Jobs lost in multicore process

Hi Matthew,

>
> Is there some way to prevent compute_widths from running on multiple cores, to avoid losing jobs when running on a batch farm?

you can edit the input/mg5_configuration.txt file
and set inside
run_mode to 0
and nb_core to 1

> Secondly, when running compute_widths on the default settings, I encounter cases where the reported error on the 2-body decays is large, but the 4-body decays are still not calculated. My understanding is that the 4-body decays should be calculated whenever the error is larger than 0.0025 according to the default settings, however in one case madgraph outputs "current estimated error: 15545.3799846 go to 4-body decay:" but only the 2-body decays are written to param_card.dat.

If the error is large, it is forcing the generation of the 4 body decay diagram.
Then a second estimation is done based on the actual Feynman diagram, this second estimation is actually much more precise than the second one.
So looks like that in your case the first estimation over-estimate the width by a huge factor (which we knows might happen since we use a very conservative estimator)
and that the computation is discarded by the second estimator due to an estimated negligeable value.

Cheers,

Olivier

On Jun 17, 2014, at 5:51 AM, Matthew Cahill-Rowley <email address hidden> wrote:

> New question #250333 on MadGraph5_aMC@NLO:
> https://answers.launchpad.net/mg5amcnlo/+question/250333
>
> Hello,
>
> I'm trying to calculate 4-body decays in the MSSM using MadWidth within MadGraph. I have run into a couple of issues. First, when submitting jobs to a batch farm, I received the following error (in 4 of 5 jobs):
>
> CRITICAL: Some jobs have been lost in the multicore treatment.
> The results might be incomplete. (Trying to continue anyway)
> Command "import /scratch/mrowley.531646/MG5_aMC_v2_1_1/mgdecay.cmd" interrupted in sub-command:
> "compute_widths 2000011 2000013 --body_decay=4 --path=./slha.dat" with error:
> IOError : [Fail 5 times]
> [Errno 2] No such file or directory: '/scratch/mrowley.531646/MG5_aMC_v2_1_1/tmpZXfs2d/temp_decay/SubProcesses/P0_erm_tamemn1tap/G1.8/results.dat'
>
> Note that /scratch/mrowley.531646/MG5_aMC_v2_1_1/ is the temporary batch system directory, created by uncompressing the MadGraph tarball in this directory.
>
> Is there some way to prevent compute_widths from running on multiple cores, to avoid losing jobs when running on a batch farm?
>
> Secondly, when running compute_widths on the default settings, I encounter cases where the reported error on the 2-body decays is large, but the 4-body decays are still not calculated. My understanding is that the 4-body decays should be calculated whenever the error is larger than 0.0025 according to the default settings, however in one case madgraph outputs "current estimated error: 15545.3799846 go to 4-body decay:" but only the 2-body decays are written to param_card.dat.
>
> Thank you for your help with these questions,
>
> Matthew Cahill-Rowley
>
>
> --
> You received this question notification because you are an answer
> contact for MadGraph5_aMC@NLO.