Added applgrids for ATLAS WMU measurements

[enocera]

As the title says.

[cschwan]

Are those the grids corresponding to the measurements Mandy was talking about yesterday?

[enocera]

Yes, they are. These correspond to https://arxiv.org/pdf/1904.05631.pdf.

[cschwan]

Thanks, I already had a look at it. Unfortunately this analysis is useless for our NLO EW PDF fit.

[enocera]

Incidentally, I've also produced FK tables out of these, and I made a quick data/theory comparison https://vp.nnpdf.science/I4I4-rnpSEu6U5OtNsPrnw==. I'd be grateful if you could check these applgrids independently with a madgraph/PineAPPL run. Thanks.

[enocera]

Thanks, I already had a look at it. Unfortunately this analysis is useless for our NLO EW PDF fit.

Indeed!

[enocera]

Nevertheless, a double check of the pure QCD NLO predictions is welcome. Thanks.

[cschwan]

@enocera : I'm getting slightly larger results. The only difference that I've found is that your runcards as declare also a pseudo rapidity cut on the neutrino, which I think should not be there. Can you confirm that this makes a difference?

[enocera]

@cschwan Thanks for checking so quickly. Let me rerun MCFM. Incidentally, are the PineAPPL grids or the numbers from your run available somewhere? Thanks.

[cschwan]

Here you go:

$ pineappl --silence-lhapdf convolute ATLAS_WMU_8TEV_WP.pineappl.lz4 NNPDF31_nlo_as_0118
bin                   x1                       diff        integ    neg unc pos unc
---+-------------------+-------------------+-----------+-----------+-------+-------
  0                   0 0.21000000000000002 6.2495254e2 1.3124003e2  -3.24%   2.81%
  1 0.21000000000000002 0.42000000000000004 6.2713114e2 1.3169754e2  -3.27%   2.82%
  2 0.42000000000000004                0.63 6.2745362e2 1.3176526e2  -3.24%   2.83%
  3                0.63  0.8400000000000001 6.3325532e2 1.3298362e2  -3.27%   2.82%
  4  0.8400000000000001                1.05 6.3636563e2 1.3363678e2  -3.29%   2.86%
  5                1.05                1.37 6.4577614e2 2.0664836e2  -3.35%   2.84%
  6                1.37                1.52 6.5113269e2 9.7669903e1  -3.40%   2.87%
  7                1.52                1.74 6.5671102e2 1.4447642e2  -3.53%   2.94%
  8                1.74                1.95 6.5267181e2 1.3706108e2  -3.57%   3.01%
  9                1.95                2.18 6.5671682e2 1.5104487e2  -3.73%   2.94%
 10                2.18                 2.4 6.3581167e2 1.3987857e2  -3.81%   3.00%

$ pineappl --silence-lhapdf convolute ATLAS_WMU_8TEV_WM.pineappl.lz4 NNPDF31_nlo_as_0118
bin                   x1                       diff        integ    neg unc pos unc
---+-------------------+-------------------+-----------+-----------+-------+-------
  0                   0 0.21000000000000002 4.8735733e2 1.0234504e2  -3.54%   2.89%
  1 0.21000000000000002 0.42000000000000004 4.8793438e2 1.0246622e2  -3.62%   2.90%
  2 0.42000000000000004                0.63 4.8325947e2 1.0148449e2  -3.58%   2.86%
  3                0.63  0.8400000000000001 4.7483794e2 9.9715968e1  -3.59%   2.91%
  4  0.8400000000000001                1.05 4.6722195e2 9.8116609e1  -3.57%   2.85%
  5                1.05                1.37 4.5417244e2 1.4533518e2  -3.59%   2.82%
  6                1.37                1.52 4.3888037e2 6.5832056e1  -3.58%   2.80%
  7                1.52                1.74 4.2558052e2 9.3627714e1  -3.61%   2.84%
  8                1.74                1.95 4.1008364e2 8.6117564e1  -3.58%   2.79%
  9                1.95                2.18 3.9357751e2 9.0522827e1  -3.63%   2.81%
 10                2.18                 2.4 3.7119255e2 8.1662361e1  -3.60%   2.88%

[enocera]

@cschwan The cut in on the neutrino pseudo rapidity seems to be inconsequential; however I understand that I used different values of the physical parameters (the W mass and the W width, above all). Can you please confirm that the numbers reported above were produced with the values of the physical parameters reported in the run.sh script in the runcard folder? Thanks.

[cschwan]

Yes, they were! The relevant parameters should be mw, gf, and ww, all remaining parameters shouldn't play a role at NLO QCD. I'm not quite sure, but I doubt that different parameter yield a difference of 2-3%.

[cschwan]

I've just uploaded the corresponding runcards here: NNPDF/pinecards#96. Note that results above were generated using set req_acc_FO 0.0005.

