Wrong systematic weights with custom scale in 2.5.1

Dear Andrey,

Yes this problem is already fixed in 2.5.2.
I do not remember what the problem was and what was the associated fix.
But I ran today this exact run with three version
2.5.1, 2.5.2 and the development version of 2.5.3 (which should be released this week)

Indeed I observe this unphysical behaviour in 2.5.1 but not in the two other version.
(You can check by yourself the (long) result —at the level of the cross-section— below for each version.

Thanks,

Olivier

 Results for 2.5.1
# mur muf alpsfact dynamical_scale pdf cross-section
1.0 1.0 1.0 -1 247000 473.189057477
0.5 0.5 1.0 -1 247000 612.887375922
0.5 0.5 1.0 1 247000 625.583704316
0.5 0.5 1.0 2 247000 504.220552572
0.5 0.5 1.0 3 247000 657.309499616
0.5 0.5 1.0 4 247000 473.189057515
0.5 1.0 1.0 -1 247000 532.745999295
0.5 1.0 1.0 1 247000 585.706191655
0.5 1.0 1.0 2 247000 467.512720944
0.5 1.0 1.0 3 247000 615.442341686
0.5 1.0 1.0 4 247000 439.631601469
0.5 2.0 1.0 -1 247000 496.155957037
0.5 2.0 1.0 1 247000 547.150462193
0.5 2.0 1.0 2 247000 434.461435278
0.5 2.0 1.0 3 247000 570.694746182
0.5 2.0 1.0 4 247000 409.361696789
1.0 0.5 1.0 -1 247000 473.189057484
1.0 0.5 1.0 1 247000 514.067892032
1.0 0.5 1.0 2 247000 425.32914424
1.0 0.5 1.0 3 247000 538.605795853
1.0 0.5 1.0 4 247000 400.456204592
1.0 1.0 1.0 1 247000 480.953681863
1.0 1.0 1.0 2 247000 394.347303764
1.0 1.0 1.0 3 247000 504.220552572
1.0 1.0 1.0 4 247000 372.021896802
1.0 2.0 1.0 -1 247000 409.361696761
1.0 2.0 1.0 1 247000 448.526749114
1.0 2.0 1.0 2 247000 366.454791568
1.0 2.0 1.0 3 247000 467.512720944
1.0 2.0 1.0 4 247000 346.378637723
2.0 0.5 1.0 -1 247000 400.456204568
2.0 0.5 1.0 1 247000 432.410007977
2.0 0.5 1.0 2 247000 363.653134827
2.0 0.5 1.0 3 247000 454.365400669
2.0 0.5 1.0 4 247000 343.304014099
2.0 1.0 1.0 -1 247000 372.021896779
2.0 1.0 1.0 1 247000 404.371732772
2.0 1.0 1.0 2 247000 337.151712123
2.0 1.0 1.0 3 247000 425.32914424
2.0 1.0 1.0 4 247000 318.902367639
2.0 2.0 1.0 -1 247000 346.378637702
2.0 2.0 1.0 1 247000 376.765003085
2.0 2.0 1.0 2 247000 313.295064406
2.0 2.0 1.0 3 247000 394.347303764
2.0 2.0 1.0 4 247000 296.899870078

Systematics Result for 2.5.2:

# mur muf alpsfact dynamical_scale pdf cross-section
1.0 1.0 1.0 -1 247000 473.189057477
0.5 0.5 1.0 -1 247000 612.593230434
0.5 0.5 1.0 1 247000 625.016816867
0.5 0.5 1.0 2 247000 504.028350297
0.5 0.5 1.0 3 247000 656.726758186
0.5 0.5 1.0 4 247000 473.189057519
0.5 1.0 1.0 -1 247000 573.303863895
0.5 1.0 1.0 1 247000 585.488464124
0.5 1.0 1.0 2 247000 467.591590416
0.5 1.0 1.0 3 247000 615.200688816
0.5 1.0 1.0 4 247000 439.870440581
0.5 2.0 1.0 -1 247000 533.049100651
0.5 2.0 1.0 1 247000 547.26085461
0.5 2.0 1.0 2 247000 434.764781108
0.5 2.0 1.0 3 247000 570.785584736
0.5 2.0 1.0 4 247000 409.799255743
1.0 0.5 1.0 -1 247000 505.768297502
1.0 0.5 1.0 1 247000 513.612906404
1.0 0.5 1.0 2 247000 425.168462091
1.0 0.5 1.0 3 247000 538.135444597
1.0 0.5 1.0 4 247000 400.451860017
1.0 1.0 1.0 1 247000 480.786381133
1.0 1.0 1.0 2 247000 394.414827927
1.0 1.0 1.0 3 247000 504.028350297
1.0 1.0 1.0 4 247000 372.219451569
1.0 2.0 1.0 -1 247000 439.87044054
1.0 2.0 1.0 1 247000 448.627912691
1.0 2.0 1.0 2 247000 366.711279205
1.0 2.0 1.0 3 247000 467.591590416
1.0 2.0 1.0 4 247000 346.744158826
2.0 0.5 1.0 -1 247000 428.080047472
2.0 0.5 1.0 1 247000 432.02647975
2.0 0.5 1.0 2 247000 363.517098914
2.0 0.5 1.0 3 247000 453.970536069
2.0 0.5 1.0 4 247000 343.297333846
2.0 1.0 1.0 -1 247000 400.451859985
2.0 1.0 1.0 1 247000 404.230176857
2.0 1.0 1.0 2 247000 337.210467392
2.0 1.0 1.0 3 247000 425.168462091
2.0 1.0 1.0 4 247000 319.068609426
2.0 2.0 1.0 -1 247000 372.219451538
2.0 2.0 1.0 1 247000 376.848624011
2.0 2.0 1.0 2 247000 313.515100564
2.0 2.0 1.0 3 247000 394.414827927
2.0 2.0 1.0 4 247000 297.209955627

