Using multi_run with NLO processes

Dear Junape,
with the aMC@NLO runs, you can directly ask for a large number of events in the run_card.
if you have many cores that you can use, you can split the event generation on the different CPUs.
In order to do so, there is this entry in the run_card
    -1 = nevt_job! Max number of events per job in event generation.

so suppose you want to generate 1M events, you can generate them in ~20 jobs of 50k each if you have 20 cores/cluster nodes that you can use.
just set nevt_job to 50000

another thing that you can do to speed up event generation is this:
suppose you have generated 1M event, then you want another 1M
Normally this would mean to recompute the grid and reintegrate the cross-section, which is quite a waste of time, if you have already done it with the same parameters and just want some more events.

So you can type
./bin/generate_events -o
where the -o means 'only generation'. This will use the grids from the LAST RUN you did. Note that no check is done for the consistency of the two runs, and it is up to you to check that you indeed are using the same parameters.
Let me know if this answer your question.

Cheers,

Marco

Hi Marco,

thanks a lot for you quick answer.

I am running in cluster mode so in principle I have no issue in dividing the jobs. As far as I know the problem of huge number of events was with pythia. Pythia will run also over each set of 50k?

I will try your suggestion and come back to you if I find any issue.

cheers,
Juanpe.

Hi Junape,
pythia will run at the end, on the final sample.
I have never had problems in running pythia over a large event sample, so it should be ok…
Cheers,

Marco

On 19 Sep 2014, at 12:56, Juanpe <email address hidden> wrote:

> Question #254718 on MadGraph5_aMC@NLO changed:
> https://answers.launchpad.net/mg5amcnlo/+question/254718
>
> Status: Answered => Open
>
> Juanpe is still having a problem:
> Hi Marco,
>
> thanks a lot for you quick answer.
>
> I am running in cluster mode so in principle I have no issue in dividing
> the jobs. As far as I know the problem of huge number of events was with
> pythia. Pythia will run also over each set of 50k?
>
> I will try your suggestion and come back to you if I find any issue.
>
> cheers,
> Juanpe.
>
> --
> You received this question notification because you are a direct
> subscriber of the question.

Hi Juanpe,

The problem with the number of the size of the stdhep (which was the reason of the limit on the number of events) is not present anymore.
I think however that it would be great to have a real multi-run option since this is way to linearize the problem and then to gain in speed for large generation.

Cheers,

Olivier

On Sep 19, 2014, at 1:03 PM, marco zaro <email address hidden> wrote:

> Question #254718 on MadGraph5_aMC@NLO changed:
> https://answers.launchpad.net/mg5amcnlo/+question/254718
>
> Status: Open => Answered
>
> marco zaro proposed the following answer:
> Hi Junape,
> pythia will run at the end, on the final sample.
> I have never had problems in running pythia over a large event sample, so it should be ok…
> Cheers,
>
> Marco
>
> On 19 Sep 2014, at 12:56, Juanpe <email address hidden>
> wrote:
>
>> Question #254718 on MadGraph5_aMC@NLO changed:
>> https://answers.launchpad.net/mg5amcnlo/+question/254718
>>
>> Status: Answered => Open
>>
>> Juanpe is still having a problem:
>> Hi Marco,
>>
>> thanks a lot for you quick answer.
>>
>> I am running in cluster mode so in principle I have no issue in dividing
>> the jobs. As far as I know the problem of huge number of events was with
>> pythia. Pythia will run also over each set of 50k?
>>
>> I will try your suggestion and come back to you if I find any issue.
>>
>> cheers,
>> Juanpe.
>>
>> --
>> You received this question notification because you are a direct
>> subscriber of the question.
>
> --
> You received this question notification because you are an answer
> contact for MadGraph5_aMC@NLO.