Point spread function tails and the measurements of diffuse stellar halo light around edge-on disc galaxies

Mon. Not. R. Astron. Soc. 000, [HI] (2008) Printed 1 July 2008 (MN JAT^ style file v2.2)

Point spread function tails and the measurements of
diffuse stellar halo light around edge-on disc galaxies

oo
O

o

o

>

(N
(N

O

o

oo

o

%

Roelof S. de Jong

Space Telescope Science Institute, 3700 San Martin Drive, Baltimore, MD 21218, USA

Accepted 2008 May 28. Received 2008 May 15; in original form 2008 April 3

ABSTRACT

Measuring the integrated stellar halo light around galaxies is very challenging. The
surface brightness of these haloes are expected to be many magnitudes below dark
sky and the central brightness of the galaxy. Here I show that in some of the recent
literature the effect of very extended Point Spread Function (PSF) tails on the mea-
surements of halo light has been underestimated; especially in the case of edge-on disc
galaxies. The detection of a halo along the minor axis of an edge-on galaxy in the
Hubble Ultra Deep Field can largely be explained by scattered galaxy light. Similarly,
depending on filter and the shape one assumes for the uncertain extended PSF, 20
to 80 per cent of the halo light found along the minor axis of scaled and stacked
Sloan Digital Sky Survey (SDSS) edge-on galaxy images can be explained by scat-
tered galaxy light. Scattered light also significantly contributes to the anomalous halo
colours of stacked SDSS images. The scattered light fraction decreases when looking in
the quadrants away from the minor axis. The remaining excess light is well modelled
with a Sersic profile halo with shape parameters based on star count halo detections
of nearby galaxies. Even though the contribution from PSF scattered light does not
fully remove the need for extended components around these edge-on galaxies, it will
be very challenging to make accurate halo light shape and colour measurements from
integrated light without very careful PSF measurements and scattered light modelling.

Key words: methods: data analysis — galaxies: fundamental parameters — galaxies:
halos — galaxies: spiral — galaxies: structure.

1 INTRODUCTION

In recent years we have begun to appreciate the impor-
tance of galaxy stellar envelopes as tracers of the hierar-
chical galaxy formation process. Hierarchical models in a
ACDM context predict that the stellar envelopes around
galaxies are created from many disrupted satellites, where
size, shape, amount of substructure, and metallicity of the
envelope principally depend on the primordial power spec-
trum, the reionisation epoch, the star formation history of
accreted dw arfs, and the total dark matter mass of the host
galaxy fc.g..'Bckki fc Chiba"2005': 'Bul lock fc Johnstonl2005l :
[Abadi ct al. 2006; PurccU et al. 2007). The discovery of the
Sagit tarius Dwarf curr ently being disrupted by the Milky
Way lllbata et al.lll994|) and the highly structured envelope
around M31 (e.g.. [Ferguson et all l20o3 : llbata et all |2007| )
has given much credence to the hierarchical model. While
these observations were performed using resolved stars,
many measurements of gala xy haloes have been attempted
using integrated light (e.g., Morri son et <^ 1994: Frv et al.l
1999l:IWu et al.ll2002l: [Zibetti et al. 2004; Z ibetti fc FergusonI

200J). These observations are very difficult, as the halo light

is typically at least a factor of 10* below sky level, and there-
fore careful attention has to be paid to flat fielding and sky
subtraction. Here I investigate another effect that has some-
times been underestimated in integrated light studies: the
effect of scattered light in extreme PSF tails when examin-
ing edge-on galaxies.

The effect of convolving a PSF with a spherical light
distribution can be fairly well estimated. For instance, with
an elliptical galaxy the light in the central region will be
dispersed by the generally broader PSF shape. Further out,
the fight distribution is nearly unaffected as the PSF shape
is steeper than the slowly declining galaxy profile. By over-
plotting the PSF on the measured light distribution nor-
malised to the same central brightness, one gets a fairly good
impression of which radii are affected by light convolution
as modelled by the PSF.

For edge-on galaxies the procedure is not as simple
as the luminosity profile perpendicular to the disc can be
steeper than the outer tails of the PSF. Close to the mid-
plane of the galaxy the vertical light distribution will still be
modified by a point-source-like convolution; however, farther
away we are less dominated by the local light. Instead the

2 Roelof S. de Jong

light scattered from the whole disc — not just the central
core — significantly contributes to the measured distribu-
tion at large scale heights. To estimate this contribution at
more than 2 scale lengths above the disc (about 6-15 scale
heights) one should not use a PSF profile normalised to the
central surface brightness of the edge-on galaxy, but instead
normalised to the total brightness of the galaxy. Calculat-
ing the actual scattered light contribution at any point is
obviously best determined by convolving a 2D model of the
intrinsic light distribution with a full 2D PSF.