Results for 2.5.3 (development version)
# mur muf alpsfact dynamical_scale pdf cross-section
1.0 1.0 1.0 -1 247000 472.697205018
0.5 0.5 1.0 -1 247000 611.616451224
0.5 0.5 1.0 1 247000 619.056440226
0.5 0.5 1.0 2 247000 501.668690718
0.5 0.5 1.0 3 247000 653.013513161
0.5 0.5 1.0 4 247000 472.697204986
0.5 1.0 1.0 -1 247000 572.785601786
0.5 1.0 1.0 1 247000 580.507255709
0.5 1.0 1.0 2 247000 465.747266577
0.5 1.0 1.0 3 247000 612.164832803
0.5 1.0 1.0 4 247000 439.669850811
0.5 2.0 1.0 -1 247000 532.87829038
0.5 2.0 1.0 1 247000 541.944041807
0.5 2.0 1.0 2 247000 433.334592345
0.5 2.0 1.0 3 247000 568.394463384
0.5 2.0 1.0 4 247000 409.821402875
1.0 0.5 1.0 -1 247000 504.892947096
1.0 0.5 1.0 1 247000 508.986274104
1.0 0.5 1.0 2 247000 423.223965761
1.0 0.5 1.0 3 247000 535.234875437
1.0 0.5 1.0 4 247000 400.009249085
1.0 1.0 1.0 1 247000 476.858914556
1.0 1.0 1.0 2 247000 392.900313918
1.0 1.0 1.0 3 247000 501.668690718
1.0 1.0 1.0 4 247000 372.025233997
1.0 2.0 1.0 -1 247000 439.669850841
1.0 2.0 1.0 1 247000 444.872411061
1.0 2.0 1.0 2 247000 365.542031591
1.0 2.0 1.0 3 247000 465.747266577
1.0 2.0 1.0 4 247000 346.740102266
2.0 0.5 1.0 -1 247000 427.310823238
2.0 0.5 1.0 1 247000 428.288267498
2.0 0.5 1.0 2 247000 361.88774412
2.0 0.5 1.0 3 247000 451.574947503
2.0 0.5 1.0 4 247000 342.900401356
2.0 1.0 1.0 -1 247000 400.00924911
2.0 1.0 1.0 1 247000 401.05603804
2.0 1.0 1.0 2 247000 335.945375358
2.0 1.0 1.0 3 247000 423.223965761
2.0 1.0 1.0 4 247000 318.885900478
2.0 2.0 1.0 -1 247000 372.025234021
2.0 2.0 1.0 1 247000 374.012594349
2.0 2.0 1.0 2 247000 312.542305181
2.0 2.0 1.0 3 247000 392.900313918
2.0 2.0 1.0 4 247000 297.191383463

