Problem with Drell-Yan generation at NLO

Asked by attilio santocchia on 2017-09-12

Hello,
I'm trying to generate Drell-Yan events at NLO and I got some errors which I'm not able to solve. Here is the situation:

1) CMS collaboration use a specific script to run on LSF at cern the gridpack generation and different cards have been used several times to produce different gridpack process.

2) The specific process is a Drell-Yan production at NLO with 1GeV<m(ll)<10GeV. CMS has already produced with the same cards 2 gridpack (10GeV<m(ll)<50GeV and m(ll)>50GeV). The lower cut on m(ll) is defined in the run.card while the upper bound is defined in a cut.f file.

3) The process is include the ll+0jet, ll+1jet and ll+2jet.

4) The error and the card used are in the link [0] below

5) I tried to split the process and the ll+0jet is the problem, the ll+1jet is correctly run and the ll+2j is running right and I don't know yet if it will complete or not but I already know that the ll+0jet process is the problem

Could you give some indication on how to solve the problem?
Thanks

Attilio

https://www.dropbox.com/sh/hx05ta179ld5dl3/AABhoeR_Q1kUCU3Z2Eb6cjzNa?dl=0

Question information

Language:
English Edit question
Status:
Needs information
For:
MadGraph5_aMC@NLO Edit question
Assignee:
Rikkert Frederix Edit question
Last query:
2017-09-21
Last reply:
2017-09-22
attilio santocchia (santocch) said : #1

Hello,
I did some search and test and that's what I found...

1) The errors seems due to too stringent cuts... I found this old thread which report an error similar to what I got (comparing the end of the log.txt file) [0]

2) The only cuts I'm using are the m(ll) cuts and so I did some test on the 0jet sample... so the proc_card now has just the following code:
define p = p b b~
define ell+ = e+ mu+ ta+
define ell- = e- mu- ta-
generate p p > ell+ ell- [QCD] @0
and I tried to see what was going on changing the m(ll) upper bound which is set in the cut.f file:
1<m(ll)<10 fails
1<m(ll)<50 fails
1<m(ll)<70 works...

3) I'm going to try now the 2jets sample to see what is the upper bound for m(ll) compatible with the m(ll)=1 lower bound.

So now I have an additional question... is there any way to increase the number of points that pass the cuts mentioned in answer #3 of the [0] thread?

Thanks

Attilio

[0] https://answers.launchpad.net/mg5amcnlo/+question/446723

attilio santocchia (santocch) said : #2

Hello,
I'm still trying to find a way to have DY events with m(ll) in the range 1-10GeV... I tried to run the full process:

generate p p > ell+ ell- [QCD] @0
add process p p > ell+ ell- j [QCD] @1
add process p p > ell+ ell- j j [QCD] @2

without any upper bound on m(ll) and using:
  1.0 = mll_sf ! Min inv. mass of all opp. sign same-flavor lepton pairs
in the run card.

The gridpack has been produced but in the log file I see the following lines:

INFO: Updating the number of unweighted events per channel

      Intermediate results:
      Random seed: 33
      Total cross section: nan +- nan pb
      Total abs(cross section): nan +- nan pb

Is that a problem? The final cross section seems to me strange:

INFO:
   --------------------------------------------------------------
      Summary:
      Process p p > ell+ ell- [QCD] @0 ; p p > ell+ ell- j [QCD] @1 ; p p > ell+ ell- j j [QCD] @2
      Run at p-p collider (6500.0 + 6500.0 GeV)
      Number of events generated: 300
      Total cross section: 2.564e-01 +- 3.1e-02 pb
   --------------------------------------------------------------
  Number of loop ME evaluations (by MadLoop): 5377389
    Stability unknown: 0
    Stable PS point: 5372960
    Unstable PS point (and rescued): 4424
    Unstable PS point (and not rescued): 5
    Only double precision used: 5372960
    Quadruple precision used: 4429
    Initialization phase-space points: 0
    Reduction methods used:
      > Ninja (double precision) 5372772
      > Ninja (quadruple precision) 4405
      > CutTools (double precision) 188
      > CutTools (quadruple precision) 19
      > Not identified (CTModeRun != -1) 5
  Total number of unstable PS point detected: 5 (0.00%)
    Maximum fraction of UPS points in channel processtmp/SubProcesses/P2_dd_epemdd/GF1 (0.01%)
    Please report this to the authors while providing the file
    /pool/lsf/santocch/88864664/dyellell012j_5f_NLO_FXFX_M1toInf/dyellell012j_5f_NLO_FXFX_M1toInf_gridpack/work/processtmp/SubProcesses/P2_dd_epemdd/GF1/UPS.log

The gridpack is dyellell012j_5f_NLO_FXFX_M1toInf_slc6_amd64_gcc481_CMSSW_7_1_28_tarball.tar.xz in the following directory:

https://www.dropbox.com/sh/hx05ta179ld5dl3/AABhoeR_Q1kUCU3Z2Eb6cjzNa?dl=0

Could you have a look and give me some hint on how to proceed? I really don't know what to try now and... if this is a mission impossible... well just tell me and I will try to find another solution for the small m(ll) range.