Another problem arises when studying galaxy haloes
from stacked galaxy images, scaled to a common size, in
order to reach fainter levels. One can estimate the PSF of
the stack by combining appropriate stellar images from each
field, scaled by the same factor as the galaxy in the frame.
However, for a typical sample selection the distribution of
scaling is not symmetric and highly biased toward the scales
near the selection limit of the sample. The distribution of
scaled PSF images is therefore strongly skewed, meaning
that combining the PSFs using median or mean values gives
very different results. Assuming that the galaxies are very
similar after scaling, no such skewed distribution is present
and median and mean combining should give very similar
results.

In this research note I investigate a few cases in the
literature where the effects of PSF convolution with edge-
on galaxies has been underestimated. I will also show that
the colour measurements of the envelopes around galaxies
can be strongly affected by scattered light. In particular, in
Section [2] I will show the effect of PSF convolution on the
light distribution around an edge-on galaxy i n the Hubble
Ult ra Deep field (HUDF:lBeckwith et al.ll2006h as measured
by (jZibetti fc FergusonI |2004 ). In Section |3] I estimate the
scattered light effec t on the stacked SDSS image analysis of
IZibetti et al.l (|200J). Finally, in Section |4] I summarise my
conclusions and give a few recommendations.

2 HUBBLE ULTRA DEEP FIELD GALAXY

Comparing the PSF and minor-axis galaxy profile nor-
malised to the same central brightness leads to an underesti-
mate of the contribution from PSF scattered light. However,
this method has appeared in the literature and been used
to argue that scattered light will not contribute to the halo
light r neasurement (e.g . , iFrv et al.lll999l : IZibetti fc FergusonI
I2OO4I ') . iFrv et ahl l|l999l ') were unable to detect a halo around
NGC4244 to their limiting surface brightness and there-
fore their discussion of scattered light was rather moot.
(The NGC4244 halo has now been detect e d by resolved
star c ount measurements ; ISeth et al.l l2007l : Ide Jong et al.l
l2007l '). [zrbetti fc FergusonI l|2004l ') do however detect an extra
component around a disc galaxy in the HUDF and report

anomalous colours for this extended co mponent.

The HUDF galaxy studied by IZibetti fc FergusonI

(I2OOJ) is a highly inclined disc gala xy at a redshift of
about 0.32 with at best a small bulge. IZibetti fc FergusonI
(|200J) carefully removed the contribution from contaminat-
ing (mainly background) sources around the galaxy and ex-
tracted the luminosity profiles in wedges along the major
and minor axes. They determined profiles for all HUDF
HST/ACS passbands (F435W, F606W, F775W, F850LP,

termed B, V, i, and z hereafter for simplicity) and these
data points are reproduced in Fig.[T]

To estimate the contribution from scattered light at
lar ge radii I use a stan dard disc galaxy model (e.g. as used
by Ivan der Kruid Il988l ) . I represent the radial distribution
with an exponential disc projected edge-on, which accu-
rately matches the inner profile. However, the observed dis-
tribution shows a clear break in exponential scale length
at 2.9 arcsec independent of waveleng th. Such breaks have
been seen many time s before (e.g., Ivan der KruitI Il979l :
IPohlen fc Truiilldbood '). The light beyond the break I model
with another exponential distribution, but with shorter scale
length. The inner scale length varies from ~2.3 arcsec in
B to 0.9 arcsec in i and z. The scale lengths beyond the
break do not vary with wavelength within the uncertainties.
I model the vertical distribution with a sech light distribu-
tion, which is a reasonable compromise between the theoret-
ical sech^ distribution and the more centrally concentrated
exponential-like dist ribution observed in the near-infrared
(jde Griis et al.lll997l ). Choosing a different vertical profile
does not significantly change the results. Major and minor
axis profiles of this model before PSF convolution are shown
as dashed lines in Fig.[T]

To calculate the observed distribution from the model
distribution the PSF needs to be determined out to a radius
of about 10 arcsec. Such an extended PSF cannot be accu-
rately measured from the HUDF itself as there are too few
bright stars. I therefore used the Tiny Tim HST PSF mod-
elling software to create artificial PSFs for each passband.
These model PSFs were checked against some bright stars
in a number of unrelated F606W and F814W images and
found to match the outer profiles accurately to at least 5
arcsec. The model galaxy was rotated with respect to the
model PSFs by approximately the same amount as the real
galaxy. Cross-cuts through the centre of the Tiny Tim PSFs
in the direction of the galaxy major and minor axes are
shown for the F435W and F850LP fiffers in Fig.[T] Compar-
ison between the two filter profiles shows that the PSFs are
quite similar within 0.8 arcsec, but at larger radii they begin
to diverge. This disparity can lead to artificial colour gra-
dients when the convolved light distribution is significantly
contaminated or dominated by scattered light.

