Äîêóìåíò âçÿò èç êýøà ïîèñêîâîé ìàøèíû. Àäðåñ îðèãèíàëüíîãî äîêóìåíòà : http://www.stsci.edu/ftp/instrument_news/FOC/Foc_isr/foc_isr083.ps Äàòà èçìåíåíèÿ: Tue Feb 28 22:59:39 1995 Äàòà èíäåêñèðîâàíèÿ: Sun Dec 23 05:41:49 2007 Êîäèðîâêà: Ïîèñêîâûå ñëîâà: arp 220

INSTRUMENT SCIENCE REPORT
FOC­083
TITLE: The Photometric Accuracy Of The FOC DATE: 28 February 1995
AUTHOR: Gerhardt R. Meurer (The Johns Hopkins Universiy)
ABSTRACT
A comparison is made between the expected photometric accuracy of the FOC, and that ob­
served. The presently known systematic effects are listed and quantified. Typically when both
internal and external factors are included one only expects to obtain photometry accurate to
¸0.2 mag in the ultraviolet. This expectation is tested by comparing magnitudes extracted
from F220W images of nine starburst galaxies with those extracted from IUE spectra. After
removing one discrepant observation, it is found that relative to the IUE, the FOC external
errors are Ÿ 0:05 mag, three times lower than expected. By carefully removing some of the
internal zeropoint variations, one should be able to reduce the total systematic errors to ¸ 0:10
mag.
0 DISTRIBUTION:
FOC Project: B. G. Taylor, R. Thomas
IDT: R. Albrecht, C. Barbieri, A. Boksenberg, P. Crane, J.M. Deharveng, M.J. Disney, P. Jakobsen,
T. Kamperman, I.R. King, C. Mackay, G. Weigelt
SIB/SOB C. Cox, P. Greenfield, M. Lallo, W. Hack, A. Nota, S. Osmer, F. Paresce,
All Instrument Scientists
SCARS: P. Hodge
SESD: M. Miebach, W. Safley
SPD: J.C. Blades, R. Jedrzejewski, F. Macchetto
RSB: D. Gilmore, D. Golombek
USB: A. Saha
Me mates: D. Maoz, E. Zirbel

1 Introduction
Since becoming involved with FOC data, I've been told that one can only expect to extract
photometry accurate to about 20%. Here I test this bit of folklore in two ways. First I attempt
to list all factors that may systematically effect the photometric accuracy of the FOC, and
quantify (where possible) the magnitude of the effects. I then compare FOC magnitudes of
starburst galaxies with IUE measurements. I find that a formal accounting of likely systematic
errors does indeed suggest only ¸ 20% photometry is typically obtainable when both external
and internal errors are considered. However, I come to pleasant conclusion that for extended
objects and through one particular filter, the actual external errors are !
¸ 5%, about 1/3 of
that expected. The total systematic errors should be able to be reduced to !
¸ 10% through
careful removal of some of the sources of ``internal'' systematic errors -- those that cause spatial
variations in the photometric performance.
2 Expected sources of systematic errors
Here I list everything I can think of that may be a source of systematic uncertainties in pho­
tometry with the FOC. By ``systematic'' I mean effects other than the limits placed by the
number of photons detected. Some of these effects can be removed, whereas others can not.
The discussion is weighted towards the data I am working with: pre­COSTAR f/96 data, using
the 512z \Theta 1024 format, and F220W filter. When discussing stellar aperture photometry, I
adopt an extraction aperture of radius r = 4 pixels, with background subtraction taken as the
mode of values in an annulus of radii r = 4 \Gamma 7 pixels.
1. Pipeline sensitivity calibration. The absolute flux calibration is given by the PHOT­
FLAM keyword, whic is inserted in the image headers during pipeline calibration. It is
determined from the in­orbit detector quantum efficiency (DQE) curve as a function of
wavelength, with the assumption that the source spectrum is flat. The rms of the DQE
calibration of Sparks (1991) is 7%. However, there are star to star offsets in the calibra­
tion of up to ¸ 20%, especially in the far UV. Over the central \Delta– = 2W 50 of the F220W
filter the rms of the residuals is ¸13%.
2. Deterioration of detector sensitivity with time. There is a very slow deterioration
with time of the DQE. For the F210M filter (close enough in wavelength to the F220W
filter for our purposes) the decay is about 1.4% per year, measured over a base line of
nearly three years (Greenfield, et al., 1993). The pipeline calibration is that given by
Sparks (1991) using observations taken around 1990.9.
