File: xcsao.tex

package info (click to toggle)
iraf-rvsao 2.8.3-1
  • links: PTS, VCS
  • area: main
  • in suites: bullseye, buster, sid
  • size: 16,456 kB
  • sloc: ansic: 963; lisp: 651; fortran: 397; makefile: 27
file content (255 lines) | stat: -rw-r--r-- 12,214 bytes parent folder | download
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
% Lines starting with "%" are comments; they will be ignored by LaTeX.

\documentstyle[11pt,paspconf]{article}

\begin{document}

\title{XCSAO: A Radial Velocity Package for the IRAF Environment}
\author{Michael J. Kurtz, Douglas J. Mink, William F. Wyatt, Daniel G.
Fabricant, and Guillermo Torres}
\affil{Harvard-Smithsonian Center for Astrophysics, Cambridge, MA 02138}

\author{Gerard A. Kriss}
\affil{Department of Physics and Astronomy, Johns Hopkins University,
Baltimore, MD 21218}

\author{John L. Tonry}
\affil{Department of Physics, MIT, Cambridge, MA 02139}

\begin{abstract}
XCSAO is a software package for obtaining radial velocities of stars and
galaxies from optical spectra using the cross-correlation method.  It has been 
written to work within the IRAF environment, and is compatible with other
packages being developed at SAO for optical spectra.  We discuss the individual 
components of the package in terms of the decisions required to set up the
package for a new instrument/observing program.  Using artificial spectra we
show the procedures necessary to optimize the set up parameters, and
demonstrate the robustness of the package when applied to spectra from both
stars and galaxies.
\end{abstract}

\keywords{Data Analysis, Spectra, Radial Velocities, Redshifts}

\section{Introduction}

During the decade of the 1980s all radial velocities produced at SAO, both
stellar and extra-galactic, were reduced using a suite of FORTH routines on 
NOVA computers written by John Tonry (Tonry and Davis 1979) and maintained and
extended by Bill Wyatt (Tonry and Wyatt 1988).  Moving our reductions onto
modern computers in the IRAF (Tody 1986) environment required a total
rewritting of the code; this has resulted in the RVSAO package, the main parts
of which are XCSAO, for cross-correlation velocities, and EMSAO for emission
line velocities.  

Copies of the source and IRAF installation scripts
can be obtained by anonymous FTP from sao-ftp.harvard.edu as
pub/iraf/pkg/rvsao/rvsao.tar.Z.  A README file is also in the directory.
XCSAO has been extensively tested on Decstations and Sparcstations and is
expected to experience no major changes.  EMSAO is still undergoing substantial
revision.  Both tasks have been in production use at SAO for the past year, 
and are currently in production use by several large redshift and radial 
velocity programs worldwide.

XCSAO has its origins in the XCOR program written by Gerry Kriss, as well as
the FORTH code, and a FORTRAN rendition of it by Tonry.  Eventually we included
the continuum estimation methods from Mike Fitzpatrick's FXCOR.  Our code was
written and is being maintained by Doug Mink.

This paper presents a very brief overview of the capabilities of XCSAO, and
some of the testing procedures we have developed for it.  An extensive
description of the program is in preparation for PASP (Kurtz, et al. 1992).


\section{Elements of XCSAO}

The following subsections describe briefly the elements of XCSAO in the order
in which they occur in the code.  We indicate those elements which we believe
should receive critical attention when setting up XCSAO for a new instrument.
               
\subsection{Input Formats}

XCSAO takes lists of 1-dimensional wavelength calibrated spectra in the
standard format used by the ONEDSPEC task in the NOAO package.  Spectra may be
either log or linear lamda binned.  Two lists are required, one for the
unknowns and one for the templates; each unknown will be compared separately
with each template.  Figure~I shows the parameter file for XCSAO.

\begin{figure}
\vspace {20cm} 
\caption{Parameter File for XCSAO.}\label{fig-1}
\end{figure}

\subsection{Rebinning}

Spectra are rebinned into the specified number of log wavelength bins (NCOLS) so
that there is the maximum possible overlap between the unknown and the template 
in the rest frame.  This is achieved by iterating on the radial velocity a
total of NZPASS times; in addition the user can choose an initial guess CZGUESS
by setting CZINIT=yes.  The error in the estimate of radial velocity can often
be halved on the second pass.  