Finally, I show in Fig.[T] the model galaxy convolved
with the PSFs for all passband as thick solid lines. The ex-
tended minor-axis profile seen at heights greater than 1 arc-
sec from the midplane in the i and z-bands can be fully
explained by scattered light from the inner disc. On the
southeast (negative z) side there is a small excess in the V-
band that might be real, for instance due to small blue back-
ground galaxies seen here, but this component would have
very remarkable colour properties (e.g., V-i<0). It could
also be due to the asymmetry in the main disc light distri-
bution not modelled here or to some imperfections in the
PSF models at that scale. The B-band shows excess light
beyond my model on both sides of the disc that could be
a real halo detection, but these data points are very close
to the sky background limit and would again result in some
remarkable colours, this time mainly on the northwest side.
Clearly, a large fraction of the light seen on the minor axis
in excess of the sech distribution is due to scattered light
from the extended PSF.

The major axis profiles tell a different story. There

PSF tails and galaxy halo light 3

20

I 25

a;

, 30

35

jT^

- W J \ m-

^''/f / \ vf-^

'M^ Xv^

r arcscc

Figure 1. Minor (left) an d major (right) axis surface brightness profiles of the edge-on HUDF galaxy. The data points are derived from
IZibetti fc FergusonI l|2004h . The profiles for the different bands are coded in different colours and symbols and are offset in order to avoid
confusion, as indicated by the legend. The input galaxy model is shown by black dashed lines for the different passbands. Thin solid lines
indicate crosscuts through the B and ^-band PSFs (arbitrarily normalised), whilst thick lines show the convolved galaxy model using
the same colours for the different passbands as used for the data points.

seems to be a clear excess of light beyond 6 arcsec radius
compared to the truncated disc model. The PSF models
would need to be significantly wrong to explain this excess.
The points beyond 6 arcsec radius have large uncertainties,
but assuming they are correct, we cannot tell from these
measurements whether this light is due to a more spherical
halo, or due to the disc changing scale length yet again. It
does however show that, in cases where scattered light may
be significant, the best place to look for a (flattened) halo
is, maybe counter-intuitively, not along the minor axis, but
in the four quadrants away from the major and minor axes.
These are the areas where the contribution from the main
disc and from scattered light will be smallest.

Zibetti fc Ferguson (2004 ) also presented colour profiles
of this HUDF galaxy and argued that the halo had anoma-
lous colours starting at 1 arcsec above the disc, i.e., the
radius where the scattered light contribution starts to be-
come significant. While the observed colour profiles could in
principle be checked against the convolved model, it turns
out that the expected colour profiles depend critically on
the details of the calculated PSFs (especially the diffraction
spikes that are hard to model) and the assumed galaxy light
and colour distribution. The galaxy has clear small scale
colour variations due to dust extinction and local star forma-
tion that is not incorporated into the model. Therefore, the
colour gradients observed in the main disc are not predicted,
even though the PSF can induce some colour gradients at
scale height less than 1 arcsec. Additionally, the observed i-
z colour gradient cannot be explained in detail here, as the
PSFs only deviate significantly beyond 2 arcsec. Unfortu-
nately, both the galaxy and PSF models lack enough detail
to make firm colour predictions on such small scales, espe-
cially because I do not have the background galaxy mask
to select exactly the same halo areas. Because the light at
large radii is so much dominated by scattered light, even
small changes in PSFs, galaxy model, or masks can lead to

large changes in derived halo colour. The models are, how-
ever, consistent with the data, given the (large) errors of the
observations and taking these limitations into account. Scat-
tered light most likely contributes to the observed anoma-
lous colours.

3 SDSS IMAGE STACKING

IZibetti et al.l l|2004l ) combined 1047 images of edge-on galax-
ies to reach /i~31 r-mag arcsec"'^, with similar depths in the
g and i-band. The SDSS images were scaled, using a char-
acteristic scale size for the galaxies, and combined using an
approximation for the mode of the pixel values in the image
stack. The resulting galaxy light distribution was modelled
by a double exponential disc (exponential in radial and ver-
tical direction) convolved with the effective PSF, derived
from similarly stacked star images. An extra component in
addition to the convolved galaxy model was clearly detected
that followed a power-law radial l ight distribut i on wi th an
axial ratio of about 0.6. However, IZibetti et al.l (|2004l ) may
have underestimated the effective PSF at large radii, prob-
ably due to the procedure used to stack the stellar images
and due to the way they extrapolated the PSF profile to

larg e radii.

