Long calculation time for subprocesses with 0 xs

Asked by Andreas on 2017-10-19

Dear experts,

I am trying to generate Z+DM events in Uli Haisch's 2HDM + pseudoscalar model [1].

Generally, I find that it takes very long times O(3 days) to run the subprocess surveys for a single parameter point (pertinent log file at [2], cards I used at [3]). Interestingly, all relevant non-zero contributions seem to be calculated after ~1h. MG5 then spends the remaining ~60h calculating that the remaining subprocesses give 0 contribution. While the exact duration varies, the effect is consistent between different MG5 versions and process definition choices.

Is there a way to speed this up? I imagine one could reduce the required precision for deciding that a subprocess does not contribute.

Any hints would be appreciated.

Thanks

Andreas

[1] Model files at https://gitlab.cern.ch/lhc-dmwg-material/model-repository/tree/master/models/Pseudoscalar_2HDM

[2] Log file at https://cernbox.cern.ch/index.php/s/cKUizHwX94v1ojE

Interesting part is starting at line ~840. There is a madspin crash in
the very end of the log file, but that is unrelated to what I describe here.

[3] Cards at https://cernbox.cern.ch/index.php/s/wSFrp9THn3aQFn0

The "customizecards.dat" file includes all the parameter changes with
respect to the model defaults.

Question information

Language:
English Edit question
Status:
Answered
For:
MadGraph5_aMC@NLO Edit question
Assignee:
No assignee Edit question
Last query:
2017-10-19
Last reply:
2017-10-20

Hi,

I assume that you are discussing about at LO computation.
In this case, the parameter controlling the zero results are hardcoded
in Source/dsample.f around line 1447, you will see the following:

      parameter (max_events=500000) !Maximum # events before get non_zero

 reducing such number will return the zero result in a faster way (but less secure way as well).

I would also advise, to check those zero result to ensure that they are indeed returning zero at the first iteration and not at a later stage of the run --which would indicate an issue--.

Cheers,

Olivier

________________________________________
From: <email address hidden> <email address hidden> on behalf of Andreas <email address hidden>
Sent: Thursday, October 19, 2017 4:10 PM
To: Olivier Mattelaer
Subject: [Question #659652]: Long calculation time for subprocesses with 0 xs

New question #659652 on MadGraph5_aMC@NLO:
https://answers.launchpad.net/mg5amcnlo/+question/659652

Dear experts,

I am trying to generate Z+DM events in Uli Haisch's 2HDM + pseudoscalar model [1].

Generally, I find that it takes very long times O(3 days) to run the subprocess surveys for a single parameter point (pertinent log file at [2], cards I used at [3]). Interestingly, all relevant non-zero contributions seem to be calculated after ~1h. MG5 then spends the remaining ~60h calculating that the remaining subprocesses give 0 contribution. While the exact duration varies, the effect is consistent between different MG5 versions and process definition choices.

Is there a way to speed this up? I imagine one could reduce the required precision for deciding that a subprocess does not contribute.

Any hints would be appreciated.

Thanks

Andreas

[1] Model files at https://gitlab.cern.ch/lhc-dmwg-material/model-repository/tree/master/models/Pseudoscalar_2HDM

[2] Log file at https://cernbox.cern.ch/index.php/s/cKUizHwX94v1ojE

Interesting part is starting at line ~840. There is a madspin crash in
the very end of the log file, but that is unrelated to what I describe here.

[3] Cards at https://cernbox.cern.ch/index.php/s/wSFrp9THn3aQFn0

The "customizecards.dat" file includes all the parameter changes with
respect to the model defaults.

--
You received this question notification because you are an answer
contact for MadGraph5_aMC@NLO.

Can you help with this problem?

Provide an answer of your own, or ask Andreas for more information if necessary.

To post a message you must log in.