Search for new physics in final states with a soft hadronic tau, a high-transverse-momentum ISR jet, and missing transverse momentum.

Analysis Review TWiki

Latest Version of Analysis Note

AN-20-188_v5.pdf

Current Iteration of Questions

AN-SECTION 3

1. Have you presented your trigger efficiency measurement, together with its uncertainty, to the JetMET POG to get their endorsement?

• ISR+tau team:

The trigger we use is the same used by the invisible Higgs, and we are seeing similar uncertainties and a similar turn-on curve in our studies.

(CMS AN -2021/094 Fig. 2)(https://journals.aps.org/prd/pdf/10.1103/PhysRevD.105.092007)

( https://cms.cern.ch/iCMS/jsp/db_notes/noteInfo.jsp?cmsnoteid=CMS%20AN-2019/257 Fig. 1 and 2)

• Convener follow-up:

Cecile: Please check with our contact Changgi.

• ISR+tau team:

Our trigger studies have been reviewed by Changgi, and everything looks good to him. He has officially reviewed it and given us his green light.

AN-SECTION 4

2. You don’t provide details in the AN regarding the signal scans performed and the parameter variations chosen in such a scan. Therefore, it is hard to evaluate if the chosen mass points for optimization (Fig 5, 6, and 7) are sufficiently representative of the signal topology, and if they fall/comprise the corner/edges of that phase space. Please add a section covering the signal generation and explain the choice of signal points to perform these optimization studies.

• ISR+tau team:

Thanks for your comments and suggestions. As requested, we have added more details regarding the signal scans to section 4 in the AN. To summarize, the previous iteration of this analysis (SUS-19-002) had an upper limit of 290 GeV on the chargino1/n2 masses for a delta-m = 50 GeV benchmark. For this reason, we have chosen to optimize the analysis near that boundary. In Figures 5-7, we specifically have chosen chargino1 masses of 270, 360, and 450 to be at the edges of sensitivity. Of course, the optimal selections vary depending on the delta-m values as well, but we cannot derive an optimal set of cuts for each signal point.

• Convener follow-up:

Danyer: I can not see those details in the AN v3, please provide tables, figures, or anything that helps us understand the reasoning to choose such a mass scan.

• ISR+tau team:

We have added a new section (4.1) in the AN to expand on the model choices and parameter scans. Thanks for the comment

3. What is the criterion to select the binning in the final discriminant variable (MT)? From this figure it seems rather arbitrary and subject to potential issues due to the non-smooth transition across bin.

• ISR+tau team:

Thanks for the comment. The binning strategy has been studied previously, and the binning had been determined through the optimization process, and picking the strategy that gave best signal significance. We have gone ahead and studied different binning scenarios under the conditions of our analysis. We are happy to add these studies to the AN if necessary.

• Convener follow-up:

Cecile: Please add the studies to the AN.

• ISR+tau team:

We presented these studies during the SUS 3rd generation meeting on 20/10/23, and have added details and plots of our optimization strategy to the AN Optimization section 4.2.

AN-SECTION 5

4. Please switch to NanoAODv9.

• ISR+tau team:

As we were asked to transition from pre-legacy to UL, the only available complete set of samples were NanoAODv8, and we were given the ok from the SUSY conveners at the time to continue with these. We had some concerns about being told to move to UL NanoAODv9 when they would become available, and we were specifically told by the conveners and physics coordination that we could proceed with v8. We would kindly request that we continue with v8, as moving to a new version of samples always comes with surprises and new results to be understood, which would significantly delay the analysis.

• SUS 3rd Gen meeting 20/10 notes:

This was discussed during the SUS 3rd generation meeting on 20/10/23, with the conclusion that using NanoAOD v8 would be ok, with the caveat that we reach out to the MC team and discuss the best way to handle getting v8 signal samples. We are currently in touch with Sara about this, testing some small runs of v8 test signal samples for new usage as well as testing some potential fixes to the issue the MC team is seeing in intermediate SUSY particle masses (bug observed in several other analyses). Sara believes we can converge this month.

5. Table 9: Please switch to the latest NNLO single top cross sections. Please add the references for all the cross sections. The diboson cross sections do not align with the latest higher-order values (eg 118.3 pb for WW), please correct. Why do you include only ZH->bb Higgs samples (H->tautau and H->WW should contribute more)? The EWKW samples are buggy, please check that their predictions are reasonable.

