Implications of 'Event target could not be reached'

Asked by Marco H. on 2021-05-25

Hello,

I created my own model, which is basically SM + a scalar mediator, which can decay into electron neutrinos. With this model, I try to generate events of the form

med > e+ e- ve ve~

which can happen, if one of the final-state neutrinos radiates off a Z boson. I do this, because I need to know the energy distribution of the resulting electrons, i.e. d\Gamma/dE_e.

Now, my problem is as follows:

When I try to generate 10'000 events, this event target is always reached. However, when I try to generate 100'000 events, I oftentimes only get ~70'000 events and the message that the 'Event target could not be reached'. However, the result for the total width is always the same within the errors.

From other questions I know that this happens if there is a problem with the Monte-Carlo integration. For example, this also happens with the process z > e+ e- \gamma, which is divergent at LO. However, I am pretty sure that my process is not divergent, since the radiated Z boson is massive (The mediator has a mass of O(100GeV), i.e. Z cannot be considered massless). Also, I am really confused that I can generate 10*10'000 events, but not 1*100'000 events.

So my question is, can I still trust the resulting energy distribution, or is it likely that the Monte-Carlo integrator gets a wrong result due to the difficulties? Am I save if I just generate 10'000 events instead of 100'000 events? And why does this make a difference in the first place? And if I want to have N events, but only get M, do I have to correct the width by a factor M/N?

Thanks a lot for your help!

Best regards,
Marco

Question information

Language:
English Edit question
Status:
Solved
For:
MadGraph5_aMC@NLO Edit question
Assignee:
No assignee Edit question
Solved by:
Olivier Mattelaer
Solved:
Last query:
Last reply:
Revision history for this message
Olivier Mattelaer (olivier-mattelaer) said :
#1

Hi,

My suggestion here would be to generate 2 sample of 50k events.
this can be done via the multi_run command:
./bin/madevent
then type "multi_run 2" and ask for 50k events.

The two samples will be merged into a single one file of 100k at then end of the run.

> Am I save if I just generate 10'000 events

Depends what you means by save obviously, statistical error will be larger but otherwise it should be ok.

> And why does this make a difference in the first place?

This is something that is difficult to answer without running the process and have to full set of log available.
The most likely reason is that one channel of integration under-performs and that if you ask too much events out of it you hit some internal limitation in term of number of events per iteration and maximum number of iteration. In that case the above trick should work.

> And if I want to have N events, but only get M, do I have to correct the width by a factor M/N?

No the cross-section does not need any re-scaling and you can technically use those 70k events without any issue.
Some old analysis of CMS have shown that when it does happens the statistical error on the sample compare to nominal is a bit larger compare to a sample with the same number of events.

Cheers,

Olivier

> On 25 May 2021, at 14:50, Marco H. <email address hidden> wrote:
>
> New question #697239 on MadGraph5_aMC@NLO:
> https://answers.launchpad.net/mg5amcnlo/+question/697239
>
> Hello,
>
> I created my own model, which is basically SM + a scalar mediator, which can decay into electron neutrinos. With this model, I try to generate events of the form
>
> med > e+ e- ve ve~
>
> which can happen, if one of the final-state neutrinos radiates off a Z boson. I do this, because I need to know the energy distribution of the resulting electrons, i.e. d\Gamma/dE_e.
>
> Now, my problem is as follows:
>
> When I try to generate 10'000 events, this event target is always reached. However, when I try to generate 100'000 events, I oftentimes only get ~70'000 events and the message that the 'Event target could not be reached'. However, the result for the total width is always the same within the errors.
>
>> From other questions I know that this happens if there is a problem with the Monte-Carlo integration. For example, this also happens with the process z > e+ e- \gamma, which is divergent at LO. However, I am pretty sure that my process is not divergent, since the radiated Z boson is massive (The mediator has a mass of O(100GeV), i.e. Z cannot be considered massless). Also, I am really confused that I can generate 10*10'000 events, but not 1*100'000 events.
>
> So my question is, can I still trust the resulting energy distribution, or is it likely that the Monte-Carlo integrator gets a wrong result due to the difficulties? Am I save if I just generate 10'000 events instead of 100'000 events? And why does this make a difference in the first place? And if I want to have N events, but only get M, do I have to correct the width by a factor M/N?
>
> Thanks a lot for your help!
>
> Best regards,
> Marco
>
> --
> You received this question notification because you are an answer
> contact for MadGraph5_aMC@NLO.

Revision history for this message
Marco H. (hep-hufnagel) said :
#2

Great! Thanks a lot for the quick reply. This should answer my questions.

Just to be sure. If I take the generated events and use them to fill a normalized histogram with the energy of the final-state electron, then the resulting normalized histogram is a proxy for the quantity

1/\Gamma d\Gamma/dE_{e-}

where \Gamma is the rate for the process that I am looking at? Is this correct?

Cheers,
Marco

Revision history for this message
Best Olivier Mattelaer (olivier-mattelaer) said :
#3

yes

Olivier

> On 26 May 2021, at 17:20, Marco H. <email address hidden> wrote:
>
> Question #697239 on MadGraph5_aMC@NLO changed:
> https://answers.launchpad.net/mg5amcnlo/+question/697239
>
> Status: Answered => Open
>
> Marco H. is still having a problem:
> Great! Thanks a lot for the quick reply. This should answer my
> questions.
>
> Just to be sure. If I take the generated events and use them to fill a
> normalized histogram with the energy of the final-state electron, then
> the resulting normalized histogram is a proxy for the quantity
>
> 1/\Gamma d\Gamma/dE_{e-}
>
> where \Gamma is the rate for the process that I am looking at? Is this
> correct?
>
> Cheers,
> Marco
>
> --
> You received this question notification because you are an answer
> contact for MadGraph5_aMC@NLO.

Revision history for this message
Olivier Mattelaer (olivier-mattelaer) said :
#4

yes

Olivier

> On 26 May 2021, at 17:20, Marco H. <email address hidden> wrote:
>
> Question #697239 on MadGraph5_aMC@NLO changed:
> https://answers.launchpad.net/mg5amcnlo/+question/697239
>
> Status: Answered => Open
>
> Marco H. is still having a problem:
> Great! Thanks a lot for the quick reply. This should answer my
> questions.
>
> Just to be sure. If I take the generated events and use them to fill a
> normalized histogram with the energy of the final-state electron, then
> the resulting normalized histogram is a proxy for the quantity
>
> 1/\Gamma d\Gamma/dE_{e-}
>
> where \Gamma is the rate for the process that I am looking at? Is this
> correct?
>
> Cheers,
> Marco
>
> --
> You received this question notification because you are an answer
> contact for MadGraph5_aMC@NLO.

Revision history for this message
Marco H. (hep-hufnagel) said :
#5

Thanks Olivier Mattelaer, that solved my question.