IZibetti et al.l (|2004l ) selected their edge-on galaxy sam-
ple from the "Large-Scale Structu re Sample 10" of galaxies
compiled by iBlanton et all l|2005l ) from the SDSS in April
2002 (about the size of SDSS DR2). Their sample selection
was based on cuts by minimum luminosity and isophotal
diameter as well as an isophotal flattening criterion. Visual
inspection removed further unsuitable candidates, reducing
the flrst cut sample of 1221 galaxies to a final sample of 1047.
After removal of contaminating sources the galaxy frames
were rotated to align the major axis and scaled radially to
a common characteristic size in the j-band. The final halo

4 Roelof S. de Jong

results were nearly independent of whether Petrosian, half-
light, isophotal radii or exponential scale lengths were used
as reference, and I will use Petrosian r adii from her e on.
As a common reference for radial scaling IZibetti et alj used
the median scale size of the sample as expressed in pixels,
where each pixel is 0.396 arcsec in the SDSS. The images
were combined using an approximation of the mode of all
pixel values in the stack at a given position:

approximate mode = 3 x median — 2 x average.

(1)

where the median and average images were created from
the median and (sigma clipped) average values of the stack

values.

IZibetti et al.l (|200J) followed the same procedure to cal-
culate an effective PSF for the image stacks. They present
their final stacked PSFs for 4 passbands as modelled by a
Gaussian core and an exponential tail. However, their PSF
models are presented out to only 15-22 pixels (depending
on filter), while their observations stretch to about 100 pix-
els. The raw PSF profiles are not shown and we cannot be
sure whether the exponential tail is a good approximation
at radii larger than 20 pixels. I have therefore recreated the
stacked PSFs in order to calculate galaxy models out to the
last measured point.

3.1 Determining SDSS PSFs

The largest galaxies in the sample are several times larger
than the median value used for scaling reference and so both
the galaxy and associated PSF will be shrunk by a large fac-
tor before stacking. Thus, to create a stacked PSF out to 100
pixels, as used in the galaxy stacks, we need to determine
the PSF on the individual frames to even larger radii. To
accomplish this, we need to combine the profiles of both
saturated an d unsa turated stars. From the sample used by
IZibetti et al.l (|200J) I selected at random 20 galaxies and
downloaded their full SDSS frames in all passbands. From
these frames I extracted the azimuthal median radial lu-
minosity profiles of a bright but unsaturated star and the
brightest isolated star, which was always saturated at the
core. By matching these profiles in the high signal-to-noise
overlap region I reproduced one very extended PSF profile,
often reaching 200 pixels in radius and 16 magnitudes below
peak brightness.

Comparing all 20 stellar profiles by normalising the fiux
within 10 pixel radius, I find they are well described by a
central, Gaussian-like core surrounded by a faint envelope
(Fig.[21l . The width of the core depends on camera column
and time, i.e. the camera optics and the blurring by atmo-
spheric seeing. The surrounding envelope is very similar in
shape across all frames, with a brightness that depends only
on the total brightness of the star. This halo is most likely
caused by scattering off of dust particles on the telescope
optics. The shape and brightness of the halo is the same
in all filters with a near power law profile of slope about
-2.6. The only exception is the i-band filter, which shows
an additional exponential component dominating between
35 and 180 pixels. This feature is typical for backside illu-
minated thinned CCD s as de scribed in more detail in for
instance ISirianni et al.l (|2005r ). This extra PSF component
in the i-band filter explains the red haloes seen around all
bright stars in SDSS colour images, irrespective of the star's

--^

.

'^^fc

2

^

4

-

\ ^

^ 6

\ \

-

o

\ ^

'-•■.■. \ *

S 8

-

J^ ,

•- \*.v">

3,

<i 10

-

%

*

-, \. >

^*^

12

^\\/

14

-.

\^^

3 10 30

r [pixel]

100

Figure 2. SDSS Point Spread Functions colour coded for the g, r,
and j-bands as indicated by the legend. Solid lines show typical
SDSS PSFs derived from faint and bright stars as described in
the text. Dashes lines show the average PSFs after scaling and
stacking. The dotte d lines show the stacked PSFs as presented by
IZibetti et"al] l l2004h .

intrinsic colour. The additional i-band scattered light is, as
I will show below, important when measuring colours of dif-
fuse haloes around edge-on galaxies.

I determine the final radial PSF profile for each filter by
taking the weighted mean of all star profiles (irrespective of
central core variations) . The weights were set by the square
root of the relative brightness of the bright stars used to
determine the outer halo of the PSF. This results in PSF
profiles with S/N>2 out to about 180 pixels. This radial ex-
tent is still not enough, as the scale sizes of some galaxies are
6 times larger than the median value and for these galaxies
the PSF will be scaled down to approximately 30 pixels ra-
dius. I chose to extend the profiles beyond 180 pixels with
a power law of slope -2.6 as seen between 40 and 180 pixel
radius. Choosing for instance an exponential profile with a
sharper cut off does not make a significant difference to the
final stacked PSF.

3.2 PSF stacking and effective PSFs

As far as p o ssible I have reproduced the procedure of
IZibetti et"al] (|2004h to create effective PSFs. I applied the
same selection criteria to the Large-Scale Structure Sample
10 (minus eye inspection), used the Petrosian radii to scale
the ID PSFs radially, and used the central galaxy luminosi-
ties to scale the PSFs in intensity. The radial scaling was
done so as to conserve surface brightness, not total luminos-
ity of the stellar PSF profile, as was done for the galaxies.