3. Flatfielding Errors. For a given image, the flatfield used by the pipeline data reduction
is a highly smoothed version (typical smoothing lengths of 15 pixels) of either one of the
high resolution flatfields of Greenfield (1992) or other internal LED flats. Only a limited
number of filter combinations are included in the archival flats. There are four causes of
flatfielding uncertainties:
(a) Flatfield smoothing. The smoothing of the flatfields is done because of the difficulty
in obtaining high S/N flatfields and the difficulty in aligning the data and flatfield.
2

Photometric errors will then occur due to the low resolution of the flatfielding. I
have measured the dispersion in the pixel values in the central region of the high
resolution UV flat image of Greenfield (1992) and find oe=mean = 0:06. Individual
features commonly have depths or peaks of 15%, with some features being virtually
black.
(b) Flatfield color mismatch. This occurs frequently because of the limited number of
filter combinations for which pipeline flatfields exist. I examined the ratio of the
pipeline blue and UV flats. In the central 512 \Theta 512 region the maximum deviation
from the mean ratio is 10%, and oe=mean = 0:04.
(c) Un­modeled vignetting at the edge of frames reduced with internal flats. F220W
observations use the ``sky'' UV flatfield of Greenfield (1992), so they do not suffer
from this problem. Flatfield vignetting is expected to be worse for the f/48 camera
than the f/96.
(d) Turn­on effects. Greenfield (1992) notes that shortly after turn­on the shape of the
flatfield can change by ¸ \Sigma2%. The turn­on effect is most clearly seen in the first
exposure of both the f/96 and f/48 series of Greenfield. These each commenced
1.4 hours after high voltage (HV) turn­on (UT times were kindly obtained for me
by P. Hodge). The second exposure in each series began 2.7 hours after HV turn­
on; the effect is not apparent in the f/48 exposure but weakly evident in the f/96
exposure. The turn­on effect may then be important for frames obtained before the
FOC has warmed up for two to three hours. Note that the nominal warm­up time
is 75 minutes, so some science frames will likely be affected.
4. Format dependent sensitivity variations. The absolute flux scale is determined for
the 512 \Theta 512 format. However, other formats have different sensitivities perhaps because
of their different scan rates (changing the effective pixel size; Greenfield 1994b). The
512z \Theta 1024 format is more sensitive than the 512 \Theta 512 format by a factor of 1:25 \Sigma 0:03.
The error is the external error from the two methods used to determine the level of this
variation. In addition this factor varies spatially with a dispersion of 0.035 over the area in
common between the two formats. The spatial variations have been mapped and should
be able to be removed.
5. High frequency pattern noise. There are two types of high frequency (non­fixed)
pattern noise: diagonal stripes having a period of 3.35 pixels and an amplitude of ¸5%;
and vertical stripes having a period of 4 pixels and amplitude of ¸ 2:5% (Nota et al.,
1993). King et al. (1994) present an algorithm for removing this nose in the Fourier plane.
6. Geometric distortion uncertainties. The distortion pattern of the F96 relay is not
stable. The detector distortion changes rapidly during the first 30 minutes after warm­up,
and more slowly thereafter; the distortion pattern is still slowly changing eight hours after
turn­on (Baxter, 1990). In addition there is a long term drift of a few pixels over several
years (Baxter, 1993). Any variations that change the plate scale will alter the photometric
zero­point. However, the long term variation in plate scale is so small (¸ 0:2%) that the
zeropoint drifts are negligible (0.004 mag). The variation in zeropoint will not be uniform,
3

and likely be worse at the edges of the detector where the geometric distortion is largest
and least well defined.
7. Flatfield non­linearity. This is the non­linearity over spatial scales of about 10 pixels
or larger. The amount of the non­linearity is measured by Jedrzejewski (1992) who also
shows that there is a positional variation due to the change in electron focus across the
photocathode. For the 512z \Theta 1024 format of the f/96 camera the linearity parameter a
has a dispersion of 24% about its mean value. Baxter (1994a,b) presents an algorithm
to remove non­linearity. Since the positional variation in a has been mapped, a position
dependent linearity correction is feasible.
