Event generation efficiency and weights for unweighted events

[Frank Krauss]

Running a simple process (g b -> mu+ mu- b) I see a difference of about a factor of 5 in event generation time between weighted and unweighted. That's fine, but anecdotic evidence points that some of my unweighted events have huge weights > 1000, most of them seem to have weights above 1.

See the attached Run card.

[Stefan Hoeche]

Could you please try and generate a weight distribution for this setup? It might be that this problem is nothing but a manifestation of our changes in the default choice of Max_Epsilon, see ticket #133. Another option to check this is to explicitly set 'Max_Epsilon 0' in the process block.

[Frank Krauss]

Checked the weights again, they seem to be constant. I do not see any connection to the xsec (they say about 800 with an xsec of about 32 pb).

I degrade the thing to "trivial" and postpone to rel-perfect.

[Stefan Hoeche]

The total cross section is determined as (sum weights)/(n trials), and in the new methode for unweighted event generation the number of trials is different from one. This is why the constant weight is (in most cases) much larger than the total cross section.