When I combine these scaled PSFs with the mode ap-
proximation of Eq.[T] I get negative values at most radii.
The empirically derived mode approximation of Eq.[T] only
produces reasonable results if the value distribution is not

PSF tails and galaxy halo light 5

z [pixel]

z [pixel]

z [pixel]

Figure 3. Light profiles parallel to the minor axis at four dist ances from the ga laxy centre for the g-band (left), r-band (middle), and
i-band (right). Symbols represent the data as presented by iZibetti et alj 1120041) with distances from the centre and offsets in surface
brightness indicated by the legends. The dashed line shows the minor axis light profile of the galaxy model before convolution. The
dotted line shows the average stacked SDSS PSF. The thick coloured solid lines show the vertical cuts at the four different radii through
the convolved disk only model, the thin lines have an added Sersic profile halo as described in the text.

too skewed teen dall fc StuartlHoTTJ ') . While this is probably
true for the scaled galaxy stack (assuming that the galaxies
are self-similar), this is not true for the scaled stars in the
absence of noise. The apparent scale size distribution of the
galaxies is strongly skewed to small angular sizes, and hence
the scaled stellar profile intensities will be strongly skewed,
exacerbated by the steep profile shape. PSFs stacked with
mode approximation are clearly a poor representation of the
true effective PSF. Using the median intensity at each ra-
dius instead results in a profile nearly identical to the input
PSF. This is to be expected as everything is scaled to the
median characteristic galaxy scale size and just as many
PSFs are scaled smaller as are scaled larger. The intensity
scaling is a small effect compared to the radial scaling due to
the steepness of the PSF profile. Using the average value at
each radius results in a profile that is a smoothed version of
the original PSF, with some of the central light distributed
to larger radii.

Figure[2] shows typical SDSS PSFs for the g, r, and i-
bands (which are, as mentioned before, virtually identical to
the median stacked PSFs), together with the av erage stacked
PSFs and the stacked PSFs as determined bv lZibetti et al.l
(I2OOJ) using the mode approximation. The average stacked
PSFs are clearly more extended than th e original (an d hence
median stacked) PSFs. However, the IZibetti et ahl (|2004l )
stacked PSFs are much more compact than the typical PSFs
of SDSS images, especially between 3 and 10 pixels and at
radii larger than 20 pixels. This is unlikely to happen when
stacking radially scaled PSFs with conservation of surface
brightness. The compactness of their PSFs at small radii
therefore seems to be the result of the mode approximation
that is incorrect for such skewed intensity distributions. The
sharper cutoffs of their PSFs at larger radii is due to the ex-
ponential extrapolation where a power law seems more ap-
propriate. It is not entirely clear which effective stacked PSF
method is most representative of the final galaxy stack. The
only correct way of doing this is to create models of each

individual galaxy, convolved with the image PSF, and stack
these models. This is beyond the scope of this paper and
I will use the averaged stacked PSFs as the best represen-
tation of the final galaxy stack PSF. In the analysis in the
next section I will indicate how galaxy halo measurements
are affecte d by using a d ifferent choice of PSF (i.e., median
stacking or lZibetti et al.i (,2004. ) stacking).

3.3 Modelling SDSS stacked galaxy profiles

In Fig.Ol show the data from figure 6 of lZibetti et"aLl l|2004 ).
which shows the luminosity profiles of the stacked galaxy
images parallel to the minor axis at four distances from the
centre. I model these light distributions with a similar model
galaxy as used for the HUDF case, with an exponential ra-
dial distribution and a sech vertical distribution. In this case
I do not include a break in the radial luminosity distribu-
tion, even though I will show it is necessary to explain the
light profiles. For simplicity I use the same scale length and
height for all filters, even though the fit could be somewhat
improved by varying the scale lengths with wavelength. The
minor axis input light profile is shown, as well as the effective
PSFs used to convolve the models. Finally, I show profiles
of the convolved 2D model at each of the four vertical cuts.
The models show that significant amounts of light are
found at large radii due to PSF scattering. This is especially
true for the i-band, where a large fraction of the light at
heights greater than 15 pixels above the midplane could be
due to scattered light. The cuts farther away from the centre
and profiles in the g and r-passbands have more light in ex-
cess of the scattered light profiles, but are still often within
a magnitude (a factor ~2) of the model profile. Between 15
and 40 pixels above the midplane, where the data are most
reliable, about 50% of the light comes from scattered in g
and r, increasing to about 80% in i. Clearly, any accurate
modelling of this excess light requires a thorough under-
standing of the scattered light profile. The most successful

6 Roelof S. de Jong

Figure 4. The minor axis colour profiles. The symbols with error-
bars show the q — r (orange diamonds ) and r-i (red circles) data
as presented by IZibetti et "aD l|2004l ). The PSF convolved disk
only models are represented by thick orange dotted {g—r) and red
solid (r— i) lines, thin lines have an added Sersic profile halo as
described in the text.

detection is most likely to occur in the four quadrants away
from the major and minor axes. The model over-predicts the
luminosity near the midplane of the profile farthest from the
centre, whilst matching the data near the midplane for the
other three radii. A break in the radial profile is necessary
between the third and the fourth profiles to explain the hght
distribution. Such a break will somewhat increase the sig-
nificance of the halo detection in the outer most profile.