[enocera]

@cschwan Thanks, this is very useful.

[cschwan]

The differences are pretty much the same size as the one I've observed here: NNPDF/pinecards#54 - not sure if this is helpful.

[enocera]

@cschwan Thanks.

[enocera]

@cschwan I've recomputed the applgrids now. For some reason MCFM was not using the Gmu scheme. I obtain a better agreement with the PineAPPL grids (you can check it yourself by pulling from the applgrids repository). Here I paste the relative difference between MCFM and madgraph. I'm still a fraction of percent below your results (and for some bins about 1%, which is not acceptable). I'll continue to investigate this issue (and possibly I'd also run more events with MCFM).

W-
MCFM [fb]		madgraph [pb]		rel diff [%]
485905		4.87E+02		-0.29800105807377
483543		4.88E+02		-0.899993970500706
479706		4.83E+02		-0.735313060704216
472798		4.75E+02		-0.4296076257091
464389		4.67E+02		-0.606339235560318
450585		4.54E+02		-0.789885004911338
435769		4.39E+02		-0.708933507324571
422863		4.26E+02		-0.638544264197059
405198		4.10E+02		-1.19137647139497
389580		3.94E+02		-1.01568557613974
369217		3.71E+02		-0.532217039377538
W+
MCFM [fb]		madgraph [pb]		rel diff [%]
619202		6.25E+02		-0.920156272986751
623055		6.27E+02		-0.649966129891112
626656		6.27E+02		-0.127120152721407
629020		6.33E+02		-0.668817121031048
634592		6.36E+02		-0.278712412548114
643054		6.46E+02		-0.421529974768021
642065		6.51E+02		-1.39260248168466
650923		6.57E+02		-0.88136483532742
652860		6.53E+02		0.028833787075919
648311		6.57E+02		-1.27997635266902
634839		6.36E+02		-0.152980834717935

[cschwan]

@enocera : OK, that's not great, but at least it's better than before. What are your MC uncertainties? For the result that I've posted above they are 0.1-0.2% for WP and 0.2-0.3% for WM. To make sure we can trust these uncertainties I'm rerunning the grids (started it already yesterday) with our default precision of 0.0001. The results should be finished later today or early tomorrow.

[cschwan]

@enocera Which version of MCFM are you using? According to the latest manual the Gmu scheme (ewscheme=+1) should be the default and correct one (see p. 12, sec. 4.1 and tab. 2). But then, I've recently learned to distrust the MCFM manual ...

[enocera]

Well, the scheme was set to +2 instead of +1. I'm using v6.8 (which is clearly outdated, but that's the default used by NNPDF for use in conjunction with applgrid and the mcfm-bridge, and also I think that it's perfectly fine for W production).

[cschwan]

Here are the updated numbers with higher statistics:

 $ pineappl convolute ATLAS_WMU_8TEV_WP-20210324201438/ATLAS_WMU_8TEV_WP.pineappl.lz4 NNPDF31_nlo_as_0118
bin                   x1                       diff        integ    neg unc pos unc
---+-------------------+-------------------+-----------+-----------+-------+-------
  0                   0 0.21000000000000002 6.2424667e2 1.3109180e2  -3.24%   2.82%
  1 0.21000000000000002 0.42000000000000004 6.2572968e2 1.3140323e2  -3.24%   2.82%
  2 0.42000000000000004                0.63 6.2874064e2 1.3203553e2  -3.26%   2.82%
  3                0.63  0.8400000000000001 6.3332656e2 1.3299858e2  -3.28%   2.83%
  4  0.8400000000000001                1.05 6.3773634e2 1.3392463e2  -3.30%   2.84%
  5                1.05                1.37 6.4456335e2 2.0626027e2  -3.35%   2.86%
  6                1.37                1.52 6.5127718e2 9.7691578e1  -3.42%   2.89%
  7                1.52                1.74 6.5435656e2 1.4395844e2  -3.49%   2.93%
  8                1.74                1.95 6.5649824e2 1.3786463e2  -3.58%   2.96%
  9                1.95                2.18 6.5221944e2 1.5001047e2  -3.70%   2.99%
 10                2.18                 2.4 6.3771574e2 1.4029746e2  -3.82%   2.99%

