00018

Weight of silver in vacuum 6.25318

Excess weight quartz dish over counterpoise 0.215 10

Excess weight dish and fused AgCl over counterpoise 8.52279

Weight of argentic 'chloride in air 8.30769

Excess weight cover glass over counterpoise 0.09972

Excess weight cover and spattering over counterpoise 0.09974

Spattering 0.00002

Weight argentic chloride in air 8.30771

Vacuum correction 0.00061

Weight of argentic chloride in vacuum 8.30832

AgCl found in desiccator by nephelometer 0.00002

Total weight of argentic chloride 8.30834

AgCl = 6.25318 : 8.30834 = 100.000 : 132.866

This second method, which avoids any transference of material
whatever, is an unusually searching check upon the method employing
the Gooch filter ; and that the two methods confirm each other to the
limit of weighing is a most satisfactory criterion of each, supporting
the previous similar work with sodic chloride as well.

THE SYNTHESIS OF ARGENTIC CHLORIDE.

65

The following table is complete, including all the experiments
made with the purest material, and every precaution. The experi-
mental work, as well as the preparation work, was the product of
more than one experimenter, the senior author having made, from
beginning to end, the syntheses numbered R. 2 and R. 3 (as well as
R. 1 in the preliminary series),* and the junior author having made all
the others. The two experimenters used different balances, different
weights separately standardized, and different preparations throughout,
in order to secure additional certainty.

For No. 69, Baxter's silver fused on lime in hydrogen was used;
for Nos. 77, 78, 80, 81, 83, R. 2, sample T, fused in the same way; for
R. 3, sample V, fused on lime in a vacuum, and for 84 and 85 sample
W, fused on lime first in hydrogen and then in a vacuum.

Synthesis of argentic chloride.
FINAL SERIES.

Exp.

Observed
wt. Ag

(vac).

First

approx.

weight

AgCl

(vac).

Corrections to AgCl.

Corrected

weight

AgCl in

vac.

Parts

AgCl

pro-
duced by

100.000
parts Ag.

Asbes-
tos
residue.

Loss

on

fusion.

Gain
in Cl2.

Cor.
for
solu-
bility.

69

77
78
80
81
R. 2
83

Average .

R. 3
84
85

Average .
Total . . .

Grams.

7.24427

Grams.

9.62517

Mg.

+0.16

Mg.
-0.40

Mg.
+0.09

Mg.
+0.06

Grams.

9.62508

Per ct.

132.865

8.30502
7.29058
8.58472
8.01318
9.77160
7.98170

11.03312
9.68547
11.40578
10.64632
12.98283
10.60528

+ 1.10
1.23
0.57
0.34
0.74

-0.14
0.12
0.32
0.32
0.37

+0.67
0.09
0.02
0.05
0.02

+0.09

0.09
0.09
0.09
0.13

11.03484
9.68676
11.40614
10.64648
12.98335
10.60528

132.870
132.867
132.866
132.862
132.868
132.870

132.867

11.49983
6.25318
7.72479

15.27978

8.30834

10.26360

+0.44

-0.66

+0.03

+0.05

15.27964

8.30834

10.26360

132.868
132.866
132.866

132.867

82.66887

109.83951

132.867

This result, namely, 132.867 parts of the chloride from 100,000 of
silver, has a " probable error," computed according to the method of
least squares of about 0.0005. This so-called probable error of course

*The hydrochloric acid used in these experiments was three times succes-
sively treated with small amounts of permanganate and boiled to eliminate bro-
mide, and each time distilled.

66 ATOMIC WEIGHTS OF SODIUM AND CHLORINE.

gives no clue as to possible constant errors involved. A better idea of
the chemical trustworthiness of the result is got by comparing the
data furnished by the two different methods of synthesis. The averages
of the seven determinations by the method of filtration is 132.8666,
while that of the three by the method which involves no transference
of material (83, 84, 85) is 132.8673. These are both within the range
from the mean 132.8668 indicated by the " probable errors" of the
respective averages, and hence may be considered as identical.

If the comparison be made according to the environment of the
silver during fusion, a similar result is obtained. The silver fused in
hydrogen gives an average of 132.8668, while that fused in a vacuum
gives 132.8668. Again this is essential identity.

