MG5_aMC_v2_6_1 Madevent generation efficiency for WW scattering with same sign

Hi Andre,

Sorry for the late reply, I'm only working part time for the physics department and therefore I can not snap fix MG5aMC all the time.

Can you try the following path and tell me
1) if you have the same cross-section (I'm slightly worry about a missing jacobian)?
2) if this fix the efficiency issue?

I'm actually testing such patch for a different issue but I guess that this might be the same patch for both problem.
If this is the real fix (up to the potential jacobian issue, you are actually lucky since this type of efficiency issue are something that I try (and fail) to fix for years. So if this is the solution we can really open a bottle of champaign here.

Cheers,

Olivier

and the patch

=== modified file 'Template/LO/SubProcesses/genps.f'
--- Template/LO/SubProcesses/genps.f 2017-08-09 12:56:57 +0000
+++ Template/LO/SubProcesses/genps.f 2018-03-21 20:44:02 +0000
@@ -849,43 +849,43 @@
          m12 = m(itree(2,ibranch))**2
          mn2 = m(ibranch-1)**2
 c write(*,*) 'Enertering yminmax',sqrt(s1),sqrt(m12),sqrt(mn2)
          call yminmax(s1,t,m12,ma2,mb2,mn2,tmin,tmax)
 c
 c Call for 0<x<1
 c
 c call sample_get_x(wgt,x(-ibranch),-ibranch,iconfig,
 c & .5d0*(tmin/stot+1d0),
 c & .5d0*(tmax/stot+1d0))
 c t = Stot*(x(-ibranch)*2d0-1d0)
 c
 c call for -1<x<1
 c

 c write(*,*) 'tmin, tmax',tmin,tmax

- tmax = max(tmax,0d0) !This line if want really t freedom
+c tmax = max(tmax,0d0) !This line if want really t freedom

          call sample_get_x(wgt,x(-ibranch),-ibranch,iconfig,
      $ -tmax/stot, -tmin/stot)
          t = stot*(-x(-ibranch))
 c
 c now reset tmax if messed it up for t freedom 3 lines above
 c
- call yminmax(s1,t,m12,ma2,mb2,mn2,tmin,tmax)
+c call yminmax(s1,t,m12,ma2,mb2,mn2,tmin,tmax)

 c write(*,*) tmin,t,tmax
 c if (t .eq. 0d0) then
 c jac = -3
 c return
 c endif
          if (t .lt. tmin .or. t .gt. tmax) then
             jac=-3d0
             return
          endif
 c
 c tmin and tmax set to -s,+s for jacobian because part of jacobian
 c was determined from choosing the point x from the grid based on
 c tmin and tmax. (ie wgt contains some of the jacobian)
 c
          tmin=-stot
          tmax= stot

Hi Andre,

I have some test on this patch (and will run more on it) and so far so good
here is a test:
generate p p > w+ w+ j j , w+ > l+ vl
output
launch
set mmjj 500
set deltaeta 2.5

And here are the results before/after the above patch:

Survey before patch: (2m 53s) 6.0e-3 +- 1.6e-4 (13 channels to improve -> 19 jobs submitted)
Survey after patch: (1m48s) 5.6e-3+-2.2e-4 (11 channels to improve-> 19 jobs submitted)
Refine before patch: (16m 25s) 5.9e-3+-1.4e-5 (10k events)
Refine after patch: (11m51s) 5.9e-3+-1.4e-5 (10k events)

So the code is ~ 30% faster. This sounds cool but might not be enough for your process. (at the same time, adding the decay like I did complexify the process even more.) I'm now running the same process at 100TeV to see the impact of the patch.

Cheers,

Olivier

For 100TeV,

This is improving the situation but not that much

Before the patch, I was generating only 97 events,=.
After the patch, I'm generating 1111 events, much smaller than the number of events requested actually.

Now the gain is actually more than a factor of 10 since it was taking me a total of 3h to generate 97 events and less than 2h to generate the 1111 events after the patch.

