CHAPTER 12 WAVELENGTH CALIBRATIONS
Note: This chapter is out of date and is scheduled for
1) TAPE TRANSFER
i) Transfer your raw data to disk as multiple maps of one
I2 file :
FITS use MTREAD
1DPCA use PCASAD
In both cases it is preferable to store your data as multiple
maps of one file, although there is a hard SAD limit of ~240
maps per file.
ii) PCASAD transfers data as 1 row maps. This is OK if you
prefer to handle the upper and lower rows seperately. Usually
it is easier to treat observations as a whole, so use option
SH (shuffle) to shuffle your data into 2 row maps in a new (I2)
NEW<1>=0. - create the new file
Format=I2 - as I2
SIZE=NI,2 - NI is rowlength - same as raw data
FILI=OLD - 1 row maps
FILO=NEW - 2 row maps
MAPI=1:N2:2 - 1 to N2 in steps of 2, N2 is no. of
observations x 2 = no of maps in OLD
FACM= c/r - default factors are unity
MAPO=1:N - N is no of observations
FACR=1 - (default)
ROWO=1 - put odd input maps into first row
Then : .SH(MAPI=2:N2:2,ROWO=2)
This completes the transfer.
iii) COPY your transferred raw data to a SAVE tape for easy
2) OBTAINING ROUGH INSTRUMENTAL SHIFTS BY CROSSCORRELATION
i) It is nice to be able to see graphically the instrumental
movement with time during the night. This is achieved by
correlating all the central arc rows against the first such row.
ii) Use SHuffle to shuffle out the central (or top, say for 1DPCA)
rows into a new file with no. of rows = no. of arc maps NARC.
FILI=RAW - raw data
SIZE=NI,NARC - new file of NI pixels by NARC rows
MAPI=1,3,5:21:2,... - the arc maps
ROWI=1 or 50 or... - whichever row you select
ROWO=2:,1 - put the first arc central row
into the last row - the
iii) Now do :
FT on file CENTRE - creating files REAL and COMPLEX
CR on REAL and COMPLEX
BT " " "
CN on REAL
- for rough shifts (which are in any case only an indication
of instrumental position, essentially the shift of the strongest
arc line, since the dispersion varies also) - it is not necessary
to MASK prior to FFT'ing.
iv) Now do :
CQ - produces numeric shifts (at terminal and on REDSHIFT.LOG)
- which can be stored in a file and plotted (WP).
3) CREATING AVERAGE ARC FILE
i) On the basis of the instrumental shifts obtained, and any
other phenomenological considerations which you wish to include,
create an average arc map for each observational map to be
scrunched. Usually this is just a straight average of the
before and after arcs, although weights can be included.
Use AR to do this :
AVERAGE=0.4*RAW+0.6*RAW for example
The DO loop can not be used with AR (ref: G QUINN), option AD
can be used, but all input maps must first be copied (CM or AR)
to the new AVERAGE file before AD (where FILI=FILO) can be used.
AR is neat but watch your typing.
4) WAVELENGTH CALABRATION SET UP
i) This is the most difficult part and requires substantial
user input. LA will not produce good fits regardless, but depends
on your judgement.
Read through the following documentation on LA and skim through
the detailed documentation in LA.DOC before starting.
COEF - LAMCOEF (OPTION LA)
Fits wave coefs to all rows of arc spectra.
A 1 D fit is used for maps which have each spectral
row from a different origin (eg - extracted aperature plate
spectra or many spectra from a single or double row instur-
ment such as the IDS). A 2-D fit (3rd order in Y, 2nd to
7th in X (lamda)) is best for true 2D spectra.
The user is required to carefully fit a selected row (usually
a sum of several rows from the centre of the map). COEF
then uses the coefficients so obtained as a first try fit to
each subsequent row, producing the best fit for each row by
checking the centres of arc lines found in each row against
a user selected arc line list. The order in which rows are
processed is from the STARTROW to the last row, and then from
STARTROW-1 backwards to the first. STARTROW is defaulted to the
map centre but can be changed to allow easy re-fitting of
A large scale plot of an arc row, with lines correctly identified,
is necessary for the first, careful fit. COEF tells you the
centres of the found lines, and you set up the first, linear
fit by typing in some channel, lambda pairs. The value of WIDE
is used as a test for valid correspondences between found line
centres (and their wavelengths calculated from the linear fit)
and the wavelengths given in the library of arc line wavelengths
(stored in SCANLAM.DAT).
You are then required to carefully prune out mismatches and
faulty lines - where faults may be due to the line being blended,
or saturated (arc lines counted at rates > 5 Hz on the IPCS are
susceptible to beam bending, and should not be used).
It is usual to fit coefficients to summed (ie. before obsn. and
after) arcs, since these most closely represent the instrumental
position midway during the observation. You can check the arc
drift by SHuffling out the central rows of all arc maps into
a single map, and correlating them against the first row.
(see 2 above)
ii) Files used
SCANLAM.DAT - a comprehensive list of arc line wavelengths
for arcs used at MSSSO and AAO.
COPY from [develop.spt] onto your area.
Select a list from here, and prune out blends
or saturated lines to obtain your
REDLIST.DAT - a reduced list of lines which you want to use
for a particular observing run. SAVE this when
asked in LA during setup, and thereafter say NO
otherwise you will get repeated versions, each
being the last used lines.
5) LA - AUTOMATIC
i) After setting up and obtaining coefficients in a file (COEF
say), LA is designed to allow fairly automatic fitting to all
other arc maps. Use :
a) REDLIST - say YES to reduced list ?
b) OPTION 3 - read old coeffs from file - this is OK
for a first try provided that the instrument
has not FLOPPED.
c) SAVE - say NO - use the initial reduced list for all
It is best to go through a second map, setting up the answers
for auto fitting in the .SVE file, and then try :
In most cases the output coef map (MAPCO) numbers are in 1:1
correspondence with the input arc map (MAPI) mapnumbers, but
this is not essential.
Leaving DEL and DLINES blank allows you to examine each fit
at the first proper attempt (press c/r after first fit which
is merely the fit to the input coefficients) and delete any
spurious matches which may arise.
1) Small sigmas can always be obtained by deleting all lines which
fit poorly, however good (accurate) fits really require 20-30
lines to be considered valid.
Never delete lines on the basis of a linear fit to
a few chosen lambda-channel pairs. Let the program
try a polynomial fit first.
2) Be wary of deleting bad matches at the edges, these lines tie
down the polynomial fits at the edges, and ensure good fits in
the central regions : if you delete them, the range of validity
will merely be narrowed.
3) Don't use high order fits if you have few lines.
4) Each row is initially fitted with the previous best fit, and
arc lines accepted or rejected according to the value of WIDE2
(in channels) : this needs to be large for 1DPCA data where the
shift between rows can be large.
5) In the automatic fit, lines are automatically rejected if
their wavelength difference (between calculated and true) is
> 2 * MAX( TOL, SIGMA), therefore allow a reasonable tolerance
so as not to invariably delete lines at the edge of the fit,
where differences can be high (~ 1 channel).
6) Reasonable sigmas are of order 0.2 - 0.5 channels, where
the high values are tolerable if large differences occur only
at the edges.
7) Check fits by scrunching your arc, and displaying both it
and the raw arc on the TV (or in the absence of a TV
by crosscorrelating one row with the rest - see redshifts for
a discription of the crosscorrelation routines.)
If a blow up occurs, this is usually due to the shift being
too great for the auto fit to work, and you will have to set up