Thanks

Attilio

Rikkert Frederix (frederix) said : #3

Dear Attilio,

Sorry for not replying to you earlier.

I believe the problem is the very low invariant mass cut on the leptons. In this region perturbation theory doesn't work very well, and I'm not surprised that the code doesn't converge properly and you get either zero, NaN, or other non-sensical results. I think you seriously need to increase this cut.

I also see that the lower generation cut on the jet pT is 10 GeV. I guess that this is for a merging scale of 20-30 GeV or so, right? However, this cannot be consistent: with the 1 GeV < m_ll <10 GeV cuts, the typical scale of your process is only a couple of GeV. And the merging scale should be smaller than the typical hard scale of the process. So, the merging scale should be 1 or 2 GeV at most, which also means that the generation cut on the jet pT should be reduced considerably. However, this clearly goes into the non-perturbative region which cannot be correctly simulated with methods based on perturbation theory, such as matrix elements and FxFx merging (and it definitely does not make sense to try to include NLO corrections).

Hence, to be short, I don't think it's possible to go well below the 10 GeV lepton invariant mass cut. Maybe down to 5 GeV, but certainly not below that.

Best,
Rikkert

attilio santocchia (santocch) said : #4

Dear Rikkert

Thanks for the clarification… just let me ask one more question. The situation I have is the following… I’m trying to have a reasonable prediction of the background due to DY events with mu+mu- in the mass range 1<m(ll)<10. I cannot rely on MC prediction because of the limitation you explained to me and I need to find another solution. I was however able to produce (without any error) a gridpack with madgraph at LO with the following request:

m(ll)>1GeV
ht>50GeV

I had a look at these events with delphes and there are (as expected) few events with m(ll) in the range I’d like to study… not many but producing a lot of events I could have enough of them to do some basic study… So my question is:

The events I could produce with this gridpack are OK for a study where I try to demonstrate that a measurement of the background from data is correct?

I know that the xSection is not the correct one (I could correct it using the k-factor we use in CMS which is 1.23) but I don’t really need it… In other words what I would like to do is the following:
1) Generate events using the LO gridpack
2) Verify that one of the method we use in CMS for background evaluation from data is OK using the MC events generated in (1)
3) Apply the method with CMS data

Do you have any comments, suggestion on this idea? Or it’s definitely not a good way to proceed?
Thanks for helping

Attilio

> Il giorno 20 set 2017, alle ore 09:09, Rikkert Frederix <email address hidden> ha scritto:
>
> Your question #657903 on MadGraph5_aMC@NLO changed:
> https://answers.launchpad.net/mg5amcnlo/+question/657903
>
> Status: Open => Answered
>
> Rikkert Frederix proposed the following answer:
> Dear Attilio,
>
> Sorry for not replying to you earlier.
>
> I believe the problem is the very low invariant mass cut on the leptons.
> In this region perturbation theory doesn't work very well, and I'm not
> surprised that the code doesn't converge properly and you get either
> zero, NaN, or other non-sensical results. I think you seriously need to
> increase this cut.
>
> I also see that the lower generation cut on the jet pT is 10 GeV. I
> guess that this is for a merging scale of 20-30 GeV or so, right?
> However, this cannot be consistent: with the 1 GeV < m_ll <10 GeV cuts,
> the typical scale of your process is only a couple of GeV. And the
> merging scale should be smaller than the typical hard scale of the
> process. So, the merging scale should be 1 or 2 GeV at most, which also
> means that the generation cut on the jet pT should be reduced
> considerably. However, this clearly goes into the non-perturbative
> region which cannot be correctly simulated with methods based on
> perturbation theory, such as matrix elements and FxFx merging (and it
> definitely does not make sense to try to include NLO corrections).
>
> Hence, to be short, I don't think it's possible to go well below the 10
> GeV lepton invariant mass cut. Maybe down to 5 GeV, but certainly not
> below that.
>
> Best,
> Rikkert
>
> --
> 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/657903/+confirm?answer_id=2
>
> 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/657903
>
> You received this question notification because you asked the question.

Rikkert Frederix (frederix) said : #5

Dear Attilio,

It all depends on the details of what you are trying to do, I think. In general, the considerations relevant for FxFx also apply to the LO sample. Hence, one has to be extremely careful with introducing these small scales (m(ll)~1GeV) into a merged sample.

However, the ht>50GeV cut you introduce above, is that on the jets? If so, this is not consistent. You cannot put any cuts on the jets other than the cuts relevant for the merging --hence you have to include a sample with ht<50GeV as well.

Best,
Rikkert

Can you help with this problem?

Provide an answer of your own, or ask attilio santocchia for more information if necessary.

To post a message you must log in.