The various steps define the data to be fitted, set up a spectral model, fit it, create a plot file showing the resulting fit, draw this to a graphics device and finally evaluate error intervals for the best fit parameters, if this level of detail is required. When fitting a single spectrum, the first step is not needed.SDATA -> SMODEL -> SFIT -> SPLOT -> DRAW -> SERROR
The spectral analysis package makes use of the parameter system so that most program parameters are defaulted or passed on from previous programs. Where it is required to override defaults, this may be done on the command line. Two global parameters, FIT_DATA and FIT_MODEL, are much used. The former contains the name of the current dataset to be fitted, or the name of a file containing references to a number of datasets (for fitting of multiple spectra) as set up by SDATA. FIT_MODEL contains the name of a dataset specifying the current fit model. Current values of these global parameters (amongst others) can be viewed by typing GLOBAL.
Spectral fitting can only be achieved when information about the instrument energy response is available to the system. The response information is stored in a standard data structure which must be attached to the spectrum itself. For example EXORESP performs this function for EXOSAT spectra. The fitting software will inform you if you have failed to attach a response to your spectrum.
For analysis of extragalactic sources it is possible to set a value of redshift to be applied to the source spectrum. See HELP on `Redshift' for details.
For raw spectra with 10 counts in a significant fraction of the spectral bins, the implicit assumption of Gaussian errors made in chi-squared fitting is no longer valid. In this case both the best fit and the confidence interval may be misleading. There are two possible solutions: the data can be rebinned, using SBIN, to give fewer bins with more counts, or you can abandon chi-squared in favour of likelihood fitting.