 $ pineappl convolute ATLAS_WMU_8TEV_WM-20210324201441/ATLAS_WMU_8TEV_WM.pineappl.lz4 NNPDF31_nlo_as_0118
bin                   x1                       diff        integ    neg unc pos unc
---+-------------------+-------------------+-----------+-----------+-------+-------
  0                   0 0.21000000000000002 4.8911139e2 1.0271339e2  -3.57%   2.88%
  1 0.21000000000000002 0.42000000000000004 4.8690335e2 1.0224970e2  -3.57%   2.88%
  2 0.42000000000000004                0.63 4.8249643e2 1.0132425e2  -3.57%   2.87%
  3                0.63  0.8400000000000001 4.7550499e2 9.9856048e1  -3.57%   2.86%
  4  0.8400000000000001                1.05 4.6748535e2 9.8171923e1  -3.58%   2.85%
  5                1.05                1.37 4.5385642e2 1.4523405e2  -3.58%   2.82%
  6                1.37                1.52 4.3886350e2 6.5829525e1  -3.58%   2.81%
  7                1.52                1.74 4.2631268e2 9.3788790e1  -3.59%   2.80%
  8                1.74                1.95 4.0998254e2 8.6096333e1  -3.59%   2.80%
  9                1.95                2.18 3.9210463e2 9.0184065e1  -3.59%   2.81%
 10                2.18                 2.4 3.7270186e2 8.1994410e1  -3.57%   2.81%

And the corresponding differences in per mille for WP:

8.147051  4.292847  3.326610  6.846460  4.954900  2.347159 14.347737
5.274910  5.572772  6.028650  4.531448

and WM:

6.598800  6.949434  5.816959  5.725468  6.667578  7.260384  7.101239
8.157914 11.807906  6.480389  9.438514

So overall this doesn't change the picture.

[enocera]

Thanks. It seems that there's a systematic difference between madgraph and MCFM of about 6 per mille - let's see whether the upcoming runs will solve the issue. I've also run DYNNLO (which I need to compute the NNLO corrections) and I'd say that here the agreement is generally better. For instance, for W- I get

bin	DYNNLO	madgraph	rel diff [%]
0.105	488442	4.8911139	-0.137045954279107
0.315	487172	4.8690335	0.05514479485684
0.525	482470	4.8249643	-0.005478060812058
0.735	476581	4.7550499	0.225776940331248
0.945	467510	4.6748535	0.005272614489522
1.21	453139	4.5385642	-0.158322280801252
1.445	438408	4.388635	-0.103898651484462
1.63	425172	4.2631268	-0.268286716905157
1.845	409272	4.0998254	-0.173610703883979
2.065	391389	3.9210463	-0.182843667042253
2.29	371571	3.7270186	-0.304345602859208

[cschwan]

The DYNNLO numbers look quite well. I did a little big of digging, and MCFM 6.8 (and DYNNLO? It seems to have copied a few of the MCFM sources?) actually uses a different Gmu scheme than Madgraph5. The latter uses the complex masses and a modulus to make alpha real (see arXiv:1804.10017, eqs. (5.50) and (5.7) vs. src/Need/coupling.f:85 in MCFM or the LUSIFER paper, hep-ph/0206070, eqs. (6.2) and (6.1)). This accounts for roughly half a per mille in alpha or a global per mille (from the alpha^2) on the (NLO) cross section.

[enocera]

@cschwan I think that I'm content with the latest run of MCFM, which seems to reasonably agree with mcfm:

bin (W-)	MCFM [fb]	Madgraph(*10^5) [fb]	rel. diff. [%]
0.105	490078	4.8911139	0.197235950195692
0.315	487435	4.8690335	0.109070953050143
0.525	483082	4.8249643	0.121215445824946
0.735	476783	4.7550499	0.268048567167887
0.945	467278	4.6748535	-0.044374012900251
1.21	453585	4.5385642	-0.059838839467792
1.445	438751	4.388635	-0.025640967200075
1.63	426160	4.2631268	-0.035826919466865
1.845	408793	4.0998254	-0.290988348626331
2.065	391773	3.9210463	-0.084648508192245
2.29	371437	3.7270186	-0.340531503323575
bin (W^+)
0.105	623935	6.2424667	-0.04995231875114
0.315	626328	6.2572968	0.095528221634681
0.525	629593	6.2874064	0.135382699617052
0.735	632501	6.3332656	-0.130523113797439
0.945	638921	6.3773634	0.18541572432273
1.21	645367	6.4456335	0.124526044870597
1.445	651381	6.5127718	0.015938444627637
1.63	654041	6.5435656	-0.048247739820601
1.845	657757	6.5649824	0.191371585555153
2.065	652661	6.5221944	0.067655337150518
2.29	638238	6.3771574	0.081828408838084

I will rerun MCFM with the values of the parameters consistent with the NNPDF4.0 theories and perhaps with some more statistics. Thanks.

[enocera]

Incidentally, I've noticed that the value of the Z mass is set to 91.176 [GeV] in the run.sh file. I guess that there's a typo here and you meant 91.1876 [GeV]?

[cschwan]

I'm not sure why I've chosen 91.176. Why do you say it's a typo?

[enocera]

Because 91.176 is very similar to 91.1876, which is the PDG value?

[cschwan]

Oh, I see - that's probably what happened.