Measurement of the differential cross section of W/Z bosons in the full hadronic channel in the high transverse momentum in proton-proton collisions at 13 TeV

Contact person: Jindrich Lidrych (jindrich.lidrych@desyNOSPAMPLEASE.de)

Conveners in normal font (always in black), authors in italic font with colour code:

• red = to be discussed
• green = done
• blue = work in progress

Comments received about AN v2

Guillelmo

• Section 14 is missing, I think it's crucial to complete it
• Fig. 8 right. Do you know the shape of the ratio at low values with some minimum at ~400GeV?
  • See this figure. Maybe one comment, these corrections are provided by JMAR group.
• Section 4.4.4. The tracking scale factors are not 1. It's quasi-irrelevant for you, but they should be ~0.99.
  • Good to know. Thank you for info.
• Table 12. Why are the ttbar cross sections not consistent among the samples? in fact, you should use the very same values (or the sum of the three decay samples should sum the inclusive one)
  • These numbers are calculated by GenXsecAnalyzer tool. I am not sure whether different numbers matter or not. If we apply some overall normalisation factor N to MC sample, then this factor will cancel out when we calculate the tagging efficiency (epsilon = N*pass/(N*pass+N*fail)) for MC sample, right?
• Section 6.4. If I understand well, you are making an inclusive (pt>200GeV) analysis, see Table 13. Is there a uncertainty due to a possible shape variation with Pt of the scale factors?
  • Yes, it is inclusive, but pt > 300 GeV. So far, we don't have uncertainty due to a possible shape variation with pt.
• Figure 48. There are two important nuisances pulled by ~2sigma, any comment?
• Results in Section 12 and 13. You quote the binned signal strengths in Section 12, Fig. 51. Nevertheless, I miss that in Section 13. You show the cross section results in Fig. 55, can you provide the signal strengths? The should be very similar w.r.t. the values reported in Section 12.
• To clarify, are you using at all events with pt>1200GeV?
  • Yes, but above pt>1200 GeV we have lack of stats, so we perform fits for 500 < pT < 1200 GeV.

Vieri

• there are many minor grammar problems spread out in the paper (like singular/plural, commas and similar) that I guess will be scrutinized by the LE in proper time, still I would suggest a further reading to take care of it!
  • Work in progress, I will try my best to fix all typos .
• impacts of PDFs and tunes: I think this is something that it is not really possible to make, but it would be very interesting to see how the non-production (non-CP5) tunes may change your signal. Some hight pt dependence of the tunes in leptonic V+Jets has been shown in GEN-17-001 for example. I think this could be done in a latter stage using RIVET (I see that you already have a final empty chapter for this in the end!).
  • We like your idea and we are planning to do this check. We have already first version of rivet plugin and we are working on the validation of plugin. So far, we observe reasonable agreement(within1-2%) within results from rivet plugin and CMSSW. See this figure. .
• Maybe more discussion on the pulls is needed in the cases where the relative points are really off (especially on the ttbarSF and JMRstat and Vtagstat). Is there any idea or study to check it? We discussed it also during one of the last meetings. I am sure this will be one of the main issues potentially during the approval/ARC review.
  • Work in progress. I will add more discussion on the pulls in AN. The jet mass resolution(JMR) is a bit tricky, I guess. Since we measure the JMR in the semileptonic sample of ttbar, we have statistics for jets with 200 < pT <400 GeV. But the actual analysis starts at pt > 500 GeV ... As shown in ATLAS-CONF-2020-022, in W/Z- mass peak in inclusive W/Z+jet topology we can be sensitive to the jet mass response...
  • In case of ttbarSF, I am tryed to investigate whether we can include ttbarSF as a rateParam into the fit. I constructed simple datacard for single-muon+jet topology(ala bin counting experiment) and now ttbarSF and ttbarnor are treated as a rateParam. As shown here (example for 2017), now only JMRstat is pulled more than one sigma in the fit.
• Are scale/alphaS uncertainties missing? It looks like I see only see PDFs uncertainty
  • We added these plots in AN, see Appendix L. Since our MC is just LO, we are not sure that scale uncertainties (muF, muR) give a realistic estimation of uncertainties. We are planning to add the theory uncertainties on MC predictions, when we will compare unfolded data with MC .
• Final result of xsec by year: it looks like the very high pT shows a disagreement which is consistent through the years, it is true that the stat is poor but this looks in fact independent of time. Do you have any idea about what could affect this last bins?
  • This is good question and we don't have answer. But as you say, the stat is poor and I believe that also uncertainties related to NLO QCD and EW corrections are not small at TeV scale .

