Overweighting: Use ME event multiple times

[Frank Siegert]

Instead of applying a weight>1 to an overweighted event, rather shower+hadronise the same ME event multiple times (and hit-or-miss the remaining fractional weight).

[Frank Krauss]

make it switchable to keep Stefan's DIS friends happy.

[Stefan Hoeche]

This feature is implemented on trunk r17150. An additional event generation mode, called "PartiallyUnweighted" was added, which keeps the overweight as an additional factor in the event weight. Note that additional events from overweight have number of trials zero. This is _NOT_ a mistake! It is needed to restore the total cross section.

[Frank Siegert]

The related commit in r17157 contains the following diff which breaks ME+PS (huge cut variations, even inclusive XS of Z+0,1,2 jets is suddenly >10% off, ... very clearly visible in benchmark):

Index: PHASIC++/Process/Single_Process.C
===================================================================
--- PHASIC++/Process/Single_Process.C   (revision 17156)
+++ PHASIC++/Process/Single_Process.C   (revision 17157)
@@ -257,17 +257,11 @@
       if (ampls.size()>2)
        THROW(fatal_error,"Invalid number of amplitudes");
       if (p_int->InSwaped()) {
-       Cluster_Amplitude *ampl(ampls.back());
-       if (ampls.size()>1) ampls.front()->Delete();
-       ampls.clear();
        msg_Debugging()<<"select 2nd.\n";
-       return ampl;
+       return ampls.back()->Copy();
       }
-      Cluster_Amplitude *ampl(ampls.front());
-      if (ampls.size()>1) ampls.back()->Delete();
-      ampls.clear();
       msg_Debugging()<<"select 1st.\n";
-      return ampl;
+      return ampls.front()->Copy();
     }
   }
   PDF::Cluster_Definitions_Base* cd=p_shower->GetClusterDefinitions();

The problem goes away when reverting this change, but assuming that the commit was made for a reason, I tried to fix it with a combination of old and new, but have failed so far (crashes during integration from CoreScale). Stefan, can you take a look?

[Stefan Hoeche]

This should be fixed on trunk r17188. Thanks a lot Frank, for finding the problem.

[Stefan Hoeche]

Milestone rel-1.3.0 deleted