Strange treatment of MENLOPS K-Factor for unweighted events
In ttbar + 0j@NLO + 1j@LO I observe a 5% difference in the total cross section when using weighted or (partially) unweighted events (run card attached). I could trace this back to events having a negative MENLOPS-Kfactor. In case of unweighted events, there is a special treatment implemented which does not seem to take care of possible negative weights:
in SHERPA/PerturbativePhysics/Perturbative_Interface.C, line 203 onwards:
if (p_me->EventGenerationMode()!=0) {
if (m_weight>=ran->Get()) {
if (m_weight < 1.0) {
*p_localkfactorvarweights *= 1.0 / m_weight;
m_weight = 1.0;
}
} else {
return Return_Value::New_Event;
}
}
After changing m_weight
to dabs(m_weight)
in the second and third line, both cross sections agree.
Edited by Johannes Krause