MadGraph5_aMC@NLO

FxFx differences between 3.4.2 and 3.5.0

Asked by Hannes

Hi,

in our routine validation in ATLAS we found some differences in FxFx merging between 3.4.2 and 3.5.0 that we don't understand.

We simply generate (in 5fs)
    generate p p > l+ l- [QCD] @0
    add process p p > l+ l- j [QCD] @1

then set
    3 = ickkw
    1.0 = jetalgo
    1.0 = jetradius
    8 = ptj

and run Pythia 8.308 with njmax=1 and qcut=20.

In the generated sample we find small differences in jet pT and mass of the softer jets (3 and 4 ordered by pT), as well as the highest jet multiplicity bins. In general 3.5.0 has slightly harder spectra.

The initial suspicion was that this is due to a different shower starting scale in 3.5.0.
Following a suggestion from Rikkert we set:
'improved_scale_choice=.false.'
in 'SubProcesses/fks_singular.f'
in 3.5.0. Strangely this changed the Q value used in the PDF but not much else.

The mentioned plots based on 3.4.2, 3.5.0, and 3.5.0 with 'improved_scale_choice=.false.' can be found here:
https://cernbox.cern.ch/s/hg2W8xj06AMQpGm
(I hope links are allowed)

We would appreciate insights as to whether these differences in 3.5.0 wrt older versions are expected and how we could identify their origin.

Cheers,
Hannes

Question information

Language:
English Edit question
Status:
Solved
For:
MadGraph5_aMC@NLO Edit question
Assignee:
Rikkert Frederix Edit question
Solved by:
Hannes
Solved:
Last query:
Last reply:
Revision history for this message
mattelaer_bot (mattelaer-bot) said : 		#1

Just to ping Rik, since looks like the "assignee" button is currently bug on launchpad...

Revision history for this message
Rikkert Frederix (frederix) said : 		#2

Hello Hannes,

Thanks for the contact. I'll have a look at it next week.

Best,
Rikkert

Revision history for this message
Rikkert Frederix (frederix) said : 		#3

Dear Hannes,

I had a careful look into the code and there were some small things for which the code was fixed. These are all fixed related to the shower starting scale; in particular, for events that are either in the dead-zone, or for some other reason the shower subtraction terms were not necessary to evaluate. For these events, there is no a priori setting for the shower starting scale and small changes should be within the usual uncertainties related to such calculations. Indeed, for your process, which has merging up to and including the first jet, you write that you see small differences for the 3rd and 4th jets. The pT and mass distributions for the jets beyond the merging are indeed quite sensitive to the shower starting scale, and the differences you see are within uncertainties one would expect for such calculations.

It is not really possible to pin down the exact differences. The way the shower scale was assigned for the events that do not require shower subtraction contributions in 3.4.2 was not entirely optimal; this part of the code is quite old and confusing. And we changed the order that some of these terms were computed and that allowed us to improve some of the treatment for the 3.5.0 release.

Note that we are actually refactoring this part of the code again, to really simplify it and clean it up completely. (It is not clear if and when this will be released). And that might again induce these kind of small difference in some observables.

Best,
Rikkert

Revision history for this message
Hannes (hannes3) said : 		#4

Dear Rikkert,

thanks a lot for the detailed reply, this is very helpful.

Best wishes,
Hannes