8. Point­source non­linearity. Point sources and extended sources have different linearity
responses due to the intricacies of how the onboard pattern recognition logic distinguishes
photon from ion events and noise. At a given count rate per pixel, the cores of point
source are less affected by non­linearity than are extended sources. Baxter (1994a,b) and
Greenfield (1994a) discuss methods of correcting for the point­source component of non­
linearity. Baxter notes a format dependent difference in response to point sources relative
to extended structure, with point sources having 90% of the response relative to diffuse
structure in zoomed mode images. Greenfield does not find such a difference. It may be
that this additional response difference is an artifact of the two step linearity correction
algorithm of Baxter (Greenfield's approach involves one step). The results of Baxter and
Greenfield will probably have to be recalibrated for data taken with COSTAR. For the
512z \Theta 1024 format, Baxter finds no indications that non­linearity has become important
for integrated countrates up to ¸ 4:1 Hz in 5 \Theta 5 centered boxes. This corresponds to
¸ 4:8 Hz in the adopted aperture after background subtraction.
9. Focus changes. The focus drifts from the nominal focus due to desorption of the OTA
(Optical Telescope Assembly) and ``breathing'' (Baxter et al., 1993). The 3­4¯m focus
changes due to breathing have a noticeable effect on the PSF of stars (Baxter et al., 1993).
A PSF mismatch would cause a systematic uncertainty in stellar photometry using either
of the two most popular techniques:
(a) Aperture photometry. Usually aperture photometry with HST data is done with
a fairly small aperture (radius of a few pixels), with background subtraction from an
adjoining annulus. With pre­COSTAR data the aperture corrections can be quite
large, ¸ 2 mag or more. To estimate the changes in aperture correction, I used the
TinyTim software (v4.0 Krist 1993) to make synthetic F220W PSFs (pre­COSTAR)
where the focus differed by 4¯m. The PSFs were calculated for a B star spectrum
normalized to unity total flux. The difference in magnitude, and thus aperture
correction, is 0.04 mag for the adopted aperture.
(b) Core profile fitting. This is the method used by packages such as DAOPHOT
and DoPHOT. Since the profile is only fit out to a limiting radius, an aperture
correction will have to be applied, and it will be uncertain to PSF mismatches as
discussed above. The cores will also be affected by a PSF mismatch that becomes
more noticeable (in terms of ü 2 ) for brighter stars. I have not modeled this effect,
but examination of the cores of the two model PSFs generated above, suggests the
4

differences are not very large. Intuitively, the maximum error should be on the order
of the aperture correction difference, for a given fitting radius.
With COSTAR the PSF will vary with position in the field due to the tilt in the new
FOC focal plane and astigmatism (Jedrzejewski et al., 1993). In addition to telescope
focus effects, the electron focus across the photocathode varies across the photocathode
plate (Jedrzejewski, 1992), as mentioned under item 7. It is not clear how this will affect
photometry.
10. Red leak (color term). Flux errors can arise from the spectral width and shape of the
filter in two ways:
(a) The pipeline calibration assumes a flat spectrum, and will yield the wrong flux
density for objects with different spectra. Most UV filters have some problems
with ``redleak''; that is the filter response drops off slowly to the red. I used the
IRAF/STSDAS package SYNPHOT to estimate the redleak expected for the F220W
filter. Spectra of stars covering a wide range in effective temperature (4000 !
¸ T eff
!
¸
50000 K) were measured. I define the red leak as the STMAG difference between the
F220W measurements and that within a box filter having them same nominal central
wavelength and width (2270 š A and 482 š A respectively). I find that redleak ! 0:05 for
T eff ? 8000K.
(b) When doing stellar photometry, if the color of the PSF star does not match the
stars being measured, their PSFs will differ. I again used TinyTim to estimate the
magnitude of this effect, this time generating PSFs for an O star and an M star, with
the other model and aperture parameters as above. Their core magnitudes differ by
0.02 mag.
Table 1 summarizes the expected variances, in magnitudes, for the particular case of pre­
COSTAR f/96 observations with the 512z \Theta 1024 format, and with F220W being the only
color filter. The variances are split into two categories. The first under the column heading
``external'' are the uncertainties induced in the overall zeropoint of a frame. These are the
only relevant errors when one integrates over large enough areas that small scale variations
are smoothed over. Under the column ``internal'' is listed the uncertainties resulting from the
position of the object in the frame or the exact PSF shape. These are the uncertainties that
apply to relative photometry of stellar sources within a frame. When doing frame to frame
comparisons of stellar sources both the external and internal errors must be considered. When
only external errors are considered one should only expect to be able to do photometry accurate
to 15%. If internal errors are also important only ¸ 20% photometry is achievable.
