[ Basic Info | User Guide ]
Basic Information on gaufit
Purpose: Fit gaussians to profile
GAUFIT fits gaussians to a profile and can write the output to
a Miriad dataset, a logfile or the terminal.
The fitting is done using an adapted version of fitting routines
in numerical recipes.
Obligatory parameters are:
a) either or both of 'in=' and 'parinp='
c) and either 'estim=' or 'options=findestim'
The ease of fitting depends strongly on the initial estimates.
These can be given using the estim= keyword, or they can be
automatically found (options=findestim). The latter is usually
preferable, though the former may be necessary in pathological
If multiple gaussians are fit but the fit is bad, another try is
made with one less gaussian. This is repeated until the fit
works. A fit is considered bad if the rms of the residual is
higher than 1.8 times the rms estimate, or if the parameters lie
outside the range given by cutoff=, crange= and wrange=, or if
the fitting routine does not converge. If after all possible
ways of retrying a fit the rms is between 1.8 and 5.4 times the
rms estimate, accept the fit after all; maybe a low-level
component increased the rms, or the profile is not perfectly
gaussian, but the fit is somewhat reasonable.
Automatic initial estimates are made by first finding the
velocity and amplitude of the peak. An estimated width is found
by looking to both sides of the peak for the nearest zero of the
2nd derivative and at where the half-maximum lies. The most
consistent combination gives the width estimate. For low S/N
profiles the width is found from the integral out to the nearest
zero. This estimated component is then subtracted from the
profile and the process is repeated until the maximum amplitude
is too low, or until the maximum number of gaussians has been
If parinp= is used gaussian parameters are taken from this
dataset for all pixels outside the specified region. Inside the
region new fits are made. All results are written to the
params= dataset. The number of fitted gaussians does not have
to be the same between the parinp= and params= datasets. This
creation of an extra dataset is needed because Miriad very-deep-
down disallows opening an existing dataset for writing. So it
is not possible to add new fits to an existing parameter set.
If parinp= is present, but in= is not, then no new fits are
made, but the gaussians in parinp are sorted as specified by the
cmpsort= keyword. They can also be transformed as specified by
Standard keyword in. See the help on in for more information.
Input dataset with spectra to be fit.
Standard keyword region. See the help on region for more information.
Note: the region=mask option is not implemented. The mask of
the input dataset is used however. If there is one, the profile
value at masked datapoints is set to zero before doing the fit.
Give a value for the rms of the profile or a dataset from which
it can be read. No default. Used by the fitting procedure to
determine when convergence occurs. If a single real value is
given, it is used for each profile. If the name of a dataset is
given, the rms at each pixel is read from that dataset; this is
particularly useful when fitting a dataset created with linmos,
for which the rms will vary across the field. Create the rms
dataset using 'linmos in=list options=sens out=rms' followed by
'maths exp=rms*<rmsvalue>'. The value should be the rms of the
profile as found with imstat on signal-free regions.
Initial estimates. Give an estimate for the amplitude, velocity
and fwhm for each component (if options=integral, pixels or
dispersion is used, give integral instead of amplitude etc.).
This is quite critical and should already be reasonably close.
The same initial estimate will be used for all profiles. If
options=findestim is used, the initial estimates are determined
by gaufit and estim= is ignored
Maximum number of gaussian components to fit (maximum 10,
Optional input parameter dataset. All fits outside the
specified region are read from this dataset. If in= is not
present, the fits in the selected region are just sorted and
selected as specified by cmpsort, cutoff, vrange and wrange.
The fits outside the region are untouched.
Optional output dataset to which the fit parameters can be
written. For each fitted component six planes are written, one
with the amplitude (or integral), one with the position and one
with the fwhm (or dispersion), and three more with the errors.
The planes with errors come after the planes with all fit
results. A final plane which contains the rms of the residual
Axis along which profiles are taken. Provide the axis name as
listed by PRTHD. Alternatively, 'longitude' (or 'lon'),
'latitude' (or 'lat'), or 'spectral'; or 'x', 'y', 'z', or 'a'.