In view of these facts, the fact that all probable errors tend to
lower the result, and the multitude of precautions which eliminated
error from these figures, it is not unsafe to conclude that 100.000 parts
of the purest silver really yields as much as 132.867 parts of argentic
chloride. This conclusion confirms the comparison of the silver and
argentic chloride indicated by the work on sodium, and shows that no
great occlusion of sodic nitrate could have taken place there. It
furnishes, moreover, a means of determining by comparison the purity
of any other specimen which has been used in the quantitative synthesis
of argentic chloride by either of the foregoing methods. For example,
it shows that the silver used in the work on sodium — silver which
yielded 132.862* parts of chloride — must have contained j^ = 27000 =
0.0037 Per cent of impurity.

*See Syntheses 65, 66, and 70 given in the table on page 61. The silver used
in these syntheses was exactly similar to some of that used in the work on
sodium. In the same way, if Stas's method of synthesis is considered as com-
parable with ours, his silver must have contained (132867 — 132848) -s- 132867 =
0.015 per cent of impurity. It is doubtful, however, if the methods of synthesis
are strictly comparable.

EFFECT OF NEW ATOMIC WEIGHTS. 67

THE NEW ATOMIC WEIGHTS OF SODIUM AND CHLORINE AND
THEIR EFFECT ON OTHER ATOMIC WEIGHTS.

From the preceding description of the syntheses of argentic
chloride, it is clear that 100.000 grams of the purest metal yields
132.867 grams of chloride. If then the atomic weight of silver is
taken as 107.920* (a convenient value to assume for preliminary
calculation), the molecular weight of argentic chloride becomes by
simple proportion 143.393, and by difference the atomic weight of
chlorine becomes 35.470^ This value is about 0.05 per cent greater
than the value announced by Stas ; and the change produces a serious
effect on a number of atomic weights. It is, of course, impossible to
be perfectly certain of the accuracy of the new value, because it may
contain concealed within it some entirely unsuspected source of error,
but at least it is free from the real mistakes in the earlier work. It
was reasonably certain that at least ten more syntheses would be
needed to affect the value one unit in the third decimal place, if the
further results varied no more widely than those recorded above in the
final series. Hence it seems unlikely that chlorine is above 35.471 or
below 35.469 if silver is taken as 107.920.

What, now, is the effect of this new value upon the atomic weight
of sodium? In the table given upon page 44 it is shown that 100.000
grams of argentic chloride could be obtained from 40.780 grams of
sodic chloride. If now argentic chloride is taken as 143.390, sodic
chloride becomes 58.474, and by subtracting the new value for chlorine,
23.004 is obtained as the new value for sodium.

This is not, however, the only value for sodium which may be
calculated from our results. On page 52 has been given a table show-
ing the results of ten experiments on the comparison of sodic chloride
with silver direct, and on page 56 the results of two more. These deter-
minations were made before we had discovered the best method of
obtaining pure silver ; but the trace of impurity is allowed for in those
tables, and it is safe to conclude that 100.000 parts of the purest silver
are equivalent to 54.185 parts of sodic chloride. If, then, silver is
taken as 107.920 and chlorine as 35.470, sodium becomes 23.007.

Thus two entirely independent values, calculated from two entirely
separate series of experiments, yield respectively the values 23.004 and
23.007 for the atomic weight of sodium. Of these two, the latter is

*F. W. Clarke has for a long time accepted this value, and it is now assumed
because it is probably nearer the true value than 107.93.

tin this connection it should be pointed out that Leduc called attention to
the fact that if Stas's silver really contained as much oxygen as Dumas said it
must, chlorine would become 35.47. Compt. Rend. (1901).

68

ATOMIC WEIGHTS OF SODIUM AND CHLORINE.

somewhat the more trustworthy, because, as has been said, the occlu-
sion of sodic nitrate by the precipitate does not affect it ; hence the
mean may be taken as 23.006. This value is probably not more in
error than two units in the third decimal place, if silver is 107.920,
although of course the same remark applies to it that was made concern-
ing the possible error of the value for chlorine. If silver is taken as
107.930, sodium becomes 23.008, and chlorine 35.473. The new values
are, then, as follows :

(Ag=l07.92o).

(Ag= 107.930).

23.006

23.008

Atomic weight of chlorine ....