3 Observations
The reason for going through the above exercise is to determine the expected accuracy of
FOC observations of starburst galaxies. These were obtained with the 512z \Theta 1024 format of
the f/96 camera and employed F220W and occasionally F1ND filters. The program galaxies
were selected from the Kinney et al. (1993; hereafter K93) IUE atlas of star­forming galaxies.
5

Table 1: Expected photometric performance
item external internal notes
1 0.13 : : :
2 0.01 : : : 1
3a : : : 0.06
3b : : : 0.04
3c : : : 0.00
3d : : : 0.02
4 0.02 0.03 2
5 : : : 0.06 2
6 0.00 0.00:
7 0.03 0.03 2,3
8 : : : 0.10: 2
9 : : : 0.02 4
10a 0.05 0.05 5
10b 0.01 0.01
sum 0.14 0.15
Notes :
1. Estimated by eye from fig. 2 of Greenfield et al. (1993)
2. These internal errors can largely be eliminated; here I assume that they are not.
3. Evaluated at nominal non­linearity of 20%, assuming that the linearity parameter a has an
uncertainty of 24%, both as an overall external uncertainty, and due to spatial variations.
4. Assumes \Sigma2¯m focus uncertainty.
5. Assumes source spectrum has T eff ? 8000K.
6

Table 2: Observing log and Measurements
Galaxy Rootname Filters \Deltat C sky m FOC m IUE
(s) (10 \Gamma4 Hz=pixel) (mag) (mag)
1Zw18 x19p0201t F220W 1497 5:50 \Sigma 0:08 14:49 \Sigma 0:05 14:67 \Sigma 0:15
NGC 1705 x19p5101t F220W,F1ND 497 8:65 \Sigma 0:15 11:82 \Sigma 0:06 11:98 \Sigma 0:05
NGC 1705 x19p0101t F220W,F1ND 497 10:79 \Sigma 0:12 11:92 \Sigma 0:05 11:98 \Sigma 0:05
NGC 3310 x19p0301t F220W 197 7:67 \Sigma 0:06 12:11 \Sigma 0:03 11:97 \Sigma 0:18
NGC 3690 x19p0401t F220W 897 9:55 \Sigma 0:12 12:97 \Sigma 0:03 13:69 \Sigma 0:18
NGC 3991 x19p0501t F220W 397 6:95 \Sigma 0:12 13:13 \Sigma 0:08 13:06 \Sigma 0:10
NGC 4670 x19p0601t F220W 297 9:03 \Sigma 0:09 12:57 \Sigma 0:05 12:72 \Sigma 0:06
NGC 5253 x19p0701m F220W,F1ND 497 9:84 \Sigma 0:09 11:76 \Sigma 0:04 11:69 \Sigma 0:08
Tol1924--416 x19p0801t F220W 447 6:92 \Sigma 0:11 13:25 \Sigma 0:07 13:48 \Sigma 0:08
NGC 7552 x19p0901t F220W 997 8:20 \Sigma 0:18 13:14 \Sigma 0:05 13:12 \Sigma 0:19
Therefore the data is available to make a direct comparison between the FOC and IUE fluxes.
Table 2 presents an observing log of the observations and the measurements extracted from
them, and from the IUE spectra.
The initial pipeline processing was adequate for most of the images. However, NGC 1705,
NGC 3690, and NGC 7552 contain bright sources with raw data values that exceed the 8­bit
word length of full format images and and thus ``wrap around''. The NGC 3690 and NGC 7552
images were fixed by adding appropriate multiples of 256 to the affected pixels of the raw
images and then re­running the pipeline calibration. This is impossible for NGC 1705 because
of saturation at the center of its brightest source.
The images were corrected for field non­linearity (item 7 above) using the algorithm of
Baxter (1994a,b). A 9 \Theta 9 median filtering box was employed to extract the smooth component
image, from which the correction factor image is created. This image is set to have a minimum
of 1.0 and an arbitrary maximum of 1.7. On average the linearity correction increases the total
flux of the objects by 7% (the range is 3% to 14%). On smaller scales the correction becomes
more important. For example at the center of NGC 7552 the mean correction factor averaged
over a 25 \Theta 25 box is 1.53. This correction is too large to be reliable, but fortunately only a
small are of the image is affected.
