# Light by light scattering with MG4@nlo

Asked by Schoeffel on 2018-05-14

Dear experts,

Recently, we have considered the possibility to simulate light-by-light (LbL) event using MadGraph5@NLO (MG).
The process is pp -> pp (gamma gamma) -> p p gamma gamma.
We have shown some results/questions in:

To summarize:

1) MG seems to compute all relevant loop diagrams needed for the process.

2) we use the elastic-elastic mode: tag_2
We understand that the standard mode is beam type 2 (see page 9 of the talk).
In this case, MG does not produce reasonable results (while the diagrams considered are correct).
The cross section values computed with MG5 are completely off w.r.t. known theoretical values...
And we do not understand why.

Thanks for your help

## Question information

Language:
English Edit question
Status:
For:
Assignee:
No assignee Edit question
Last query:
2018-05-14
2018-05-14
 Olivier Mattelaer (olivier-mattelaer) said on 2018-05-14: #1

Hi,

What is tag_2?

Otherwise
"lpp1=2" indeed means elastic photon from proton.
The associated flux is taken from
Equivalent photon approximation structure function.
Improved Weizsaecker-Williams formula
V.M.Budnev et al., Phys.Rep. 15C (1975) 181
Even if this is irrelevant since you seem to have hack that flux according to your slide.
Are you sure that you hacked it correctly?

if you use lpp1=1. It means that you use the photon contribution from the pdf set.
Depending of the PDF that you use this contribution can
be zero/ inelastic component only / elastic component only / elastic+inelastic component.
I would suggest to read the documentation of the pdf set that you are using to see which convention is used.
Most of those PDF are associated to very large error as well.

Otherwise since you seem to hack MG5 (which you can obviously do) it is impossible for us to comment.
Did you try without the hack? Why do you need the hack?

Cheers,

Olivier

PS: What is pphoton?

> On 14 May 2018, at 12:03, Schoeffel <email address hidden> wrote:
>
> New question #668999 on MadGraph5_aMC@NLO:
>
>
> Dear experts,
>
>
>
> Recently, we have considered the possibility to simulate light-by-light (LbL) event using MadGraph5@NLO (MG).
> The process is pp -> pp (gamma gamma) -> p p gamma gamma.
> We have shown some results/questions in:
>
> To summarize:
>
> 1) MG seems to compute all relevant loop diagrams needed for the process.
>
> 2) we use the elastic-elastic mode: tag_2
> We understand that the standard mode is beam type 2 (see page 9 of the talk).
> In this case, MG does not produce reasonable results (while the diagrams considered are correct).
> The cross section values computed with MG5 are completely off w.r.t. known theoretical values...
> And we do not understand why.
>
> Thanks for your help
>
>
>
>
>
> --
> You received this question notification because you are an answer
> contact for MadGraph5_aMC@NLO.

 Schoeffel (laurentcea) said on 2018-05-14: #2

Hello

Thanks for the answer.

For what concerned MG5, we have used
lpp1=2 (without any modification)
the indications on the slides are just here to show that this is possible to make the calculations per se
as it is done in the literature... that we use to compare with values extracted from MG.
in practice, we compare the values extracted with a code called FPMC, itself checked with complete integral calculations.

So, as you state in your answer, we understand that we use MG with the well-known elastic flux of photons but find
non relevant values for the light by light cross section.
If fact, even the dependence of xs as a function of the mass of the gamma-gamma makes no sense.

Therefore, there is something really strange...
(as if most of the diagrams computed were damped for a reason or another).

But, most probably, this can not be solved in a discussion chain :)

Laurent

 Olivier Mattelaer (olivier-mattelaer) said on 2018-05-14: #3

Hi,

So looks like I understood even less than I thought from your slides.

So can you confirm that both computation are actually computing the same quantity?
Do I understand correctly that they miss some contribution?
Did you try to remove those diagrams in MG5 to have the same diagram in each case?

How did you set the factorization scale?
Did you use fix scale? and what is used in the other computation?
Can you also vary the factorization scale to see the dependence in that parameter?
If you can ensure that both tools use the same it would be even better.

A final thought is about numerical accuracy. Phase space integration of loop-induced process is quite tricky on its own but actually photon induced are also tricky on their own. So did you check numerical stability aver different seed?

Cheers,

Olivier

> On 14 May 2018, at 18:21, Schoeffel <email address hidden> wrote:
>
> Question #668999 on MadGraph5_aMC@NLO changed:
>
> Schoeffel posted a new comment:
>
> Hello
>
> Thanks for the answer.
>
> For what concerned MG5, we have used
> lpp1=2 (without any modification)
> the indications on the slides are just here to show that this is possible to make the calculations per se
> as it is done in the literature... that we use to compare with values extracted from MG.
> in practice, we compare the values extracted with a code called FPMC, itself checked with complete integral calculations.
>
> So, as you state in your answer, we understand that we use MG with the well-known elastic flux of photons but find
> non relevant values for the light by light cross section.
> If fact, even the dependence of xs as a function of the mass of the gamma-gamma makes no sense.
>
> Therefore, there is something really strange...
> (as if most of the diagrams computed were damped for a reason or another).
>
> But, most probably, this can not be solved in a discussion chain :)
>
> Laurent
>
> --
> You received this question notification because you are an answer
> contact for MadGraph5_aMC@NLO.

 Schoeffel (laurentcea) said on 2018-05-14: #4

Hello

Yes, the computation should be exactly equal for each line of page 9 for example.
what is under the column FPMC is correct, at least reproducible by a pure calculation of an integral.

We have not investigated about the factorization scale (so I guess, we have kept the default one): this gives us a hint to make a move in some direction. Thanks.
For the numerical accuracy, this seems to be stable at the value quoted in the talk...

L

## Can you help with this problem?

To post a message you must log in.