Justine Devin / gammalib

The original powerlaw integration functions were prone to overflows for large gammas as the function computed the pow(energy, -gamma), where -gamma can be a large positive number.

…servation definition XML files

This avoids entering oscillations in case of bad convergence behaviour.

The XML module now handles predefined entities. Special characters, such as " will now be transformed into their predefined entities, e.g. ", when writing the XML file. Upon reading, predefined entities will be converted into special chacaters, so that the user always deals with a human readable string.

A new parameter was introduce to specify the acceptable function decrease in an iteration. Acceptable means that the new parameters are kept for continuing the iterations. This behaviour existed already before, but a low threshold of 1e-6 did not really activate it. Now the threshold was raised to 1.0 and this help ctlike to converge quicker. Methods accept_dec() have been implemented to get and set the acceptable threshold.

This improves convergence, but there is not real justification for that except that it will likely improve the precision of the computations.

Linear delta binning accelerates the interpolation scheme.

Not sure why this was set to 0.1.

…try while the response computation is done in spherical trigonometry.

…ped from optimization method.

Conflicts:
	ChangeLog
	inst/cta/src/GCTAResponse_helpers.cpp
	inst/cta/src/GCTASupport.hpp

… radians; this led to convergence problems with the shell model).

Only the model-based integrations allow a proper and easy handling of the elliptical models.

This appears more natural for an ellipse.

…strange Npred profile.

The binned analysis looks okay, but unbinned shows a strange behaviour for the semiminor axis parameter which results in a strong dependency of Npred on the semiminor axis value. This seems to prevent a nice convergence in unbinned mode.

There is also an open issue to be solved related to the possibility that the minor axis is larger than the major axis. The actual code assumes that minor axis < major axis. If this is not fulfilled, the omega arc computations will be wrong.

… There still seems to be a problem with the Irf computation as the log-likelihood profiles are not smooth. The Npred computation gives however a smooth profile.

Any elliptical model should be defined by semiminor and semimajor axes, these are in fact used in the CTA response computation. Hence they should live in the baseclass, not the derived class.

…This script is for validating the code in cta_irf_elliptical_kern_rho::eval().