• ISR+tau team:

Thanks for the suggestion. We have updated to latest ST NNLO cross-section listed in the following Twiki:

https://twiki.cern.ch/twiki/bin/view/LHCPhysics/SingleTopNNLORef

We have also updated the diboson cross-sections to the higher order values. The new WW cross-section used is listed in https://twiki.cern.ch/twiki/bin/viewauth/CMS/StandardModelCrossSectionsat13TeV]]

For the inclusive WZ, ZZ, as well as the rest of the samples we use, we are attaching slides indicating the sources we are using. Namely, these are taken from analysis notes of relevant well known CMS analyses. We are adding more detailed references to these in the AN.

The rest of our cross-sections are the same ones used in EXO-19-015.

( https://cms.cern.ch/iCMS/jsp/db_notes/noteInfo.jsp?cmsnoteid=CMS%20AN-2019/082 )

The EWKW contribution in signal region overall is 3.8%, and <1% percent at MT < 100 GeV. This is expected.

• Convener Follow-up:

Cecile [regarding cross-sections]: It looks like the plots and results in the AN have not been changed after these modifications.

Cecile [regardin ewk samples]: But the MC sample is incorrect and probably cannot be used. Please check with MC experts.

• ISR+tau team:

The fact that changes weren't observed after cross-section updates was addressed during the SUS 3rg gen 20/10/23 meeting (fraction of a percent-level effect expected).

As for the EWK samples, we have reached out to the MC conveners and they are currently checking on this whether there are any issues the specific ones we use.

6. Can you provide a validation of the stitching procedure used here? (e.g. comparing the inclusive sample to the various HT-binned sub-samples)

• ISR+tau team:

You can see a comparison of the inclusive DY samples and those with the HT stitching procedure below. As expected, the low-HT area around 100 GeV (where the threshold of inclusive and HT-binned events at generator-level are stitched) shows excellent agreement. Further, the stitched distributions are smooth, showing appropriate stitching. Finally, as expected, the higher-HT regions show differences between the inclusive and stitched samples due to different levels of accuracy between samples in terms of the matrix element calculation for this phase space where momentum transfer is larger.

7. Gamma+Jet samples: Be aware that because PYTHIA still allows photons to be included in the PS for DY samples, there is some overlap in phase-space between these samples and standard DY, are you accounting for this with any sort of removal?

• ISR+tau team:

Gamma+Jet background is <0.5% in signal region, and effectively 0 in the high mT region, where we expect to see signal. Correspondingly, DY background is the order of ~3% of our signal region, where only 1.8% comprises our high MT region, therefore any overlapping would be negligible.

• Convener Follow-up:

Danyer: But you should not use MC samples with overlapping in phase-space (double counting of events) no matter how small their contribution is.

• ISR+tau team:

Thank you for the comment. We have followed your suggestion and are now removing the overlaps.

8. 2018 HEM failure: Did you check somehow that the procedure suggested there was suitable for this analysis as well? (this is an analysis-dependent effect.

• ISR+tau team:

We are following the JetMET group recommendations on this as they are for analyses with high MET and/or high pT jets. The effect in SR MC after applying the veto is about a 5% reduction in overall BG yields.

• Convener Follow-up:

Danyer: What about signal acceptance?

• ISR+tau team:

We are seeing a similar reduction in signal from applying the HEM veto: ~ 5% reduction in overall yields.

AN-SECTION 6

9. Why don't you use DeepJet?

• ISR+tau team:

Thanks for the comment. When we started the analysis, DeepJet SFs were not officially available for all years, and so we proceeded to move forward with DeepCSV since the analysis strategy had already been carried out. Since DeepJet is now available, we have performed studies with DeepJet and find that it only provides a < 3% improvement in the UL on signal cross section (for all signal points), and thus we have decided to keep the DeepCSV requirement so as to not have to re-do the background estimations and a large part of the analysis and cause further delay.

