Mg5 fails to generate all events, when aQGC is enabled

Asked by Evgeny Soldatov on 2020-02-06

Dear experts,

I'm generating the following process with MadGraph:
generate p p > vl vl~ a j j QCD=0 QED=5
Anomalous couplings are enabled using EFT model SM_LT8_LT9_UFO.

I set 1 paratemeter (FT9) equal to 3.000000e-12 and all others equal to 0.

And when I'm trying to generate events I get much less of them then I requested.

The error is the following:
INFO: fail to reach target 10000
  === Results Summary for run: run_01 tag: tag_1 ===

     Cross-section : 0.07003 +- 0.0001146 pb
     Nb of events : 1744

Param_card, run_card and the log are available here: https://cernbox.cern.ch/index.php/s/28A10yOxIAs420b

I know that this problem was mentioned before in one of the tickets: https://answers.launchpad.net/mg5amcnlo/+question/683400
but there were no clear conclusion on this.

With the best regards,
Evgeny

Question information

Language:
English Edit question
Status:
Answered
For:
MadGraph5_aMC@NLO Edit question
Assignee:
No assignee Edit question
Last query:
2020-02-17
Last reply:
2020-02-18

The minimum that you should do. is to use this method:
https://answers.launchpad.net/mg5amcnlo/+faq/2312 <https://answers.launchpad.net/mg5amcnlo/+faq/2312>

After that this model breaks is known to break perturbative theory and unitarity which is then difficult for our integrator to catch up with the bad associated behaviour.
Since in top of that you know that such model have HUGE fine tuning issue.
Therefore it is not too surprising that our phase-space integrator has some issue with this kind of model.

Cheers,

Olivier

> On 7 Feb 2020, at 01:03, Evgeny Soldatov <email address hidden> wrote:
>
> New question #688567 on MadGraph5_aMC@NLO:
> https://answers.launchpad.net/mg5amcnlo/+question/688567
>
> Dear experts,
>
> I'm generating the following process with MadGraph:
> generate p p > vl vl~ a j j QCD=0 QED=5
> Anomalous couplings are enabled using EFT model SM_LT8_LT9_UFO.
>
> I set 1 paratemeter (FT9) equal to 3.000000e-12 and all others equal to 0.
>
> And when I'm trying to generate events I get much less of them then I requested.
>
> The error is the following:
> INFO: fail to reach target 10000
> === Results Summary for run: run_01 tag: tag_1 ===
>
> Cross-section : 0.07003 +- 0.0001146 pb
> Nb of events : 1744
>
> Param_card, run_card and the log are available here: https://cernbox.cern.ch/index.php/s/28A10yOxIAs420b
>
> I know that this problem was mentioned before in one of the tickets: https://answers.launchpad.net/mg5amcnlo/+question/683400
> but there were no clear conclusion on this.
>
> With the best regards,
> Evgeny
>
> --
> You received this question notification because you are an answer
> contact for MadGraph5_aMC@NLO.

Evgeny Soldatov (esoldato) said : #2

Dear Olivier,

I've tried to use restrictions cards for the model, however still I have the same problem:

INFO: fail to reach target 10000
  === Results Summary for run: run_01 tag: tag_1 ===

     Cross-section : 0.116 +- 0.0001892 pb
     Nb of events : 783

What exactly I've done in the restrictions card:
I've set all the anomalous parameters (ANOINPUTS) except one (FT9) to zero. Only FT9 was set to other value.
Also I've tried b-quark mass to be zero also.
All these (equal to zero) parameters disappeared in the param_card, which was used for generation.
However as I mentioned it did not help.

My restriction_card can be found here: https://cernbox.cern.ch/index.php/s/qx6HLwSr5FRNVtg
Maybe I've done something in the wrong way...

