CMS_Z0_13TEV

[evagroenendijk]

Implementation of the Drell-Yan forward-backward asymmetry at high dilepton masses in proton-proton collisions at 13 TeV.

[evagroenendijk]

Hi @enocera a question came up

The observables that are available are A_{FB} for dimuons and dielectrons, and \Delta A_{FB} for dimuons and dielectrons. Should these be in separate datasets or in the same directory implemented as different observables of the same process? 

At the moment I have made two directories, CMS_DY_13TEV_DIMUONS and CMS_DY_13TEV_DIELECTRONS, both containing the two observables (A_FB and \Delta A_FB), because from past datasets it seems like it was also done like this.

Is this alright or should it be organised differently? Thanks!

[enocera]

Thanks @evagroenendijk . We have the possibility of defining two different data sets, as you propose and did. Or also to define a single data set with two different distributions (one for electrons and one for muons). We don't have strong reasons to prefer one than the other, given that these are essentially uncorrelated. So I suggest that you proceed with the implementation that you've already started.

[scarlehoff]

Two questions/comments

1. Is this pure charged current, neutral current or both? If both we need to separate W and Z (and if only one of the two, it should be either Z0 or WPWM). This is necessary for the theory covmat.

2. If the paper is the same, the rawdata is the same and the definition of the bins is the same, I think it is better to have them in one single folder with two observables like in LHCB Z0 13 TeV https://github.com/NNPDF/nnpdf/tree/master/nnpdf_data/nnpdf_data/commondata/LHCB_Z0_13TEV (in that one the kinematics are close but not the same tbf)

[enocera]

@scarlehoff This is the Z forward backward asymmetry, so NC only.

[evagroenendijk]

Two questions/comments

1. Is this pure charged current, neutral current or both? If both we need to separate W and Z (and if only one of the two, it should be either Z0 or WPWM). This is necessary for the theory covmat.
2. If the paper is the same, the rawdata is the same and the definition of the bins is the same, I think it is better to have them in one single folder with two observables like in LHCB Z0 13 TeV https://github.com/NNPDF/nnpdf/tree/master/nnpdf_data/nnpdf_data/commondata/LHCB_Z0_13TEV (in that one the kinematics are close but not the same tbf)

Thanks @scarlehoff and @enocera

1. It's only neutral current (Z or virtual gamma)
2. The mass bins for all observables are the same, so then I can put them in one single dataset

[enocera]

@evagroenendijk Incidentally, I would implement only AFB (electrons, muons), not Delta AFB.

[evagroenendijk]

@evagroenendijk Incidentally, I would implement only AFB (electrons, muons), not Delta AFB.

Ok! I thought because the uncertainties are different (the pdf uncertainty cancels out mostly for \Delta A_FB), I thought it would maybe be useful? But I can take only the A_FB!

[enocera]

The thing is that, in the SM, at parton level, Delta AFB should be zero. So this observable is better suited to study new physics than constraints on PDFs.

[evagroenendijk]

The thing is that, in the SM, at parton level, Delta AFB should be zero. So this observable is better suited to study new physics than constraints on PDFs.

Alright, clear! Thanks

And shall I put them into one dataset as Juan suggested then?

[enocera]

And shall I put them into one dataset as Juan suggested then?

Yes, that would be desirable in light of the remark made by @scarlehoff . Also, who am I to contradict @scarlehoff ?

[scarlehoff]

And please change DY to Z0!

[enocera]

Dear @evagroenendijk in view of the upcoming Morimondo meeting, I have reviewed this PR and implemented the following alterations.

• I have removed the separate electron and muon channels, and I have retained only the combined measurement, the reason being that the breakdown of systematic uncertainties is available only for this.
• I have fixed some inaccuracies in metadata.yaml, in particular in regards to the name of the observable.
• I have removed the last bin: this is indeed the integral over the invariant mass range, but we are interested in the measurement differential in the various invariant mass bins.
• I have implemented two variants of the uncertainties, one in which we only have the statistical and the systematic uncertainty, as per Hepdata, and one in which we have the breakdown of systematic uncertainties from Table 1 of the paper (under the assumption that it applies to all invariant mass bins). I have slightly modified the uncertainty treatment in this case - I agree with you that it is highly ambiguous to determine which uncertainty is correlated and which is not.

[enocera]

Tests now pass, so this PR is ready for review.

[evagroenendijk]

Dear @evagroenendijk in view of the upcoming Morimondo meeting, I have reviewed this PR and implemented the following alterations.

• I have removed the separate electron and muon channels, and I have retained only the combined measurement, the reason being that the breakdown of systematic uncertainties is available only for this.
• I have fixed some inaccuracies in metadata.yaml, in particular in regards to the name of the observable.
• I have removed the last bin: this is indeed the integral over the invariant mass range, but we are interested in the measurement differential in the various invariant mass bins.
• I have implemented two variants of the uncertainties, one in which we only have the statistical and the systematic uncertainty, as per Hepdata, and one in which we have the breakdown of systematic uncertainties from Table 1 of the paper (under the assumption that it applies to all invariant mass bins). I have slightly modified the uncertainty treatment in this case - I agree with you that it is highly ambiguous to determine which uncertainty is correlated and which is not.

Dear @enocera Thank you! I have been on holiday for the past few weeks, and I was planning to dot the i's on this pull request in these days before the Morimondo meeting (indeed I had not finished with the metadata.yaml and the uncertainties). Thanks a lot for finishing it, I completely agree with the changes!

[scarlehoff]

LGTM

Let's wait for the grids so that we have a proper comparison before merging.

