Resolve issues with excess events

[Juergen Reuter]

Report by the Siegen ATLAS group in the process pp -> bblvlvA. Roughly 50% excess events.

[kilian]

Next in line.

I'll have to look for manageable processes where this occurs, and debugging is actually possible ...

[kilian]

OK, there is a massive amount of excess events in the simple process e+ e- => b bbar W+ W-. The integration is perfect, the f_max values nicely converge. Nevertheless, in event generation the weights almost always overshoot.

There must be a problem with rescaling between integration and event generation. Interesting ...

[kilian]

Found the culprit (but not yet the exact cause). Channel weights are not initialized properly when starting event generation. The real impact is unclear.

[kilian]

This is now fixed in r6934.

When a VAMP grid is read, the channel weights have to be recalculated or imported into the MCI record. This was not done in the initialization step of event generation. The problem was probably present in all 2.2 versions, since the MCI abstract data type was introduced.

The visible effect was most prominent in processes with a strongly dominant channel, such as tt production and decay. After the revision, excess events are back to a tolerable (tiny) amount.