Using the median effective PSFs, the model profiles
are still within 1 mag of the data. The median stacked
PSF yields a slightly larger dip between 10 and 30 pixels,
which is smo othed out when using average stacking. The
IZibetti et al] (2004) model PSFs do not explain the hght
at radii larger than 30 pixels as they cut off exponentially,
but at these larger radii their effective PSF is unlikely to be
correct.

The remaining difference between the observations and
the convolved disk model can be explained by adding a halo
of a for m that is typical of what is found by the GHOSTS
survey l|de Jong et al.ll2007l . de Jong et al., in preparation).
The GHOSTS survey measures halo shapes from HST star
counts not effected by PSF convol ution. The ty pical halo
envelope can be parameterised by a lSersid (|l968l ) profile:

EB(r) = Ecffe'

^6r.([r/rrffll/"_i)

(2)

with n=5.5, the effective radius enclosing half the luminos-
ity ('"eft) equal to a third of the disk scale length, and the
effective surface brightness at this radius (Ecff) equal to a
tenth of the central surface brightness of the disk. The halo
has a fiattening of vertical-over-major axis of 0.65. The re-
sult of adding this additional halo component to the model
is shown by the thin solid lines in Fig.|3] This shows that
the remaining excess can indeed be explained by a halo pro-
file that is typical for nearby massive disk galaxies. However,
due to the uncertainty in the PSF convolution it will be very
challenging to invert this process and derive halo parameters
fro m the integrat e d ligh t measurements.

IZibetti etld] (|2004 ) found that the detected halo hght

showed very anomalous colours, especially in r-i. In Fig.|4]
I reproduce their minor axis colour profiles with my PSF
convolved model. The colour gradient seen in ) — i can be
fully ascribed to the difference in PSF between r and i, as
the intrinsic light distribution has no colour gradient at all.
The anomalous extra scattered light component in the i-
band PSF causes the strong colour gradient. The very good
agreement between r-i model and data strongly suggests
that the average stacked PSF is a fair approximation for the
effective PSF. The model also predicts a colour gradient in
g-r between 10 and 35 pixels. No clear trend is seen in the
data, but the errorbars at these radii are so large that the
model is consistent with the data. On the other hand, only
about 50% of the hght is due to scattering in these passbands
according the model, therefore the observed colour gradient
is expected to be less than the model prediction if there is
no intrinsic colour in the galaxy. Indeed, adding the typi-
cal GHOSTS Sersic halo to the models as described above
reduces the colour gradients, relatively more in g-r than in
1 — i. This shows that the scattered light still dominates the
colours in i — i for a reasonable halo profile.

Using the median stacked PSFs instead of the average
stacked PSFs yields model colour gradients on the minor
axis that are much steeper beyond 10 pixels than observed.
Such a steep gradient would need to be compensated by
a significa nt real halo component with no colour gradient.
The Zib ctti et al.l ((2004) PSFs produce a very steep colour
gradient starting at 5 pixels and a real halo would need to
dominate at these radii to reduce the scattered hght colour
effect.

4 CONCLUSIONS

Measuring the exact halo hght distribution around edge-on
disc light from the integrated light distribution is very chal-
lenging. In addition to careful fiat fielding and removal of
fore- and background sources, particular attention has to be
paid to the effect of scattered light from the central disc.
Specifically, for edge-on galaxies it is not sufficient to com-
pare the galaxy minor axis light profile with the PSF profile;
at large heights above the disc scattered light is contributed
from the whole galaxy, not just the centre, and a full convo-
lution of a 2D model is needed to assess the scattered hght
contamination. Basic modelling suggests that the minor axis
may not be the optimum place to detect a halo, but for a
fiattened halo one should look in the quadrants away from
the major and minor axis where scattered light contribution
from the disc is smaller relative to the halo.