Juan Pablo

-- START (NO NEED TO REPLY)--
General comment NOT to be considered by just mentioned for future analysis that might want to start from scratch ( that is not your case ) :
Future analysis could try to search only for W->cs by using charm tagging on the AK8 jet.
As you can see in table 1, the W -> cs decay channel amounts for half of the total W->qq BR while the impact on the reduction on the QCD background when only searching for charm should be stronger.
So, in more detail, for the next iteration of this analysis I would use the help of one of deepCSV ctagger or deepjet ctagger to eliminate the QCD (that is mainly light). I would just plot the discriminants CvsL and CvsB defined in this way :

deep = "pfDeepCSVJetTags:probb" + "pfDeepCSVJetTags:probbb"
CvsB ="pfDeepCSVJetTags:probc"/(deep+"pfDeepCSVJetTags:probc");
if "pfDeepCSVJetTags:probc" is -1. or deep is -2. then set CvsB to -1.;

CvsL = "pfDeepCSVJetTags:probc"/("DeepCSVJetTags:probc"+"pfDeepCSVJetTags:probudsg")
same as before : if ( jet->getBTaggerDiscriminant("pfDeepCSVJetTags:probc") = -1. || Jet->getBTaggerDiscriminant("pfDeepCSVJetTags:probudsg") = -1. ) CvsL = -1.;

Again, the tagging info. from AK8 (this would require running DeepDoubleX, etc) is not even the way to go. What is needed is to take into account the tagging info. for the subjets on your merged jet.
I might be wrong in this whole initial idea to start with and ploting the CvsB and CvsL as I explained above for the subjets would be great. Most probably (for your signal) you will have a charm like subjet (and the future analysis could use a deepCSV ctagger to enhance the signal over the background).

Whenever I say AK4 jets ( AK4PFchs jets ) I also mean the subjets within the AK8 jet. In principle there is no difference and one can use the b,c - algorightms in the same manner in AK4PFchs jets and subjets (not the same SFs though). Nothing really special is needed , I mean, just doing the same you do for the AK4PFchs jets in the subjets (I mean similar procedure, path, etc).

-.-.-.-.-.-.-.-.
More fun. If you have muons (iso or non iso) one could also look for a non-iso, pt<25 GeV soft-muon in your AK8 and a non-iso, pt<25 GeV soft-muon in your AK4 jet ( you said there is a recoiling jet on your event topology ) or a Secondary vertex in your AK4 jet. If you set the weight of the events with the opposite sign (for the electric charge) of those muons to 1 and set the weight of the events with the same electric charge to -1 then you should be able to see ( W->cs + c final state, a part of your signal ... low in stats but with much less background )

For the secondary vertex on the AK4 jet case you will have a bit more stats. but then you need to know the sum of the charges of the tracks that from that secondary vertex to be able to do the OS-SS subtraction above.

-- END (NO NEED TO REPLY)----

• Your major contribution is QCD , no doubt about it. Would it be worth trying a MET > X and an alignment between the MET and AK8jet ( acoplanarity or something like it, i.e. acos(-cos(AK8phi-met.phi) ) to reduce it a little this contribution? [I see you are using this idea for determining the W-tagging efficiency with the ttbar control sample]. May you have already did take a look at these variables to see if they are any different (in the two cases).
  • Work in progress. .
• About the use of top pT MC reweighting
  I understand a one-dimensional reweighting (top pT) may not be enough (multifolded mismodeling issue) but sufficient until a more accurate and stable treatment is provided but I do not necessarily have to think this improves other distributions as well. Did you check whether the description of the data by the MC in other distributions (particularly those the analysis is sensitive to) is not spoiled ?
  So, I would like to see the Data/MC plots of Figs 10-13 and also 26 before the top pT MC reweighting (because the ones you have there are after)
  However, the uncertainty you are assingning for this is done by comparing results obtained
  with and without the top pT reweighting and those not seem to be the dominant though.
  • Done, these plots are in Appendix K. As shown in the Data/MC plots of Figs 10-13, without any cuts we are dominated by QCD, so it is quite hard to make conclusions. But Figs 26 vs 157 show that top pT MC reweighting gives some improvement. Maybe I can just add, that TOP-18-013(measurement with boosted top) shows that there is a really some mismodelling in the high pT top region. .

