cross-section bias for gridpack in read-only

[Olivier Mattelaer]

Dear Olivier,

We are testing on the "readonly" gridpack mode to enable multithreading for event generation, as introduced in [1].
It seems the LHE cross-section given by this approach is abnormal for some multi-jet process. The discrepancy gets even larger as the jet number increases.

We prepare a W+3jet gridpack to reproduce this issue. The test is under v2.7.2 with a patch applied [2] to fix a "readonly" mode bug. The gridpack is uploaded to [3] as a reference.
For the "normal" mode, we use the command "./run 200 1 1".
For the "readonly" mode, we set the madevent folder to readonly, then do "../<folder>/run 200 1 1" in another empty directory.

The LHE cross-section given by two methods are: 1190.47 +/- 16.64 (normal) and 726.65 +/- 13.15 (readonly).
By reading the cross-section in "crossx.html", we get 1190.3 ± 2.76 fb . Hence the "readonly" result is abnormal.

Would you mind looking into this issue? Thank you very much for your help.

Best regards,
Congqiao

[1] https://answers.launchpad.net/mg5amcnlo/+question/662721
[2] https://bazaar.launchpad.net/~mg5core1/mg5amcnlo/2.7.3/revision/301
[3] https://coli.web.cern.ch/coli/tmp/.200515-212529_mtbug/w3jet_gridpack_272.tar.gz

[Olivier Mattelaer (olivier-mattelaer)]

This sounds an issue with the symmetry factor of a given process.
In normal mode P1_gg_ttxg/G5 returns 132pb
and in readonly P1_gg_ttxg/G5 returns 66.76pb
so this sounds a factor 2 issue

also the readonly mode seems to use non fix grid when running compare to the normal gridpack who does not change the grid anymore (which is likely better). Will have to change that as well.

More news later.

Olivier

[Olivier Mattelaer (olivier-mattelaer)]

Hi,

Thanks a lot for this report, it can be even worse than I thought originally.
(I thought that only the cross-section was affected but not the shape of the events, but I do think that both are actually impacted.

Thanks a lot for checking this features of MadGraph.
I believe this is an important feature that was left on the side for too long.
It is really nice that you spend the time to find such issues and that you force me to look back at this.

Cheers,

Olivier

[Congqiao Li (colizz)]

Hi Olivier,

Thank you very much for fixing this issue!

Cheers