With the best regards,
Evgeny

As I said this is not surprising for dimension 8 EFT model.
One idea that you can try is to pass in Feynman Gauge it sometimes improves efficiency
(set gauge Feynman)

Cheers,

Olivier

> On 16 Feb 2020, at 11:38, Evgeny Soldatov <email address hidden> wrote:
>
> Question #688567 on MadGraph5_aMC@NLO changed:
> https://answers.launchpad.net/mg5amcnlo/+question/688567
>
> Status: Answered => Open
>
> Evgeny Soldatov is still having a problem:
> Dear Olivier,
>
> I've tried to use restrictions cards for the model, however still I have
> the same problem:
>
> INFO: fail to reach target 10000
> === Results Summary for run: run_01 tag: tag_1 ===
>
> Cross-section : 0.116 +- 0.0001892 pb
> Nb of events : 783
>
> What exactly I've done in the restrictions card:
> I've set all the anomalous parameters (ANOINPUTS) except one (FT9) to zero. Only FT9 was set to other value.
> Also I've tried b-quark mass to be zero also.
> All these (equal to zero) parameters disappeared in the param_card, which was used for generation.
> However as I mentioned it did not help.
>
> My restriction_card can be found here: https://cernbox.cern.ch/index.php/s/qx6HLwSr5FRNVtg
> Maybe I've done something in the wrong way...
>
> With the best regards,
> Evgeny
>
> --
> You received this question notification because you are an answer
> contact for MadGraph5_aMC@NLO.

Evgeny Soldatov (esoldato) said : #4

Dear Olivier,

Now it is a bit better:

INFO: fail to reach target 10000
  === Results Summary for run: run_01 tag: tag_1 ===

     Cross-section : 0.1169 +- 0.0002835 pb
     Nb of events : 8664

However still not all events are generated.

What can be done to be able to generate all requested events?
1) Should we change anything in selection? (Our run_card.dat is here: https://cernbox.cern.ch/index.php/s/Pyj4O9pbMT8SmNS )
2) Or should we change anything in model? (we are in contact with author of the model and he is working on the new model at the moment)

If it is impossible to generate all events...
1) Can we trust the distributions of the generated events or not? (if integration fails in the definite phase space regions, distributions can be wrong, can't they?)
2) The resulting cross section should be scaled to the actual number of events, right? (For the case mentioned above in this message it should be scaled on 8664/10000, right?)

Just to clarify what is our target: we would like to use decomposition method in MadGraph, but we need to be able to validate it (compare the sum of quadratic, interference and SM terms with the full amplitude).

With the best regards,
Evgeny

Hi,

This helps quite a lot actually.

What I would here is to move to multi_run and/or gridpack mode for generation.
multi_run is slower but has the advantages that you have different computation of the cross-section and then you can check if the cross-section is stable or not.
(For processes where we fail to reach the target number of events, some of them have also issue with providing stable cross-section)

>
> Just to clarify what is our target: we would like to use decomposition
> method in MadGraph, but we need to be able to validate it (compare the
> sum of quadratic, interference and SM terms with the full amplitude).

If you have already this issue with the amplitude square, this is unlikely that the interference term
(where the phase-space integrator is everything but optimised) will converge/produce the expected number of events.

Cheers,

Olivier

> On 17 Feb 2020, at 01:53, Evgeny Soldatov <email address hidden> wrote:
>
> Question #688567 on MadGraph5_aMC@NLO changed:
> https://answers.launchpad.net/mg5amcnlo/+question/688567
>
> Status: Answered => Open
>
> Evgeny Soldatov is still having a problem:
> Dear Olivier,
>
> Now it is a bit better:
>
> INFO: fail to reach target 10000
> === Results Summary for run: run_01 tag: tag_1 ===
>
> Cross-section : 0.1169 +- 0.0002835 pb
> Nb of events : 8664
>
>
> However still not all events are generated.
>
> What can be done to be able to generate all requested events?
> 1) Should we change anything in selection? (Our run_card.dat is here: https://cernbox.cern.ch/index.php/s/Pyj4O9pbMT8SmNS )
> 2) Or should we change anything in model? (we are in contact with author of the model and he is working on the new model at the moment)
>
> If it is impossible to generate all events...
> 1) Can we trust the distributions of the generated events or not? (if integration fails in the definite phase space regions, distributions can be wrong, can't they?)
> 2) The resulting cross section should be scaled to the actual number of events, right? (For the case mentioned above in this message it should be scaled on 8664/10000, right?)
>
> Just to clarify what is our target: we would like to use decomposition
> method in MadGraph, but we need to be able to validate it (compare the
> sum of quadratic, interference and SM terms with the full amplitude).
>
> With the best regards,
> Evgeny
>
> --
> You received this question notification because you are an answer
> contact for MadGraph5_aMC@NLO.

Can you help with this problem?

Provide an answer of your own, or ask Evgeny Soldatov for more information if necessary.

To post a message you must log in.