Cheers,

Olivier

Hi Olivier,
Which PDF are you using for the 100TeV ?
Are you sure the PDF has no role in this problem ?
When I use NNPDF30_nlo_as_00118 the efficiency is much higher ...
Cheers,
Andre

On Mar 22, 2018, at 12:46 PM, Olivier Mattelaer <<email address hidden><mailto:<email address hidden>>> wrote:

Your question #665762 on MadGraph5_aMC@NLO changed:
https://answers.launchpad.net/mg5amcnlo/+question/665762

Olivier Mattelaer proposed the following answer:
For 100TeV,

This is improving the situation but not that much

Before the patch, I was generating only 97 events,=.
After the patch, I'm generating 1111 events, much smaller than the number of events requested actually.

Now the gain is actually more than a factor of 10 since it was taking me
a total of 3h to generate 97 events and less than 2h to generate the
1111 events after the patch.

Cheers,

Olivier

--
If this answers your question, please go to the following page to let us
know that it is solved:
https://answers.launchpad.net/mg5amcnlo/+question/665762/+confirm?answer_id=3

If you still need help, you can reply to this email or go to the
following page to enter your feedback:
https://answers.launchpad.net/mg5amcnlo/+question/665762

You received this question notification because you asked the question.

===================================================
Andre Sznajder
Professor Associado
Instituto de Fisica - Dept. DFNAE
Universidade do Estado do Rio de Janeiro ( UERJ )
Rio de Janeiro, RJ - Brasil
Tel.: (+55)(21)23340608 ramal:24 Fax: (+55)(21)23340483
Email: <email address hidden><mailto:<email address hidden>>
===================================================

Well, it can have an impact but nothing that I can really control on that side.

Cheers,

Olivier

> On 22 Mar 2018, at 17:02, Andre Sznajder <email address hidden> wrote:
>
> Question #665762 on MadGraph5_aMC@NLO changed:
> https://answers.launchpad.net/mg5amcnlo/+question/665762
>
> Status: Answered => Open
>
> Andre Sznajder is still having a problem:
> Hi Olivier,
> Which PDF are you using for the 100TeV ?
> Are you sure the PDF has no role in this problem ?
> When I use NNPDF30_nlo_as_00118 the efficiency is much higher ...
> Cheers,
> Andre
>
> On Mar 22, 2018, at 12:46 PM, Olivier Mattelaer
> <<email address hidden><mailto:<email address hidden>>>
> wrote:
>
> Your question #665762 on MadGraph5_aMC@NLO changed:
> https://answers.launchpad.net/mg5amcnlo/+question/665762
>
> Olivier Mattelaer proposed the following answer:
> For 100TeV,
>
> This is improving the situation but not that much
>
> Before the patch, I was generating only 97 events,=.
> After the patch, I'm generating 1111 events, much smaller than the number of events requested actually.
>
> Now the gain is actually more than a factor of 10 since it was taking me
> a total of 3h to generate 97 events and less than 2h to generate the
> 1111 events after the patch.
>
> Cheers,
>
> Olivier
>
> --
> If this answers your question, please go to the following page to let us
> know that it is solved:
> https://answers.launchpad.net/mg5amcnlo/+question/665762/+confirm?answer_id=3
>
> If you still need help, you can reply to this email or go to the
> following page to enter your feedback:
> https://answers.launchpad.net/mg5amcnlo/+question/665762
>
> You received this question notification because you asked the question.
>
>
> ===================================================
> Andre Sznajder
> Professor Associado
> Instituto de Fisica - Dept. DFNAE
> Universidade do Estado do Rio de Janeiro ( UERJ )
> Rio de Janeiro, RJ - Brasil
> Tel.: (+55)(21)23340608 ramal:24 Fax: (+55)(21)23340483
> Email: <email address hidden><mailto:<email address hidden>>
> ===================================================
>
> --
> You received this question notification because you are an answer
> contact for MadGraph5_aMC@NLO.