'ra', 'dec', 'freq', and 'vel' are also understood. Default is
the spectral axis.
First smooth the profile over 'smooth' pixels before fitting.
Controls the output. Defaults are the opposite of the action
specified by an option. Possible options are:
nofit: output the initial estimates, don't make fits
findestim: let gaufit determine the initial estimates
noprint: do not print the fit results on the terminal
supbad: suppress results for fits outside ranges given
by cutoff, crange and wrange and results for bad
estimout: print initial estimates
intermout: print some intermediate results for multi-
abspix: x, y coordinates on output are relative to lower
left, rather than relative to crpix
abscoo: x, y coordinates on output are absolute
wrprof: write out a file with the data and the fit so that
it at least is possible to use plotting programs
to compare them; a kludge until gaufit itself can
plot. filenames will be 'profile_at_$x_$y' (or
given by prof=)
integral: write out integral of gaussian instead of
dispersion: write out dispersion of gaussian instead of fwhm
pixel: write center and width in pixels, not in axis
units (for these three: also interpret input for
cutoff, vrange, wrange and estim keywords as
average: first make an average profile of the selected
region and then fit one single gaussian to this
summed: first make a summed profile of the selected region
and then fit one single gaussian to this profile
negative amplitudes may be both positive and negative,
instead if just positive
fixvelo: fix the velocities to the initial estimate during
fixwidth: fix the width to the initial estimate while
fitting (fixvelo and fixwidth can be combined)
This parameter specifies how to sort the resulting components
The following options exist
velocity, amplitude, integral, width, vdiff, vrange
Option 'velocity' and 'fwhm' result in components sorted on
increasing velocity or width.
Option 'amplitude' and 'integral' result in components sorted on
decreasing amplitude or integral.
If vdiff is used, then a second parameter gives a center
velocity; components are sorted based on the difference between
the fitted velocity and this center velocity.
If vrange is used, the second and third parameter give a
velocity range. If one component is within this range, it
becomes the first. If none or more than one is within this
range, they are sorted on velocity.
Usually, cmpsort is applied for every pixel of the dataset.
This is wanted when originally fitting (in= used). It is also
generally wanted when refitting part of the dataset (in= and
parinp= used), especially when more gaussians are to be added in
selected regions. However, when only parinp= is present, the
sorting is done only in the selected region and everything
outside is left alone.
Optional output dataset, to which theoretical (described by fit)
profiles can be written.
Optional output dataset, to which the difference between the
profile and the fit can be written
Optional filename for use with options=wrprof
Give a cutoff for the amplitude/integral. Can be 1, 2 or 3
values, all in units of the rms.
If one value, fits with amplitude (integral if options=integral
set) below the given cutoff are not written out. The absolute
value of the amplitude is used if options=negative was set.
If two values, fits with amplitude/integral in the specified
range are eliminated.
If three values, further eliminate fits for which the ratio of
amplitude to amplitude error is less than specified ratio
(default 1). Default: cut off amplitudes below 3 times the rms
and with amp/err<2.
When using options=average or summed, only average/sum pixels
whose intensity is above cutval (default: sum all)
Give a range (in units along profile) between which the center
should lie. Fits that result in centers outside this range are
not written out. A third value specifies to not write out fits
whose center is uncertain by more than that number of channels.
Default: cut off centers outside profile range and uncertain by
more than four channels.
Give 1, 2 or 3 values: a lower limit, and/or a range, and/or a
S/N ratio for the width (fwhm or dispersion). Fits giving
widths below the lower limit, outside the range, or with too
uncertain widths are not written out.
Default: cut off fwhms less than 1 pixel and larger than the
length of the profile, and with value/error less than 1.
If the name of a file is given, the results of the fitting are
written to this file instead of to the terminal
Revision: 1.13, 2013/01/29 03:09:28 UTC
Generated by email@example.com on 21 Jun 2016