Note that this binning suggests that it is
worthwhile to have the template cover the broadest wavelength range possible,
at a dispersion equal or better than the unknown's, the rebinning will then
choose the proper wavelength region for maximal overlap, with the resolution
determined by the unknown.  We find that it is in all cases better to rebin to
the next larger power of two than the number of actual pixels, rather than the
next smaller.

\subsection{Apodization}

The spectra are apodized by a cosine bell over a specified percentage of its
extent.  The size of the apodized region should be kept as small as possible
while still preventing ringing in the fourier transform.  We have found 5\%\ is
satisfactory for most spectra, but this should be set for each new instrument
following some testing.

\subsection{Continuum and Emission Line Suppression}

The continuum is subtracted from each spectrum after being calculated by
routines derived from the ICFIT task, and only slightly modified from their
implementation in the FXCOR task in the RV0 package.  There are a large number
of possibilities here, different observing programs require different settings. 
This is the area which we believe deserves the most experimentation when
setting up a reduction.

If EMCHOP=yes XCSAO will remove emission lines and replace them with the value
of the continuum before performing the cross
corelation.  The emission lines are found using ICFIT, with parameters
settable as for the continuum supression; the method for finding emission lines
here is similar to that used in the sister task EMSAO.  Removing emission lines
is critical to finding absorption line redshifts from galaxies with emission
lines, if the spectrum is too noisy for the emission line removal, or if it has
sharp spurious absorption features from poor sky subtraction we recommend that
the offending regions be exised by hand, using SPLOT.

\subsection{Fourier Filtering}

Before correlating the unknown with the template each is put through a bandpass
filter in the fourier domane (unless TEMPFIL=yes, then the template is
considered to have already been filtered).  We have done a number of
experiments with the fourier filter, some of the main results are:

1.  The type of ramp used in the bandpass filter is not important (assuming it
is a reasonable choice, a step function  is not reasonable).  We have therefore 
removed all the options for filter type from XCSAO, and use just a cosine bell.

2.  The exact choice of the high pass roll-off is unimportant.  in so far as it
does not impinge on the main data region.

3.  The exact choice of the low pass roll-on is critical to the reduction, it
should be chosen with great care.  This also effects the relation of the $r$
statistic with the actual error, see below.

4.  Use of a frequency weighting scheme which can be shown in the limit of
infinitly wide spectra to produce the maximum likelihood estimation for the lag
(radial velocity) (e.g. Hassab and Boucher, 1979) is not an effective technique
for spectra similar to ours.

\subsection{Cross Correlation}

The normal fourier cross correlation, the transform of one spectrum times the
conjugate of the transform of the other, is performed.

\subsection{Peak Fitting}

The user can set the function to use to fit the peak with PKMODE, normally this
makes no significant difference in the velocity.  The user also sets the
fraction of the peak height to fit the function to with PKFRAC, .5 is the
default choice.

Because the spectrum is quantized in pixels while the fractional power point to
fit to is not we perform two fits, one two pixels wider than the other, and
report the weighted mean as the result of the fit.

\subsection{Error Calculation}

Following Tonry and Davis(1979) we calculate an error as a function of their
$r$
statistic.  It can be shown analytically that for sinusoidal noise with a half
width of the sinusoid equal to the halfwidth of the correlation peak that the
mean error in the estimation of the peak of the correlation function is 
\begin{equation}
error = {3\over 8}{w \over{(1+r)}}
\end{equation}
where w is the FWHM of the correlation peak, and r is the ratio of the
correlation peak height to the amplitude of the antisymmetric noise.

