statistical equivalence of gridpack and multi_run runs

hi,

It will be equivalent since you stilll have a non zero probability to have event from those subheading channel (both in multi-run and in gridpack)

Cheers,

Olivier

On Nov 21, 2015, at 01:07, Wojciech Kotlarski <email address hidden> wrote:

New question #275030 on MadGraph5_aMC@NLO:
https://answers.launchpad.net/mg5amcnlo/+question/275030

Dear MG experts,

We're trying to generate large amount of events (let's say 10M) combined from 10k samples. Is the set of samples generated by running 1k gridpacks statistically equivalent to running 1k multi_runs? We are worried that rare process, which doesn't matter for 10k events run, will not be recognized correctly during survey in multi_run mode, and hence not represented correctly in combined 10M sample. As far as I understand grid pack should overcome this problem. Or maybe it doesn't matter and both samples would be equivalent at the end. Can you comment on that?

cheers,
Wojciech

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

Hi,

Thanks a lot for the answer. So a follow up question. If one intends to generate a lot of samples in gridpack runs, wouldn't it be beneficial to increase accuracy of survey while creating a gridpack. This by default is 1% or up to 8 iterations (whichever comes first). Naively it seems that it would allow gridpack runs to make better selection of channels.

cheers,
Wojciech

They are two important pieces that you want to have
1) a good estimate of the cross-section. Actually only the order of magnitude really matter (asking 1% is conservative here)
2) the starting grid associate to the integral
on that one at some point you have a plateau and the gain in precision is only due to the number of PS point used.

So I do not that you gain (at least not in a simple way).

Cheers.

Olivier

> On Nov 23, 2015, at 20:37, Wojciech Kotlarski <email address hidden> wrote:
>
> Question #275030 on MadGraph5_aMC@NLO changed:
> https://answers.launchpad.net/mg5amcnlo/+question/275030
>
> Status: Answered => Open
>
> Wojciech Kotlarski is still having a problem:
> Hi,
>
> Thanks a lot for the answer. So a follow up question. If one intends to
> generate a lot of samples in gridpack runs, wouldn't it be beneficial to
> increase accuracy of survey while creating a gridpack. This by default
> is 1% or up to 8 iterations (whichever comes first). Naively it seems
> that it would allow gridpack runs to make better selection of channels.
>
> cheers,
> Wojciech
>
> --
> You received this question notification because you are an answer
> contact for MadGraph5_aMC@NLO.