MadGraph5_aMC@NLO

Increasing the accuracy at fixed order makes the jobs crash

Asked by matteo maltoni

Dear MadGraph experts,

I'm generating VBF Zjj at fixed order NLO QCD on the lemaitre3 cluster.

The process works properly with req_acc_FO = 0.005, but when I set it to lower values, in order to reduce the statistical error, at least one job fails to run, invalidating the generation.

Is there something I can do to have the jobs finish, or anything else that can help me reducing the statistical uncertainty?

Thank you,

Matteo

Question information

Language:
English Edit question
Status:
Solved
For:
MadGraph5_aMC@NLO Edit question
Assignee:
No assignee Edit question
Solved by:
Olivier Mattelaer
Solved:
Last query:
Last reply:
Revision history for this message
Best Olivier Mattelaer (olivier-mattelaer) said : 		#1

You can quickly hit the walltime of the cluster with such request of precision no?
Is that the issue?

Olivier

> On 31 Jul 2021, at 17:00, matteo maltoni <email address hidden> wrote:
>
> New question #698208 on MadGraph5_aMC@NLO:
> https://answers.launchpad.net/mg5amcnlo/+question/698208
>
> Dear MadGraph experts,
>
> I'm generating VBF Zjj at fixed order NLO QCD on the lemaitre3 cluster.
>
> The process works properly with req_acc_FO = 0.005, but when I set it to lower values, in order to reduce the statistical error, at least one job fails to run, invalidating the generation.
>
> Is there something I can do to have the jobs finish, or anything else that can help me reducing the statistical uncertainty?
>
> Thank you,
>
> Matteo
>
> --
> You received this question notification because you are an answer
> contact for MadGraph5_aMC@NLO.

Revision history for this message
matteo maltoni (matteo-maltoni) said : 		#2

Dear Olivier,

That was indeed the problem.
I increased the maximum time for the subjobs, and now it works.

Do have a criteria to guess this maximum times, for the subjobs and the whole process, without spoiling computational resources while trying?

Best,

Matteo

Revision history for this message
Olivier Mattelaer (olivier-mattelaer) said : 		#3

No we do not have that.

I think that I should be now able to guess that at LO (which will then allow for optimization) but I do not have time to invest on that to be honest.

NLO is even more complex and I do not know how to do that.

Olivier

> On 4 Aug 2021, at 10:55, matteo maltoni <email address hidden> wrote:
>
> Question #698208 on MadGraph5_aMC@NLO changed:
> https://answers.launchpad.net/mg5amcnlo/+question/698208
>
> Status: Answered => Open
>
> matteo maltoni is still having a problem:
> Dear Olivier,
>
> That was indeed the problem.
> I increased the maximum time for the subjobs, and now it works.
>
> Do have a criteria to guess this maximum times, for the subjobs and the
> whole process, without spoiling computational resources while trying?
>
> Best,
>
> Matteo
>
> --
> You received this question notification because you are an answer
> contact for MadGraph5_aMC@NLO.

Revision history for this message
matteo maltoni (matteo-maltoni) said : 		#4

Thanks Olivier Mattelaer, that solved my question.