ctools

When trying to make a joint fit to simulated LAT and CTA data, I found that the optimization process is rather sensitive to initial values of the parameters. In the only case tested so far, the source model is composed of the extragalactic isotropic diffuse emission (LAT only), the instrumental background (CTA only), and a point source at the position of the LMC (both instruments). When the intensity of the extragalactic diffuse is initially set at about 10 times its true value, the fit ends correctly but with incorrect estimates for the parameters of the point source and extragalactic diffuse. If the true value is used as initial guess, everything goes fine. This is illustrated in the two attached ctlike log files.
Things to try or correct:
1- What if we relax the boundaries on the parameters ? Maybe the way to the optimum when starting from too far needs to go through presently forbidden values...
2- Could be an idea to implement seom global rescaling of the parameters values before entering the iterative optimization process, in order to have Nobs ~ Nmodel.