For large observing programs a better error estimate may be obtained.  The
coefficient ${3 \over 8} w$ is for many situations a constant, and may be
estimated by taking the mean of a number of measures.  The error estimate then
becomes a constant divided by $(1 + r)$.  This is not in all cases possible, for
example when the low pass fourier roll-on is changed the relation of $r$ with
error changes.  ${3 \over 8} w$ tracks this properly.  Figure~II shows the
calculated errors (dots) versus the actual mean errors (solid lines) as a
function of the $r$ statistic for three different low-pass filters; note that the
calculated errors do track the actual errors, but that the relation of $r$ with
error changes.
\begin{figure}
\vspace {8cm}
\caption{Individual error estimates (dots) versus actual mean errors (solid
lines) for 1200 synthetic echelle spectra.  The three sets are for three
different fourier filter roll-on points; the filter parameters, and the mean
errors for each filter are shown on the figure.}\label{fig-2}
\end{figure}

\subsection{Output Formats}

While XCSAO may be run totally as a batch proceedure, it may also be run in an
interactive mode.  Intermediate stages in the reduction may be plotted, and the
correlation peak to chose may be specified.  At the conclusion of a reduction
one may choose to redo the reduction interactively, to view a plot of the main
absorption lines, to edit the emission lines, to remove spectral regions, ....

\begin{figure}
\vspace{8cm}
\caption{Primary output plot from XCSAO}\label{fig-3}
\end{figure}
\begin{figure}
\vspace{8cm}                                    
\caption{Output plot showing absorption lines labeled}\label{fig-4}
\end{figure}

Figure~III shows the standard final plot, and Figure~IV shows an absorption line
plot, both for a galaxy spectrum observed by M. Ramella with the ESO 1.5m
telescope.

\section{Testing}

We have tested XCSAO with synthetic spectra created to have a specified number
of counts sampled from a parent distribution, which is a template spectrum. 
For tests of echelle spectra we used Kurucz models for the templates, for tests
of galaxy spectra we used high S/N observations.  Note that we can know the
true radial velocity difference between the unknowns and the templates either
exactly in the case of models, or where the template spectrum was exactly the
one used to create the synthetic spectra, or with very high accuaracy, as when
the radial velocity template is different from the template used to create the
synthetic spectra.
             
The data used to create Figure~II were 1200 synthetic spectra created from a
Kurucz model of a 5500K dwarf, and having between 5000 and 30,000 total counts. 
The template used for cross correlation was exactly the same template as was
used to create the synthetic spectra.  A complete description of the testing 
procedures will be found in Kurtz, et al. (1992).  

\acknowledgments

Steve Levine and John Morse made important contributions to the development of
the code.  Dave Latham, Alejandra Milone, Susan Tokaraz, and John Huchra have
contributed mightily of their expertise.  Much of the IRAF code we have used
was written by Mike Fitzpatrick and Frank Valdes.

\begin{references}
\reference Hassab, J.C. and Boucher, R.E. 1979, {\it IEEE Trans. Acoust. Speech
Signal Process.}, {\bf ASSP-27}, 922
\reference Kurtz, M.J., Mink, D.J., Wyatt, W.F., Fabricant, D.G., Torres, G.,
Kriss, G.A., and Tonry, J.L. 1992, \pasp, to be submitted
\reference Tody, D. 1986, in {\it Instrumentation in Astronomy VI}, ed. D.L.
Crawford, Proc. SPIE 627, 733
\reference Tonry, J.L. and Davis, M. 1979, \aj, {\bf 84}, 1511
\reference Tonry, J.L. and Wyatt, W.F. 1988, {CFA Z-Machine Data Analysis
Software}, Cambridge: Smithsonian Astrophysical Observatory
\end{references}

\end{document}