The background or ``sky'' level, C sky , was determined by taking the mean count rates in
several (8 -- 25) boxes with no apparent galaxy emission, and away from known defects at the
edges of the frame. The uncertainty in the sky level, ffl sky , is taken as the dispersion of the mean
levels of the different boxes. ffl sky is typically three or four times larger than one would expect
from photon statistics, probably due to inadequacies in the flat fielding. NGC 3310 covers too
much of its frame to use this method. Instead C sky is taken as the mean C sky from the 1Zw18,
NGC3690, NGC 3991, NGC 4670, Tol1924­416, and NGC 7552 frames.
The integrated countrate, C (in Hz), was extracted using a circular aperture with r = 6:94 00 .
7

This aperture is designed to have the same area as the IUE extraction aperture employed by
K93. The fluxes in the NGC 1705 images had to be corrected for the light lost near its saturated
central source. This was done by modeling the radial profile of this object using the wings of its
PSF. A model of it including the PSF wings was subtracted, and then the ``hole'' in the center
of the resulting image (from the oversubtracted core) was patched over. The total C was then
taken as the sum of the aperture flux from this image plus the flux of the source derived from
the modelling.
Before flux calibrating, PHOTFLAM was adjusted for the deterioration of detector sensi­
tivity and format dependent sensitivity variation (items 2 and 4 above). The mean epoch of
the observations is 1993.3, and therefore a ¸ 3:4% drop in sensitivity has occurred since the
pipeline calibration. Combined with the 25% increase in sensitivity of the 512z \Theta 1024 format
relative to the 512 \Theta 512 format there is a net increase of 22% in sensitivity relative to that
indicated by the pipeline calibration coefficients. Therefore I adopted a value for PHOTFLAM
that is 22% lower than that given in the image headers. Here I report m 220 magnitudes in the
STMAG system. For monochromatic magnitudes,
m – = \Gamma21:1 \Gamma 2:5 log(f – );
where f – is the flux density in erg cm \Gamma2 s \Gamma1 š A \Gamma1 . To derive m 220 from the integrated count rate
I take
m 220 = m 0 \Gamma 2:5 log(C \Gamma n pix \Lambda C sky ):
Here m 0 = \Gamma21:1 \Gamma 2:5 log(PHOTFLAM) is the zeropoint (m 0 = 20:844; 19:828 for the obser­
vations with and without the F1ND filter respectively), and n pix = 298885 is the number of
pixels in the extraction aperture. The errors reported for m FOC includes the photon statistics
uncertainty, and ffl sky integrated over the extraction aperture.
The IUE spectra used by K93 were obtained from their online database. The magnitudes
in the F220W passbands were extracted from the spectra using the CALCPHOT routine in the
SYNPHOT. The errors were derived from table 8 of K93. In table 2 I report C sky , m FOC , and
m IUE .
Fig. 1 compares the FOC and IUE m 220 magnitudes, while fig. 2 compares the ratio of F 220
fluxes F IUE =F FOC as a function of integrated count rate. The FOC and IUE fluxes agree very
well except for NGC 3690, which forms with IC 694 the merging system Arp 299. K93 note
``The IUE aperture aperture contains both objects'', however each is about 15 00 in diameter
with their centers separated by ¸ 23 00 (Wynn­williams et al., 1991). The IUE aperture could
not possibly fully contain the whole system. The disagreement in fluxes suggests that the IUE
aperture was not centered on NGC 3690. This view is supported by an FOS spectrum obtained
by Robert et al. (1995) in the center of NGC 3690. Their spectrum has a very different slope
from the IUE spectrum.
After rejecting the NGC 3690 observation, and taking the average of the two NGC 1705
observations, then the straight average of the eight magnitude differences is (m FOC \Gamma m IUE ) =
\Gamma0:05. The rms about the mean is 0.13 mag which is the same as the mean error in the
differences. Thus the scatter about the mean is about what one would expect from random
errors alone. Therefore, (m FOC \Gamma m IUE ) can be determined to an accuracy of the standard error
in the mean s:e: = rms= p
n \Gamma 1 = 0:05. In other words the FOC observations agree with the
8

IUE observation to within 5% on average, which is again what one expects from random errors
alone.
Fig. 1. FOC m 220 measurements compared to those from IUE. The dotted line marks where
they are equal.
Fig. 2. Ratio of IUE to FOC F 220 fluxes plotted as a function of integrated count rate.