35.470

35.473

The new value for sodium is nearly 0.2 per cent less than that
found by Stas — a percentage error much greater even than that made
by him in the case of iodine. It will be noted that the likely error of
the new result is only about a twentieth of the difference between the
new result and that of Stas. Some will attach significance to the fact
that both the new values bring the elements nearer to the demands of
Prout's hypothesis.

These new values affect greatly the figures in the second decimal
place of all other atomic weights depending directly or indirectly upon
chlorine, sodium, or silver. The number of elements thus affected is
so great and the relations so complicated as to render imperative a re-
calculation of all atomic weights. Nitrogen, as computed from am-
nionic chloride, will be especially affected, the new value approaching
more nearly that required by Avogadro's rule than the old. Probably,
however, it will be best to delay this systematic recalculation until
a few other new data shall have been obtained — in particular new
analyses of potassic chloride, argentic chlorate, the bromides, sulphides
and sulphates, and similar important compounds. Some of these are
already under way at Harvard, and others will be undertaken at once.

In conclusion, it is a pleasure, to acknowledge our great indebted-
ness to the Carnegie Institution of Washington, without whose liberal
support the present investigation could not have attained its present
thoroughness or precision.

SUMMARY. 69

SUMMARY.

The investigation consisted of a very careful quantitative study
of three ratios— namely , AgClrNaCl, Ag : NaCl, and Ag : AgCl.
The effort was made to test every operation involved in the execution
of the experiments. In the course of the work, the following points
were developed :

1. Sodic chloride, obtained from many sources and purified in
many ways, always gave the same equivalent weight.

2. Fusion in vacuum of salt already fused in air caused no change
in this equivalent weight.

3. Argentic chloride precipitated from aqueous solutions always
occludes traces of other substances present, and those traces can not
always be eliminated. Very dilute solutions must hence be used in
precipitation.

4. The conditions governing the occlusion and release of these
impurities were minutely studied, and it was shown that Stas's method
of dropping solid salt into a silver solution causes occlusion of salt.
Many considerations important in all precise chemical work are dis-
cussed.

5. A careful study of the solubility of argentic chloride was made,
and the precautions necessary in using the nephelometer in the estima-
tion of traces of chloride and silver were ascertained.

6. Fused argentic chloride probably contains traces of dissolved
air, but not enough to affect essentially its weight, since subsequent
fusion in vacuum caused no appreciable loss of weight.

7. The most difficult question in the purification of silver was
found to be the elimination of the inclosed mother liquor without
introducing other impurities. Fusion on pure lime, first in pure
hydrogen and then in a vacuum, is the safest method. Stas's silver
must have contained at least as much oxygen as Dumas claimed.

8. In ten experiments, 44.5274 grams of sodic chloride yielded
109.1897 grams of argentic chloride.

9. In twelve other experiments, entirely distinct from these,
49.5007 grams of sodic chloride were found to be equivalent to 91 .3543
grams of the purest silver.

10. In ten other experiments, again entirely distinct from the
preceding, 82.6689 grams of the purest silver yielded 109.8395 grams
of argentic chloride. Two very different methods of synthesis were
used in this series, and the silver came from various sources, these
variations being without effect on the result.

yO ATOMIC WEIGHTS OF SODIUM AND CHLORINE.

ii. If the atomic weight of silver is assumed to be 107.920,
sodium is found from the above results to have an atomic weight of
23.006 and chlorine an atomic weight of 35.470.

12. Many other atomic weights are affected, in their second deci-
mal places, by these changes. In particular, certain slight anomalies
previously noticed in Harvard work are explained by them, and the
atomic weight of nitrogen computed from ammonia is brought nearer
to the value required by Avogadro's rule. Other anomalies appear
in other places, however, and it is clear that many new atomic-weight
investigations must be instituted to explain them, with due attention
to hitherto unheeded dangers, especially of occlusion.

A REVISION

OF THE

Atomic Weights of Sodium and

Chlorine

BY

THEODORE WILLIAM RICHARDS

AND

ROGER CLARK WELLS

OF HARVARD UNIVERSITY

WASHINGTON, D.-C:

Published by the Carnegie Institution of Washington
April, 1905

MBI. WHOI LIBRARY

WH IflDZ