[enocera]

Dear @jacoterh I have (finally!) updated the implementation of the CMS AFB in this PR. The input data is now from https://arxiv.org/pdf/2408.07622, as agreed. The input data is double differential in mll and yll. The data is for A4, which is related to AFB with a 3/8 rescaling factor: AFB=3/8 A4. Can you please proceed with the computation of the grids? Thanks.

note to @enocera correlations still need to be carefully implemented.

[jacoterh]

Dear @enocera (cc @scarlehoff ), thanks again for the commondata implementation. I have obtained a satisfactory data theory comparison with NNLOJET. However, we need to keep in mind that A4 is very sensitive to the EW input scheme around the Z pole. In particular, at the Z pole, A4 is proportional to

$$A_4 \sim \frac{2g_V g_A}{g_V^2+g_A^2} \, ,$$

with $$g_V$$ and $$g_A$$ the vector and axial couplings, respectively. Indeed, $$A_4$$ is extremely sensitive to our value of $$\sin^2\theta$$. I produced two comparisons:

1. Fixing $$\sin^2\theta$$ to the best-fit value in the experimental paper, i.e. $$\sin^2\theta_{\rm eff}= 0.23154$$, and running in the $$\alpha_0$$ scheme rather than $$G_\mu$$ to avoid tension between the EW input parameters. This produced the following plot below.

As you can see, we find excellent agreement. Now, scenario 2) is as follows

2. Let $$\sin^2\theta$$ be determined by the NNPDF4.1 settings, which corresponds to $$\sin^2\theta =0.22321 $$. This results in the plot below, where the agreement gets worse especially around $$M_{\ell\ell}$$ bins around the Z pole.

In NNPDF4.1 we will have to adopt option 2). This is going to generate (artificial) large pulls around the Z pole though given that we keep the $$W,Z$$ masses fixed.

Finally, a technical comment. Rather than generating Z + 0J, I had to generate Z + J and be inclusive in the jets in order to allow for non-zero Z pT. NNLOJET supports reweighting with the fac option (see below) and I used this to generate two histograms, one unweighted, one weighted by $$4 \cos\theta_{CS}$$. The effect of the latter is to project out $$A_4$$:

mll > mll_bin1 [54, 66, 76, 82, 86, 89.5, 92.7, 96, 100, 106, 116, 150] grid=mll_bin1.pine
mll > mll_bin1_proj [54, 66, 76, 82, 86, 89.5, 92.7, 96, 100, 106, 116, 150] fac = proj_A4 grid=mll_bin1_proj.pine

so that $$A_4$$ is given by the ratio of the grids mll_bin1_proj /mll_bin1. Similarly for the other rapidity bins. Generating only Z results in an empty mll_bin1_proj grid because proj_A4 is ill-defined at zero pT.

[scarlehoff]

In NNPDF4.1 we will have to adopt option 2)

By eye it looks like the few points at the end are driving the difference.
For a fit we can opt for option 2) and add a theory error for the uncertainty that we know we have which will deweight the more problematic points.

[enocera]

By eye it looks like the few points at the end are driving the difference. For a fit we can opt for option 2) and add a theory error for the uncertainty that we know we have which will deweight the more problematic points.

Why not a cut?

[enocera]

Dear @enocera (cc @scarlehoff ), thanks again for the commondata implementation. I have obtained a satisfactory data theory comparison with NNLOJET. However, we need to keep in mind that A4 is very sensitive to the EW input scheme around the Z pole. In particular, at the Z pole, A4 is proportional to
A 4 ∼ 2 g V g A g V 2 + g A 2 ,

with g V and g A the vector and axial couplings, respectively. Indeed, A 4 is extremely sensitive to our value of sin 2 ⁡ θ . I produced two comparisons:

1. Fixing 
     
       sin
       2
     
     ⁡
     θ
    to the best-fit value in the experimental paper, i.e. 
     
       sin
       2
     
     ⁡
     
       θ
       
         e
         f
         f
       
     
     =
     0.23154
   , and running in the 
     
       α
       0
     
    scheme rather than 
     
       G
       μ
     
    to avoid tension between the EW input parameters. This produced the following plot below.

As you can see, we find excellent agreement. Now, scenario 2) is as follows

2. Let 
     
       sin
       2
     
     ⁡
     θ
    be determined by the NNPDF4.1 settings, which corresponds to 
     
       sin
       2
     
     ⁡
     θ
     =
     0.22321
   .  This results in the plot below, where the agreement gets worse especially around 
     
       M
       
         ℓ
         ℓ
       
     
    bins around the Z pole.

In NNPDF4.1 we will have to adopt option 2). This is going to generate (artificial) large pulls around the Z pole though given that we keep the W , Z masses fixed.

Finally, a technical comment. Rather than generating Z + 0J, I had to generate Z + J and be inclusive in the jets in order to allow for non-zero Z pT. NNLOJET supports reweighting with the fac option (see below) and I used this to generate two histograms, one unweighted, one weighted by 4 cos ⁡ θ C S . The effect of the latter is to project out A 4 :

mll > mll_bin1 [54, 66, 76, 82, 86, 89.5, 92.7, 96, 100, 106, 116, 150] grid=mll_bin1.pine
mll > mll_bin1_proj [54, 66, 76, 82, 86, 89.5, 92.7, 96, 100, 106, 116, 150] fac = proj_A4 grid=mll_bin1_proj.pine

so that A 4 is given by the ratio of the grids mll_bin1_proj /mll_bin1. Similarly for the other rapidity bins. Generating only Z results in an empty mll_bin1_proj grid because proj_A4 is ill-defined at zero pT.

Amazing, thanks @jacoterh .

[scarlehoff]

Why not a cut?

Because I don't think there's anything wrong with those datapoints or theory and we are still biasing the result for the points that pass the cut.