variation in parton level processes with fixed scales and no pdf's
Affects | Status | Importance | Assigned to | Milestone | |
---|---|---|---|---|---|
MadGraph5_aMC@NLO |
Fix Released
|
Undecided
|
Unassigned |
Bug Description
Hi,
I have been generating events with a fixed scale:
True = fixed_ren_scale ! if .true. use fixed ren scale
True = fixed_fac_scale ! if .true. use fixed fac scale
and without pdfs:
0 = lpp1 ! beam 1 type
0 = lpp2 ! beam 2 type
using the commands:
import model mssm
generate go go > u u~ / ul~ ur~
and then comparing to an analytic formula. There seems to be quite some variation in this process if generated for multiple mass points. One does not obtain a smooth line, and there seems to be very bad convergence of the error (fluctuations of a few percent with 10k events per mass point). This problem does not appear fro the process:
generate go go > g g / ul~ ur~
where the cross section is spot on (per-mille level errors).
The same happens for the model I am working on myself.
Changed in mg5amcnlo: | |
status: | New → Fix Committed |
Changed in mg5amcnlo: | |
status: | Fix Committed → Fix Released |
Hi,
Something realy funny seems to appear here.
Some of the helicity are sometimes extremelly suppressed while contributing for 50% of the cross-sections.
Such behavior make the helicity selector to sometimes discard those helicities and this creates some fake determination of the cross-sections.
The solution is to be more conservative on that selection of helicity. LO/Source/ genps.inc
This can be done by modifying Template/
you can modify the lines: ******* ******* ******* ******* ******* ******* ******* ******* ******* **** (LIMHEL= 1e-6) (MAXTRIES= 10)
c Parameters for helicity sums in matrixN.f
c******
REAL*8 LIMHEL
PARAMETER
INTEGER MAXTRIES
PARAMETER
I have set LIMHEL to 1e-9 and then I do not observe such behavior anymore (in 65 different consecutive run).
increasing MAXTRIES should help as well.
I will do a combination of both for the next release.
Cheers and thanks for the information,
Olivier
PS: I still have 35 run schedule, I will keep you posted if any or those have the buggy behavior.