I find that the extended component seen along the
minor axis of an HUDF edge-on galaxy can almost
fully be explained by scattered light. The observations of
Zibet ti fc Ferg uson (,200.4|) for this galaxy do show an ex-
tended component along the major axis that cannot be
explained by scattered light. Similarly, the halo detec-
tion around stacke d SDSS edge-on galaxies reported by
IZibetti et al.l l|2004l ) has a significant scattered light contam-
ination, especially along the minor axis and in the i-band.
Depending on the assumed best method to create a stacked
PSF the contamination along the minor axis in the i-band
can amount to 80% of the light, in the g and r-bands 50%.
The contamination decreases with other PSF assumptions

PSF tails and galaxy halo light 7

and going to one of the quadrants, but the PSF uncertain-
ties make it hard to derive accurate stellar halo properties.
The excess light seen on top of the disk only model is consis-
tent with a Sersic law halo wit h parameters typical of those
found by the GHOSTS survey l|de Jong et al.ll2007l . de Jong
at al., in preparation).

The anomalou s min or a xis halo colours rep orted by
IZibetti fc FergusonI (|2004 ) and'Zibct ti et all (|2004 ) are con-
sistent with originating from scattered light c ontamination.
In par ticular, the colour gradients reported bv lZibetti et al.l
i|2004l) can be modelled by an extended PSF as calculated
from the average of the observed scaled SDSS PSFs. There
seems to be no need to invoke extreme s tellar population
to exp lain these colours as proposed by IZackrisson et al.l
l|2006l ). The expected contribution from a typical GHOSTS
halo is so small in the i-band that it has only a small effect
on the predicted colour gradient.

There have b een other reports of anomalo us galaxy stel-
lar halo colours. iLequeux et al.l (|l996l , Il998l ) report thick
disk or halo _BV7-band colours of edge-on galaxy NGC5907
that are only consistent with colours of elliptical galaxies,
i.e. an old, metal rich stellar population. In this case the er-
rors are unlikely to be due to scattered light; the thick disk
is detected at about a factor 10^ below the central bright-
ness at about 80 arcsec above the midplane, while the PSF
is already lO'^ below peak brightness at 16 arcsec radius.
Not even a factor 10 enhancement of the scattered light
effect due to the edge-on disc configuration can make up
that difference. Similarly, the red haloes reported around
Blue Compact Galaxies (see e.g., IZackrisson et al.l |2006| .
and references therein) are not caused by scattered light.
The galaxies are too large compared to the PSF and the
haloes show substructure different from the central galaxy
that is unlikely to be due to the substr ucture in the PSF.

Unfort unately, the examples of IZibetti fc FergusonI

(|2004l ') and IZibetti et all (|2004l ') are close to the limit where
scattered light is no longer significant and slightly larger
objects would have avoided problems. The HUDF does not
contain a larger edge-on galaxy that would render scattered
light unimportant. However, the SDSS study could now be
repeated on a sample of larger galaxies as the area surveyed
by SDSS has nearly tripled since the IZibetti et al] (|2004l ')
study. Ideally, such a study would use only a small range
of scale sizes to avoid large spatial scaling corrections and
uncertainties in the effective PSF. Taking about a two times
as large an isophotal selection limit would probably suffice,
especially in the g and r-bands.

While measuring halo properties from integrated light
remains plagued with large uncertainties, we do have a tech-
nique that allows us to accurately measure the stellar en-
velopes around galaxies. By performing star counts of re-
solved stellar populations we can measure equivalent sur-
face brightnesses to very faint li mits as spectacul arly demon-
strated for M31 and M33 (e.g.. Ilbata etal .1120071 '). These ob-
servations are not affected by scattered light and only min-
imally by fiat fielding errors. The main limitations of this
method lie in fore- and background contaminating objects
and low number statistics (too few brights stars at very low
surface brightnesses/densities). Very few massive, highly in-
clined galaxies can be studied with ground-based resolution,
but HST allows a much larger sample of galaxies to be stud-
ied. Indeed, preliminary results from the GHOSTS survey

show conclusively that most large galaxies do have very ex-
tended stellar envelopes, with envelope size likely depending
on galaxy mass and bulge-to-disc ratio (Ide Jong et al.ll2007l ,
de Jong et al., in preparation).

ACKNOWLEDGEMENTS

I thank Stefano Zibetti, Annette Ferguson, and Simon White
for very constructive comments on an earlier version of this
manuscript that improved its final quality. I am grateful to
Michael Blanton for providing a machine readable version of
the SDSS Large-Scale Structure Sample 10. I thank David
Radburn-Smith for a critical reading of a previous version
of this manuscript. Eric Zackrisson provided feedback that
improved my appreciation of the intricacies of the red halo
phenomenon. I acknowledge Eric Bell for making this paper
see the light of day.