• Convener Follow-up:

Cecile: At this stage of the analysis it should not be very hard to switch to DeepJet, which is the latest recommendation. You anyway need to rerun to switch to NanoAODv9, update the cross sections, change the TAU ID scale factors, include 3 prongs, etc

• ISR+tau team:

As presented at the SUS 3rd gen meeting on 20/10/23, though DeepJet is recommended, they deemed DeepCSV also acceptable for our analysis. However, we mentioned that If we are required to rerun our background estimation with updated TauID SFs, we would be ok changing to DeepJet. We have now moved to DeepJet since the tau POG suggested we re-run our BG CRs with the new tau ID SFs.

10. Please switch to the new tau ID scale factors provided by the TAU POG.

• ISR+tau team:

We have contacted the tau POG, and they indeed recommend that we use the updated Tau ID scale factors. We are have implemented these into our analysis framework and are working on rerunning the control regions.

AN-SECTION 7

11. Why not use the W and DY pt binned samples? They have well calculated cross sections and EWK corrections.

• ISR+tau team:

We have twice the statistics in the HT-binned samples than the pT-binned samples where most of our SR lies (18.9 million vs 9.5 million). We also have finer binning in the HT-binned case with similarly high statistics in each bin, compared with the pT-binned samples which are divided into only two bins (100-200 GeV, and 200-inf GeV).

• Convener Follow-up:

Cecile: The MC statistical uncertainties do not seem to dominate the total uncertainty, and the pT binned samples would have smaller systematic uncertainties, so that it is not clear that HT-binned samples are the best choice. Could you try to switch to the pT-binned samples so that we can evaluate if statistics are really an issue?

• ISR+tau team:

We are not aware of any NLO HT-binned samples that exist in NanoAODv8. The errors in Table 19 are purely statistical. Thanks for the comment on the appropriate statistical errors, we have added a description of the corresponding errors in the systematics section.

• Convener Follow-up:

Danyer: You don’t necessarily need HT-binned samples, you can simply use inclusive ones, and they are plenty of those

• ISR+tau team:

We compare below the yield of W+Jets background MC in our signal region from the inclusive samples (left), and HT-binned samples (right). You can see that our MT distribution stats definitely benefit a great deal from using the HT-binned samples. Using the HT-binned samples, we are seeing MC statistical uncertainties ranging from 12-30% in the high-MT W+Jets BGs in SR. This is to be compared with 70-100% statistical uncertainty using the inclusive samples. Therefore, there is a clear advantage to using the HT-binned samples especially when you consider the systematic effects which are generally not as large at high MT. This supports our choice of continuing to use these HT-binned samples vs. the inclusive or pt-binned versions.

AN-SECTION 8

12. The recommendation for correcting the normalisation of some process in the SR using subsidiary measurements (CRs) is to perform a simultaneous fit of both regions, that way a proper correlation of systematics unc can be put in place, how do you propagate the individual sub-sources of the unc reported in Table 20 to the SR? How much of these have a negligible impact? (This applies for all data-driven (CRs) estimation in this analysis)

• ISR+tau team:

We are not doing a simultaneous fit in each of our CRs, instead we will include systematic uncertainties corresponding to the uncertainty in yield of non-targeted backgrounds in CRs we use to extract scale factors. Also, to minimize the effect due to non-targeted BGs, we take the approach outlined in our “strategy” section where we begin with highest purity CR, and correct by applying SFs down the line in lower purity CRs. This method is consistent with what many current published analyses in CMS have used, and also what had been presented and agreed upon with the previous conveners.

This is a similar approach taken in many recently published analysis (SUSY 17-007, EXO 19-015, EXO 17-016)

• Convener Follow-up:

Cecile: It is not complicated to set up a simultaneous fit with the CRs and it would improve the quality of the paper. You can get in touch with the Combine contacts if you need advice.

• ISR+tau team:

Per discussion in the SUS 3rd gen meeting on 20/10/23, we have to use our current methodology for reasons of time. We have reached out to the Combine experts with our implementation of systematic uncertainties related to non-targeted backgrounds in CRs where we extract these SFs to confirm that this a correct way to implement these.

Answers to First Set of Convener Questions

ANSWERED_SUSYconvenerQsV1.docx

Previous AN Versions

AN-20-188_v3.pdf AN-20-188_v2.pdf

-- JethroTaylorGaglione - 2023-04-12