weighting events properly in multiple-sample

Asked by Arian Abrahantes on 2012-01-15

Dear MG-Team:

Consider the simple problem,

I want to model a pt distribution of a given final particle (i.e it behaves, as many other observables, 1/pT) and I want the larger pt bins to have a statiscal error as small as lower pt bins have (so, I want the full distribution to have a similar statistical error in the full pt range). I had done this kind of study using pythia, in that case I cut on a few space paramente (let say transfer P_T in the parton interaction, pT_hard) and generate a few samples of events (10<PT_hard<50, pT_hard>50, PT_hard>100) not having the same number of events each. The issue is that in Pythia I had three final global parameters of the simulation: x-section of each sample (sigma), the tried parton configurations (ntrials) and the successes (the amplirtude of the sample I wanted, the nevents in the MG run_card). Then observable distribution for each sample needed to be weigthed before summation. In pythia's case the weight factor

sigma/ntrials

in such a way the histograms look smooth and independent of the successes asked por each sample. And this the point, in the results I get in MG simulation from unweightedevents root files the ntrials is never stored, just the success (even the weight in the tree is the average value of the x-section of succeses). So when I tried to add up al my samples in the final analysis I get the typical bumps in the distributions at the cuts I set if I cut on a given final P_T, which tells me I am making the weighting process wrong.

This is a quetsion actually for the ExRootAnalysis part but it spread from parton simulation to expermiental observables. The issue is the following:

1- How to properly weight observable distributions when adding diffrent samples of a given reaction?

I know the question looks simple but I can not figure out how to get it work properly with MG or at least find the suited explanation for weighting process in MG. Probably I am missing something .

any help is welcome and thanks in advance ,

arian

Question information

Language:
English Edit question
Status:
Solved
For:
MadGraph5_aMC@NLO Edit question
Assignee:
No assignee Edit question
Solved by:
Arian Abrahantes
Solved:
2012-01-16
Last query:
2012-01-16
Last reply:
2012-01-15

Hi,

so let_fix convention first,
10<PT_hard<50, this corresponds to a cross section noted xs_10 and
correspond to sample 1 with N_1 un-weighted event
pT_hard>50, this corresponds to a cross section noted xs_50 and
correspond to sample 2 with N_2 un-weighted event
PT_hard>100)this corresponds to a cross section noted xs_100 and
correspond to sample 3 with N_3 un-weighted event

In general the weight should be
XS / N

But here you have an overlap between the sample 2 and 3 that you need
to take into account
so the weight for the samples are
sample 1 : xs_10 / N1
sample 2 (if pt<100): xs_50 / N2 (nothing weird here, except that the
weight is valid only for pt<100)
sample 2 (if pt>100): xs_100 / (N3 + xs_100/xs_50 * N2)
sample 3 : xs_100 / (N3 + xs_100/xs_50 * N2)

(N3 + xs_100/xs_50 * N2) corresponds to the number of events generated
above pt>100.

Cheers,

Olivier

On 15-janv.-12, at 14:15, Arian Abrahantes wrote:

> New question #184781 on MadGraph5:
> https://answers.launchpad.net/madgraph5/+question/184781
>
> Dear MG-Team:
>
> Consider the simple problem,
>
> I want to model a pt distribution of a given final particle (i.e it
> behaves, as many other observables, 1/pT) and I want the larger pt
> bins to have a statiscal error as small as lower pt bins have (so, I
> want the full distribution to have a similar statistical error in
> the full pt range). I had done this kind of study using pythia, in
> that case I cut on a few space paramente (let say transfer P_T in
> the parton interaction, pT_hard) and generate a few samples of
> events (10<PT_hard<50, pT_hard>50, PT_hard>100) not having the same
> number of events each. The issue is that in Pythia I had three final
> global parameters of the simulation: x-section of each sample
> (sigma), the tried parton configurations (ntrials) and the successes
> (the amplirtude of the sample I wanted, the nevents in the MG
> run_card). Then observable distribution for each sample needed to be
> weigthed before summation. In pythia's case the weight factor
>
> sigma/ntrials
>
> in such a way the histograms look smooth and independent of the
> successes asked por each sample. And this the point, in the results
> I get in MG simulation from unweightedevents root files the ntrials
> is never stored, just the success (even the weight in the tree is
> the average value of the x-section of succeses). So when I tried to
> add up al my samples in the final analysis I get the typical bumps
> in the distributions at the cuts I set if I cut on a given final
> P_T, which tells me I am making the weighting process wrong.
>
> This is a quetsion actually for the ExRootAnalysis part but it
> spread from parton simulation to expermiental observables. The issue
> is the following:
>
> 1- How to properly weight observable distributions when adding
> diffrent samples of a given reaction?
>
>
> I know the question looks simple but I can not figure out how to get
> it work properly with MG or at least find the suited explanation for
> weighting process in MG. Probably I am missing something .
>
> any help is welcome and thanks in advance ,
>
> arian
>
> --
> You received this question notification because you are a member of
> MadTeam, which is an answer contact for MadGraph5.

Dear Oliver thanks I'll try this. At the moment I am closing the thread.