• The jet four-vector is corrected by the jet energy corrections supplied by the JetMET POG.
  Are these corrections dependent on the type of AKX jet?

  • Yes. JME group supports and provides JECs for AK4PFchs, AK4PFPuppi, AK8PFchs and AK8PFPuppi.
• 370 -> satified
  • Fixed .

• The presence of loose muon or electron in events works as event veto." Could you motivate this cut earlier ? Either I missed it or I found the motivation too late for my taste( in 8.1.8 )

  • Fixed. See section 4.6.1(summary of physics object selection) .

• Fig 12. Is there any systematics associated to the poor modelling of tau_21 or N_2^1 variables on this figure ? Are you somehow taking care of this mismodeling by constructing your transfer function(m_SD,pT) instead of using a constant when subtracting the QCD background with the control region ?

  • We are not using a constant transfer function. Actually, the mismodelling of jet pT, m_SD and N_2^1 leads to the non-constant/flat transfer function.

• Your V-tagging consist on selection those jets with N21,DDT < 0 (being N21 the variable defined in eq 17 , using the 2,3-points energy correlation functions [16-17]) as W/Z candidates but this tagging efficiency is computed for W-tagging only. You claim no significant difference is expected for W and Z-tagging efficiency, since both bosons create jets with very similar properties. I guess you have compared the jet properties ( which ones ? ) and seen this behaviour that you comment.

  • I believe this is quite standard approach. Jet properties for jets coming from W and Z boson decay were compared in JME-18-002. See these figures 1 2 . If needed, I can try to reproduce these plots.
• How the data/MC validation-comparison in section 7 looks like when there is no pt-reweighting ? In parituclar do you see any change in the data/MC in terms of overall-scale-normalization when going from 2016 to 2017-2018? Same for #MC(ttbar), so do you see any change in the predicted contribution from ttbar on 2016 vs 2017-2018 ?[The errors are very big to see a trend on the SF for ttbar from table 15 (when I compare those on 2016 vs those on 2017-2018) on the pass region but I see it on the fail region]
  • I believe that part of your question was answered in few lines above. You are right, the stat. errors are very large and it is very hard to make any conclusions. As shown in Figure 20, the older UE tune (CUETP8M1) was doing better job. Slightly different trend in the fail region for 2016 data can be maybe explanaited by different UE tune and jet substructure modelling by this tune .

• 8.1.9 when you write the name of the PDF used (NNPDF3.1), for which year do you mean, all, only 2017 and 2018 or what ?

  • All three years .

• In order to perform the likelihood based unfolding, the signal MC templates need to be splitted
  into sub-templates according the generator level distributions that include information about the migration effects from one bin to another bin and the background/fakes.
  You decided to fix the fakes/background to the MC predictions with corresponding uncertainties in order to make the final fit more stable. What was happening before doing the fix ? I mean to what extend was the fit "mis-behaving"/unstable ?

  • As shown in Figure 53 (AN v2), the fake contribution in the first pT bin also create a peak structure and this is quite hard to model for the transfer function .

• It is nice to see through the test in 10.1.5 that there is no significant bias on the extracted parameters from the fits due to the background subtraction method

  • Thank you .

• In fig 51 , the signal strength tells us we are consistent with MC predictions in all bins but the last two pT bins (in the 900-1200 GeV range). Even if, temporarily of course, you are missing in this version the comparison of the unfolded and combined results with the MC prediction and conclusions, shouldn't Fig 51 give us an idea of what is going to happen ?

  * This is good question and we don't have answer. But as you say, the stat is poor and I believe that also uncertainties related to NLO QCD and EW corrections are not small at TeV scale .
• Another doubt regarding figure 51. Are you cutting on 1.2 TeV or do you include the overflow bin as well ? It would be nice to include an overflow bin if this is not done yet.
  • Yes, we are cutting on 1.2 TeV. In the past, we tried to go to the higher pT, (one extra bin 1.2-1.5TeV), we ran out of the statistics ...

-- JindrichLidrych - 2021-01-26