4 Conclusions
Integrated magnitudes of eight starburst galaxies measured from full format FOC f/96 frames
through the F220W filter agree with IUE magnitudes to 0.05 mag on average. This is what one
expects from random errors alone. The accounting in x2 indicates that one expects systematic
errors of up to 0.15 mag. The expected systematic errors are external errors, since the fluxes
were measured with large enough of an aperture that small scale length variations in the
calibration are negligible. The better than expected performance probably indicates that the
DQE curve of Sparks (1991) around 2200 š A is determined better than one would expect from
rms of the stars measured at nearby wavelengths. It is an indication, that at least with F220W
filter data corrected for flatfield non­linearity, and care in the calibration, the external errors
can be reduced to 0.05 mag.
Comparison of stars within a frame with the same set­up are expected to be limited by
typical systematic internal errors of 0.15 mag. These can be reduced to ¸ 0:09 mag by carefully
removing the following effects: spatial sensitivity variations (item 4), high frequency ripples (5),
spatially varying non­linearity parameters (7), and format dependent lower sensitivity to point
sources (item 8; or use Greenfield's method of linearity correction). Ideally the combined
systematic (internal and external) errors can then be reduced to 10% for F220W observations.
Similar accuracies are likely for other UV observations.
Acknowledgements: This analysis benefited from many useful discussions with D. Baxter, P.
Greenfield, P. Hodge, A. Nota, B. Simon.
9

References
ffl Baxter, D.A. 1990, Instrument Science Report FOC­047 (STScI publication).
ffl Baxter, D.A. 1993, Instrument Science Report FOC­072 (STScI publication).
ffl Baxter, D.A., Greenfield, P.E., Hack, W., Nota, A., Jedrzejewski, R.I., and Paresce, F.
1993, in Space Astronomical Telescopes and Instruments II (S.P.I.E., 1945), eds. P.Y.
B'ely and J.B. Breckinridge, p252.
ffl Baxter, D.A. 1994a, Instrument Science Report FOC­073 (STScI publication).
ffl Baxter, D.A. 1994b, in Calibrating Hubble Space Telescope, eds J.C. Blades, and S.J.
Osmer, (STScI: Baltimore), p. 109.
ffl Greenfield, P. 1992, Instrument Science Report FOC­060 (STScI publication).
ffl Greenfield, P., Nota, A., Jedrzejewski, R., Hack, W., Hasan, H., Hodge, P., Baxter, D.,
Baggett, W., and Paresce, F. 1993, in Space Astronomical Telescopes and Instruments II
(S.P.I.E., 1945), eds. P.Y. B'ely and J.B. Breckinridge, p264.
ffl Greenfield, P. 1994a, Instrument Science Report FOC­074 (STScI publication).
ffl Greenfield, P. 1994b, Instrument Science Report FOC­075 (STScI publication).
ffl Jedrzejewski, R.I. 1992, Instrument Science Report FOC­062 (STScI publication).
ffl Jedrzejwski, R.I., Hartig, G.F., Nota, A., Greenfield, P.E., Baxter, D.A., Hack, W., and
Paresce, F. 1993, in Space Astronomical Telescopes and Instruments II (S.P.I.E., 1945),
eds. P.Y. B'ely and J.B. Breckinridge, p252.
ffl King, I.R., Bartels, C., Nowak, P., and Hack, W. 1994, in Calibrating Hubble Space
Telescope, eds J.C. Blades, and S.J. Osmer, (STScI: Baltimore), p. 128.
ffl Kinney, A.L., Bohlin, R.C., Calzetti, D., Panagia, N., and Wyse, R.F.G. 1993, ApJS, 86,
5 (K93).
ffl Krist, J. 1993, in Astronomical Data Analysis Software and Systems II , ASP Conference
Series 52, eds. R.J. Hanisch, and R.J.V. Brissenden, and J. Barnes, p530.
ffl Nota, A., Jedrzejewski, R., and Hack, W. (eds.) 1993, ``Hubble Space Telescope Faint
Object Camera Instrument Handbook (version 4.0)'', STScI, Baltimore.
ffl Robert, C., Leitherer, C., and Heckman, T.M. 1995, (in preparation).
ffl Sparks, W.B. 1991, Instrument Science Report FOC­053 (STScI publication).
ffl Wynn­Williams, C.G., Eales, S.A., Becklin, E.E., Hodapp, K.W., Joseph, R.D., McLean,
I.S., Simons, D.A., and Wright, G.S. 1991, ApJ, 377, 426.
10