Add an additional transition point to the stacked elliptical integration.

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

Add an additional transition point to the stacked elliptical integration.
This improves convergence, but there is not real justification for that except that it will likely improve the precision of the computations.
Jürgen Knödlseder
1 parent f8c9f87f
Showing 1 changed file with 10 additions and 0 deletions
inst/cta/src/GCTAResponseCube.cpp
@@ -1245,6 +1245,16 @@ double GCTAResponseCube::psf_elliptical(const GModelSpatialElliptical* model,
         bounds.push_back(rho_min);
         bounds.push_back(rho_max);
  
+        // Kluge: add this transition point as this allows to fit the test
+        // case without any stalls. Not clear why this is the case, maybe
+        // simply because the rho integral gets cut down into one more
+        // sub-interval which may increase precision and smoothed the
+        // likelihood contour
+        double transition_point = delta_max - rho_obs;
+        if (transition_point > rho_min && transition_point < rho_max) {
+            bounds.push_back(transition_point);
+        }
+
         // If the integration range includes the semiminor boundary, then
         // add an integration boundary at that location
         if (semiminor > rho_min && semiminor < rho_max) {