Implementation of ATLAS_WPWM_13TEV_HMT

[ecole41]

I have produced grids using NNLOJET for this dataset, but I have some questions regarding the match of the grids and the commondata. For the $m_T$ single differential, the $W^-$ results have a good match with the data:

Exp chi2: 0.6394
Exp+th chi2: 0.4909
Exp+th+pdf chi2: 0.4687

However, the $W^+$ results are showing more of a disagreement:

Exp chi2: 5.391
Exp+th chi2: 2.379
Exp+th+pdf chi2: 1.976

Altogether giving these $\chi^2$ values:

Exp chi2: 3.383
Exp+th chi2: 2.085
Exp+th+pdf chi2: 1.464

After discussing with @scarlehoff, we were wondering if this might be caused by cuts or if this might be due to EW corrections not being accounted for. If this is the case, should we apply a cut in the high-mass tail?

I was also wondering if this might be caused by a bug in the implementation of uncertainties as the uncertainties for $W^+$ are much lower than $W^-$ but I can double check this.

[ecole41]

After having a look at the commondata implementation, I believe that the uncertainties are implemented correctly and agree 'by eye' with the uncertainties shown in the figures in the paper.

[enocera]

Thanks @ecole41 for your results. My thoughts are as follows.

1. It seems to me that there are Hepdata Tables that report theoretical predictions computed with DYTurbo and NNPDF40_nnlo_as_01180, see e.g. lep_physical_plus etc. In principle a benchmark of these numbers is possible. If you want, i can liaise with the experimentalists and ask them to provide us with the DYTurbo runcards that they have used to generate those predictions.
2. If I compare your results with Fig. 17 in [arXiv:2502.21088] I seem to note the same trend as in your data/theory comparison plots, with W+ being farther from the data than W-. Not by the same amount, though: if I take the highest transverse mass bins the data-theory discrepancy is between 10 and 20% in Fig. 17, whereas it is twice that in your plots.
3. Fig. 17 incorporates EW corrections, whereas your plots do not. I am afraid that, given the precision of this measurement, one has to compute EW corrections and either apply them as K-factors or apply a cut to the data. In [arXiv:2502.21088] they say that EW corrections can be as large as -10% at transverse mass values of 1 TeV at the level of simulated event samples. This seems more or less what would bring your predictions in line with those of Fig. 17. They also say that EW corrections can be as large as -3% at lower values of the transverse mass. I'm afraid that, with a measurement with a sub-percent experimental precision, you cannot ignore EW corrections anywhere. In other words, if you were to ignore EW corrections, you would cut all the data points. I understand that EW corrections are included in the DYTurbo predictions, so again I'd say that the most conclusive check is to get the runcards that the experimentalists used to generate their predictions.

[ecole41]

Thanks for your comment @enocera. That would be great if you were able to get the runcards from the experimentalists. Would you also suggest producing EW K-factors? If so, do you know how I am able to this as I haven't made these before?

[enocera]

Dear @ecole41 let me make contact with the experimentalists, then. I will ask them to provide us with EWK K-factors as well (they must have them). If you wanted to compute EWK K-factors yourself, you should run madgraph, enabling EW corrections. The computation is NLO only. So the idea is that you run madgraph twice: once at NLO QCD (and you make sure that you reproduce the NLO that you get from NNLOjet) and once at NLO QCD+QED. Then you just take the ratio of the two cross sections, bin-by-bin. No need to produce interpolation grids, though this is also possible, in principle. There are examples of the required files at https://github.com/NNPDF/pinecards, see e.g. https://github.com/NNPDF/pinecards/tree/master/ATLAS_DY_8TEV_2D_0116_0200. Happy to assist you further.