FITSPEC N1 N2 N3 N4 NG File NP PARAMS[NP] AXES[NP] FULL -- Fitspec estimates
best-fit parameter values of some observed spectra relative
to some model spectra tabulated on a cartesian grid. It assumes
that spectra between the grid points can be represented by
bi-linear interpolation.
Parameters:
N1, N2 -- Range of observed spectra
N3, N4 -- Range of model spectra
NG -- Number of spectra in each group. The spectra should be
thought of in groups of NG spectra. e.g. if you had
observations of Halpha and Hbeta with separate spectra
you would set NG = 2. You would require models for each
of course. All spectra of the same parameter should be
in one block.
FILE -- A file of best fit values can be dumped. <CR> to ignore;
! will store the values in the headers instead which can
then be used with 'GENSPEC' to compute the best-fit models.
NP -- The number of parameters . The model spectra should on a
Cartesian grid (i.e. equal steps in each parameter).
e.g you might have a grid of log g and log T with
the spectra stored to change first in log g and then log T
This would have NP=2.
PARAMS -- Names of the parameters, which should correspond to the
storage order, so that in the example above they would be
"log g" and "log T". These should be REAL variables
with values stored in the headers of the model spectra.
AXES -- The number of points on each parameter axis. e.g. if you
stored 10 gravities at 5 temperature points, you would
enter 10, 5. *** NB *** take care over both 'PARAMS' and
'AXES' to get reliable results.
FULL -- gives fuller output with chi**2 and best-fit values
reported for every cuboid and for each observed spectrum.
Checks are made for the correct match of pixels, but not for anything
else so the onus is on the user to get the axes etc correct. The method
used is to go through every cuboid of the model grid and carry out a
non-linear optimisation assuming bi-linear type interpolation from the
corners of the cuboids. The best few results are reported, the program
distinguishing between those in and outside their defining cuboids.
Ideally one should get just one fit within its cuboid (aka "non-extrapolated"),
all others being extrapolated. The best of the extrapolated values should be
within spitting distance of the non-extrapolated values if the model grid is
fine enough. More than 1 non-extrapolated value is listed in case of
multiple minima.
Uncertainties are calculated from covariance matrix. To mask points, mask the
data only. The mask on the model spectra is ignored.
This command belongs to the classes: fitting .