Funding for the SDSS and SDSS-II has been provided
by the Alfred P. Sloan Foundation, the Participating In-
stitutions, the National Science Foundation, the U.S. De-
partment of Energy, the National Aeronautics and Space
Administration, the Japanese Monbukagakusho, the Max
Planck Society, and the Higher Education Funding Council
for England. The SDSS Web Site is http://www.sdss.org/

The SDSS is managed by the Astrophysical Research
Consortium for the Participating Institutions. The Partic-
ipating Institutions are the American Museum of Natu-
ral History, Astrophysical Institute Potsdam, University of
Basel, University of Cambridge, Case Western Reserve Uni-
versity, University of Chicago, Drexel University, Fermilab,
the Institute for Advanced Study, the Japan Participation
Group, Johns Hopkins University, the Joint Institute for
Nuclear Astrophysics, the Kavli Institute for Particle As-
trophysics and Cosmology, the Korean Scientist Group, the
Chinese Academy of Sciences (LAMOST) , Los Alamos Na-
tional Laboratory, the Max-Planck-Institute for Astronomy
(MPIA), the Max-Planck-Institute for Astrophysics (MPA),
New Mexico State University, Ohio State University, Uni-
versity of Pittsburgh, University of Portsmouth, Princeton
University, the United States Naval Observatory, and the
University of Washington.

REFERENCES

Abadi M. G., Navarro J. F., Steinmetz M., 2006, MNRAS,

365, 747
Beckwith S. V. W., Stiavelh M., Koekemoer A. M., Cald-
well J. A. R., Ferguson H. C, Hook R., Lucas R. A.,
Bergeron L. E., Corbin M., Jogee S., Panagia N., Rob-
berto M., Royle P., Somerville R. S., Sosey M., 2006, AJ,
132, 1729
Bekki K., Chiba M., 2005, ApJL, 625, L107
Blanton M. R., Schlegel D. J., Strauss M. A., Brinkmann
J., Finkbeiner D., Fukugita M., Gunn J. E., Hogg D. W.,
Ivezic Z., Knapp G. R., Lupton R. H., Munn J. A., Schnei-
der D. P., Tegmark M., Zehavi I., 2005, AJ, 129, 2562
Bullock J. S., Johnston K. V., 2005, ApJ, 635, 931
de Grijs R., Peletier R. F., van der Kruit P. C, 1997, A&A,
327, 966

B Roelof S. de Jong

de Jong R. S., Radburn-Smith D. J., Sick J. N., 2007, ArXiv
e-prints, 07105511

de Jong R. S., Seth A. C, Radburn-Smith D. J., Bell
E. F., Brown T. M., Bullock J. S., Courteau S., Dalcanton
J. J., Ferguson H. C, Goudlrooij P., Holfeltz S., Holwerda
B. W., Purcell C, Sick J., Zucker D. B., 2007, ApJL, 667,
L49

Ferguson A. M. N., Irwin M. J., Ibata R. A., Lewis G. F.,
Tanvir N. R., 2002, AJ, 124, 1452

Fry A. M., Morrison H. L., Harding P., Boroson T. A.,
1999, AJ, 118, 1209

Ibata R., Martin N. F., Irwin M., Chapman S., Ferguson
A. M. N., Lewis G. F., McConnachie A. W., 2007, ApJ,
671, 1591

Ibata R. A., Gilmore G., Irwin M. J., 1994, Nature, 370,
194

Kendall M., Stuart A., 1977, The advanced theory of statis-
tics. Vol.1: Distribution theory. London: Griffin, 1977, 4th
ed.

Lequeux J., Combes F., Dantel-Fort M., Cuillandre J.-C,
Fort B., Mellier Y., 1998, A&A, 334, L9

Lequeux J., Fort B., Dantel-Fort M., Cuillandre J.-C, Mel-
lier Y., 1996, A&A, 312, LI

Morrison H. L., Boroson T. A., Harding P., 1994, AJ, 108,
1191

Pohlen M., Trujillo I., 2006, A&A, 454, 759

Purcell C. W., Bullock J. S., Zentner A. R., 2007, ApJ,
666, 20

Sersic J. L., 1968, Atlas de galaxias australes. Cordoba,
Argentina: Observatorio Astronomico, 1968

Seth A., de Jong R., Dalcanton J., The Ghosts Team
2007, in Vazdekis A., Peletier R. F., eds, lAU Sympo-
sium Vol. 241 of lAU Symposium, Detection of a stellar
halo in NGC 4244. pp 523-524

Sirianni M., Jee M. J., Bem'tez N., Blakeslee J. P., Martel
A. R., Meurer G., Clampin M., De Marchi G., Ford H. C,
Gilliland R., Hartig G. F., Illingworth G. D., Mack J.,
McCann W. J., 2005, PASP, 117, 1049

van der Kruit P. C, 1979, A&AS, 38, 15

van der Kruit P. C, 1988, A&A, 192, 117

Wu H., Burstein D., Deng Z., Zhou X., Shang Z., Zheng
Z., et al. 2002, AJ, 123, 1364

Zackrisson E., Bergvall N., Ostlin G., Micheva G., Leksell
M., 2006, ApJ, 650, 812

Zibetti S., Ferguson A. M. N., 2004, MNRAS, 352, L6

Zibetti S., White S. D. M., Brinkmann J., 2004, MNRAS,
347, 556