> On 6 Mar 2017, at 23:13, Andrey Popov <email address hidden> wrote:
>
> New question #530730 on MadGraph5_aMC@NLO:
> https://answers.launchpad.net/mg5amcnlo/+question/530730
>
> Dear experts,
>
>
> I'm facing a strange problem with systematic weights in MadGraph5_aMC@NLO 2.5.1. When using a specific custom scale, the weight computed for the mu_F = 0.5 variation equals the nominal event weight, which I believe cannot be correct. Below are details on a simplified setup that reproduces the problem.
>
> I generate p p > t t~ in SM. All settings are kept at their defaults except for the dynamic scale, which is set to mtt / 2. The banner and the customized file setscales.f are available here [1]. This is an example generated event:
>
> <event>
> 4 1 +4.7415300e+02 3.10875300e+02 7.54677100e-03 1.07992700e-01
> 21 -1 0 0 501 502 +0.0000000000e+00 +0.0000000000e+00 +6.0325682573e+02 6.0325682573e+02 0.0000000000e+00 0.0000e+00 1.0000e+00
> 21 -1 0 0 502 503 -0.0000000000e+00 -0.0000000000e+00 -1.6020281463e+02 1.6020281463e+02 0.0000000000e+00 0.0000e+00 1.0000e+00
> 6 1 1 2 501 0 -5.4241929470e+01 +2.5027374724e+02 +2.6290089525e+02 4.0574008453e+02 1.7300000000e+02 0.0000e+00 1.0000e+00
> -6 1 1 2 0 503 +5.4241929470e+01 -2.5027374724e+02 +1.8015311585e+02 3.5771955583e+02 1.7300000000e+02 0.0000e+00 1.0000e+00
> <mgrwt>
> <rscale> 2 0.31087528E+03</rscale>
> <asrwt>0</asrwt>
> <pdfrwt beam="1"> 1 21 0.92808741E-01 0.31087528E+03</pdfrwt>
> <pdfrwt beam="2"> 1 21 0.24646587E-01 0.31087528E+03</pdfrwt>
> <totfact> 0.14305511E+04</totfact>
> </mgrwt>
> <rwgt>
> …
> <wgt id='5'> +4.7415300e+02 </wgt>
> …
> <wgt id='16'> +4.7415300e+02 </wgt>
> …
>
> The weight #16 corresponds to mu_F = 0.5:
> <weight id="16" MUR="1.0" MUF="0.5" PDF="247000" > MUF=0.5 </weight>
> As you can see, its numeric value is exactly the same as the nominal weight. This happens for all events.
>
> [1] https://cernbox.cern.ch/index.php/s/mCKveDtBaxuFKnr
>
>
>> From four momenta of particles in this event it is easy to see that the scale is indeed evaluated to mtt / 2. Moreover, weight #5 is defined as
> <weight id="5" MUR="0.5" MUF="0.5" DYN_SCALE="4" PDF="247000" > MUR=0.5 MUF=0.5 dyn_scale_choice=sqrts </weight>
> and therefore its value should equal the nominal weight, which is the case. The pdfrwt tags cite the correct factorization scale.
>
> I tried playing around a bit. If I change the scale in setscales.f to something different, e.g. mtt / 4 or fixed 100 GeV, values of weight #16 start to differ from the nominal value. Setting dynamical_scale_choice = 4 (s-hat) and sys_scalefact = 0.25, 0.5, 1 gives different values for weights for (mu_R = 0.5, mu_F = 0.5) and (mu_R = 0.5, mu_F = 0.25), while this should reproduce the nominal and mu_F = 0.5 weights from the setup discussed above. Finally, running the problematic setup in MadGraph5_aMC@NLO 2.5.2 without any changes gives different nominal and mu_F = 0.5 weights.
>
>
> Could you please explain what is going on here? Was there a bug in SysCalc in 2.5.1, which has been fixed in 2.5.2? Or am I doing something wrong?
>
>
> Best regards,
> Andrey
>
>
> --
> You received this question notification because you are an answer
> contact for MadGraph5_aMC@NLO.

Dear Oliver,

Thanks for checking. Do you know if PDF weights are also affected?

Cheers,
Andrey

Hello Oliver,

Sorry, I've got another follow-up question. The problem is that I've generated a large sample in 2.5.1 and have propagated it through the detector simulation, so redoing it in 2.5.2 is not really an option. But my process is generated at LO, so it should be easy to compute weights for alternative renormalization and factorization weights. I guess the weights should be evaluated as
  w_R = [alpha_s(k * mu) / alpha_s(mu)]^n
and
  w_F = [PDF(x1, k * mu) * PDF(x2, k * mu)] / [PDF(x1, mu) * PDF(x2, mu)],
where mu is the nominal scale evaluated per event, k = 0.5, 2 is its modification scale factor, and n is the number of QCD vertices. Could you please confirm these formulas? Does MadGraph compute alpha_s according to Eq. (4) at page 4 of [1], relating it to the specified alpha_s at the scale of mZ? Sorry for the basic questions.

[1] https://arxiv.org/pdf/hep-ph/0607209.pdf

Best regards,
Andrey

I would say so but it is difficult to guarantee.

Cheers,

Olivier

> On 7 Mar 2017, at 11:26, Andrey Popov <email address hidden> wrote:
>
> Question #530730 on MadGraph5_aMC@NLO changed:
> https://answers.launchpad.net/mg5amcnlo/+question/530730
>
> Andrey Popov posted a new comment:
> Dear Oliver,
>
> Thanks for checking. Do you know if PDF weights are also affected?
>
>
> Cheers,
> Andrey
>
> --
> You received this question notification because you are an answer
> contact for MadGraph5_aMC@NLO.