A large discrepancy between an original and reweighted cross-section

Well re-weighting can have very large statistical error.
Did you take that into account. I would not be surprised if you have HUGE statistical error when targeting interference alone (like a statistical error of the order of 100%)

Additionally, re-weighting works only if you have the same support.
This concern the phase-space support but also the helicity support (since you are using the helicity per helicity re-weighting.

Cheers,

Olivier

> On 6 Feb 2021, at 17:45, Hesham El Faham <email address hidden> wrote:
>
> New question #695405 on MadGraph5_aMC@NLO:
> https://answers.launchpad.net/mg5amcnlo/+question/695405
>
> Hi,
> I am trying to do a simple validation exercise for EFT re-weighting using mg5 v3.0.3-betternlops. I first generate an EFT process using the simple command:
> -- >
> import model dim6top_LO_UFO_each_coupling_order
> generate g g > t t~ h FCNC=0 DIM6^2==1 DIM6_ctG^2==1
> <--
> The cross-section of this process is around 0.265 pb.
> Then, I generate the same process in the SM using: generate g g > t t~ h and then set the re-weight card to:
> -->
> change mode LO
> change model /home/elfaham/Downloads/tzw/working_dir/3.0.3-betternlops/models/dim6top_LO_UFO_each_coupling_order
> change process g g > t t~ h FCNC=0 DIM6^2==1 DIM6_ctG^2==1
> change output 2.0
> launch
> /home/elfaham/Downloads/tzw/working_dir/ttH/test_ttH_EFT/Cards/param_card.dat
> <--
> I am expecting that the latter re-weighted process would generate the same cross-section as the former process. However, the cross-section of this re-weighted process is around 0.12pb (compared to 0.265pb for the former one).
> I am wondering if I am doing something wrong here, your help is appreciated.
> Thanks,
> Hesham
>
> You received this question notification because you are an answer
> contact for MadGraph5_aMC@NLO.

Thank you for the prompt reply.

It is true that the stat error is almost +- 100%. Is there a way to avoid this large stat error? Perhaps to generate way more events than I do now?

May you please elaborate a bit on the "same support" thing? I am not sure I follow.

Best,
Hesham

> It is true that the stat error is almost +- 100%. Is there a way to
> avoid this large stat error? Perhaps to generate way more events than I
> do now?

This is one way. Starting from a distribution of events closer from your final distribution is typically a smarter idea.

> May you please elaborate a bit on the "same support" thing? I am not
> sure I follow.

The best is to read the LO section of the paper describing the re-weighting feature:
1607.00763 <https://arxiv.org/abs/1607.00763>

Cheers,

Olivier

> On 6 Feb 2021, at 19:30, Hesham El Faham <email address hidden> wrote:
>
> Question #695405 on MadGraph5_aMC@NLO changed:
> https://answers.launchpad.net/mg5amcnlo/+question/695405
>
> Status: Answered => Open
>
> Hesham El Faham is still having a problem:
> Thank you for the prompt reply.
>
> It is true that the stat error is almost +- 100%. Is there a way to
> avoid this large stat error? Perhaps to generate way more events than I
> do now?
>
> May you please elaborate a bit on the "same support" thing? I am not
> sure I follow.
>
> Best,
> Hesham
>
> --
> You received this question notification because you are an answer
> contact for MadGraph5_aMC@NLO.

Thank you Olivier,

I am sorry if I am taking time to digest this. I don’t understand what do you mean when you say “starting from a distribution of events closer to the final one”?

Best,
Hesham

Read the paper should answer your question.
If you look at the formula of the statistical error after re-weighting
it is clear that you need to have to have the re-weight factor dsitribution as flat as possible.
Which means that your original sample and your targetted sample should be very similar.

Cheers,

Olivier

> On 6 Feb 2021, at 22:20, Hesham El Faham <email address hidden> wrote:
>
> Question #695405 on MadGraph5_aMC@NLO changed:
> https://answers.launchpad.net/mg5amcnlo/+question/695405
>
> Status: Answered => Open
>
> Hesham El Faham is still having a problem:
> Thank you Olivier,
>
> I am sorry if I am taking time to digest this. I don’t understand what
> do you mean when you say “starting from a distribution of events closer
> to the final one”?
>
> Best,
> Hesham
>
> --
> You received this question notification because you are an answer
> contact for MadGraph5_aMC@NLO.

Thank you, I will read the paper then.

In the re-weighting process, I get this warning message: WARNING: dynamical_scale is set to -1. New sample will be with HT/2 dynamical scale for renormalisation scale, is it something I should worry about?

Best,
Hesham

This message occurs when you change the mass of some final state particles.
When you do, we can not reproduce the dynamical scale choice set to -1 and therefore switch to HT/2,
nothing wrong but you need to be aware of that.

Now we recently change the default param_card used for the re-weighting so it is possible that you have that message because of that meaning that you are not using the re-weighitng card that you think you are using.
Now since you use some outdated alpha version of the code, it is likely that you do not yet have such change to the code.
(I actually do not know what was merged/not merged within that development branch)

Cheers,

Olivier

Thank you Olivier.