MadGraph5_aMC@NLO

The matching difference betweenh mg5-2.2.3 and mg5-2.3.2.2

Asked by lily

Hi,
In the ubuntu 12.04 system, I perform the matching of W+0,1,2,3,4jets, w > l v (w= w+, w-) using the version 2.2.3 and 2.3.2.2, respectively.

I set the same cards for the two version, and find that the speed of mg5-2.2.3 is much faster than mg5-2.3.2.2. It will take about one hour for mg5-2.2.3, and five hours for mg5-2.3.2.2 (nevents=100000, 12 cores). Compared to mg5-2.2.3, Mg5-2.3.2.2 spends the extral four hours in the process of "Refine results to 100000 and need to improve channels".

However, for xqcut=15 and qcut=25, the DJR distributions are enough smooth for mg5-2.2.3, but is unacceptable for the Mg5-2.3.2.2. The matched cross sections of the two versions are consistent within the range of uncertainty.

How do understand the differences between the two versions. Since the mg5-2.2.3 is much faster than mg5-2.3.2.2, I prefer to using the mg5-2.2.3. For the matching of w+0,1,2,3,4 jets, is the mg5-2.2.3 result correct?

Many thanks

Question information

Language:
English Edit question
Status:
Answered
For:
MadGraph5_aMC@NLO Edit question
Assignee:
Olivier Mattelaer Edit question
Last query:
Last reply:
Revision history for this message
Olivier Mattelaer (olivier-mattelaer) said : 		#1

Hi,

I'm starting to investigate the speed issue first after that i will look at the distribution, if the matched cross-section are consistent this is surprising that the DJR are significatively different. But I will look at that later.

Cheers,

Olivier

Revision history for this message
degos (degos) said : 		#2

Hi Olivier,

Are there some progresses on this problem?

Many thanks.

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

Investigation is on progress.
Not that much more to report so far.

Cheers,

Olivier

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

Hi,

So I have found two reason for the slow down:
1) some cumulative variable were reset to zero.
This is harmless in term of bias but slow down the code since the cross-section is only based on the latest iteration rather than an weighted average over iteration (with typically the latest one dominating anyway). Therefore the associate error is typically larger and force the code to add iteration to reach precision

If you want to speed up your version, you can apply the following fix:

=== modified file 'Template/LO/Source/dsample.f'
--- Template/LO/Source/dsample.f 2015-09-16 14:21:58 +0000
+++ Template/LO/Source/dsample.f 2015-10-01 10:08:45 +0000
@@ -1608,7 +1608,7 @@
       if (iteration .eq. cur_it) then
 c Add the current point to the DiscreteSamplerGrid
          call add_entry_to_discrete_dimensions(wgt)
- if (kn.eq.0) then
+ if (kn.eq.0.and.(iteration.eq.1.or.use_cut.eq.-2)) then
             ! ensure that all cumulative variable are at zero (usefull for reset)
             twgt1 = 0d0 !
             iavg = 0 !Vars for averging to increase err estimate

2) One new way to handle multi-particle in the initial state which was suppose to be neutral for normal case and improve the speed in presence of CKM. This do not create any bias at all and this can be turned off by adding in the run_card the following lines:
False = mc_grouped_subproc

Cheers,

Olivier

Revision history for this message
degos (degos) said : 		#5

Hi Olivier,
Thanks for your reply.

1. Should I replace "if (kn.eq.0) then" with "if (kn.eq.0.and.(iteration.eq.1.or.use_cut.eq.-2)) then", or
add "if (kn.eq.0.and.(iteration.eq.1.or.use_cut.eq.-2)) then"

2. Is it safe to use the old version mg5-2.2.3 for the w+jets and z+jets?

Many thanks

Revision history for this message
Olivier Mattelaer (olivier-mattelaer) said : 		#6

Hi,

> 1. Should I replace "if (kn.eq.0) then" with "if (kn.eq.0.and.(iteration.eq.1.or.use_cut.eq.-2)) then", or
> add “if (kn.eq.0.and.(iteration.eq.1.or.use_cut.eq.-2)) then"

replace the line

> 2. Is it safe to use the old version mg5-2.2.3 for the w+jets and
> z+jets?

Yes it is.

Olivier

On 03 Oct 2015, at 10:21, degos <email address hidden> wrote:

> Question #271601 on MadGraph5_aMC@NLO changed:
> https://answers.launchpad.net/mg5amcnlo/+question/271601
>
> degos requested more information:
> Hi Olivier,
> Thanks for your reply.
>
> 1. Should I replace "if (kn.eq.0) then" with "if (kn.eq.0.and.(iteration.eq.1.or.use_cut.eq.-2)) then", or
> add "if (kn.eq.0.and.(iteration.eq.1.or.use_cut.eq.-2)) then"
>
> 2. Is it safe to use the old version mg5-2.2.3 for the w+jets and
> z+jets?
>
> Many thanks
>
> --
> 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 lily for more information if necessary.

To post a message you must log in.