The detailed description of spiskymax is given in the document written by its author A. Strong, available at ISDC [7]. Here we give only a brief overview of the program.
spiskymax regards the image as a pixelated skymap, and the aim is to obtain the intensity in each pixel. The most important idea is that the data constrain the image within some limits in an N-dimensional space, where N is the number of pixels. Hence there is no unique `best' image and we have to make some choice out of all the possible images within the constrained region. Maximum Entropy method as implemented here is one way of doing this, and of quantifying the uncertainty of the result. Since the number of pixels is usually large (e.g. 10 -10 ) the techniques involved are rather different from those of model-fitting.
spiskymax is adapted to the particular needs of SPI. The background temporal variations are treated via a template prepared by , and the coefficients are fitted during the imaging process. The input count spectra contain data for many energy bins, and spiskymax analyses a subset (or all) of these energies as specified by the parameters energy_range_min, energy_range_max. The number of iterations can be specified since the automatic stopping criterion is not always appropriate and in any case may not be reached in the CPU time available. Sources to be analyzed are defined by their positions in the input source catalog (SPI.-SRCL-CAT); only those with the SEL_FLAG flag =1 are analyzed, and their fluxes and 1 errors written to the output catalog.
After the image is produced, the flux and its error for a number of user-defined sources can be optionally determined. Each source is specified in terms of a position and 'ON' and 'OFF' radii. The ON region is the circle centered on the source position with radius ON and the OFF region is the annulus between the ON and OFF radii. The source flux is defined as
(intensity in ON region)/ (intensity in OFF region) *solid angle(ON)/solid angle(OFF).
The error is based on a Bayesian analysis which marginalizes over all the unwanted degrees of freedom, so the significance of a source by this method may not correspond to a 'classical' significance level. In general significances increase for smaller skymaps since the degrees of freedom decrease.