https://hypernews.cern.ch/HyperNews/CMS/get/AUX/2017/02/23/21:11:46-11009-da11846_report_attachment_1.pdf

Color code:

Comments fully answered
Questions for ARC and LE
Comments without answer

Page 1:

1) We have changed the text as suggested.

2) We have changed the text as suggested.

3) The word "presence" has been replaced by "multiplicity".

4) 5) 6) 8) It is our convention to not use a hyphen for "top squark". However, we will follow whatever guideline the editor suggests.
(ie we keep no hyphen and ask editor guideline, is it ok?)

7) We have changed the text as suggested.

9) Lower limits on LSP mass have been added in the text.

10) We have changed the text as suggested.

11) We have changed the text as suggested.

12) At some point in the text the improvement of the new tagger should be quantified.

We have modified the text as follows: This top quark tagging algorithm is improved relative to the one described in Ref.~\cite{stop8TeV}, to enhance the sensitivity for selecting top quarks with large Lorentz boosts that cause the merging of jets among the top decay products. Evaluated from simulation the tagging efficiency for the algorithm described in Ref.~\cite{stop8TeV} is about 20\% at top quark $\pt=600\GeV$ and drops quickly to close to 0 for higher top quark \pt. Figure~\ref{fig:MethodAlpha_toptagger_efficiency} clearly demonstrates the tagger improvement with an efficiency around 55\% at the same top quark $\pt=600\GeV$ and rises for top quarks with larger boosts.

13) We have changed the text as suggested.

Page 2:

1) We have changed the text as suggested.

2) We have changed the text as suggested.

3) We have changed the text as suggested.

4) We have changed the text as suggested.

5) We have changed the text as suggested.

6) 7) 8) It is our convention to not use a hyphen for "b quark" and "lead tungstate". However, we will follow whatever guideline the editor suggests.
(ie we keep no hyphen and ask editor guideline, is it ok?)

9) It is our convention to write "high-level trigger" (and not "high-level-trigger"). However, we will follow whatever guideline the editor suggests.
(ie we keep it like this and ask editor guideline, is it ok?)

Page 3:

1) Perhaps this one should be with an hyphen. To be checked with CMS publication guidelines.

2) It is our convention to not use a hyphen for "b tagging". However, we will follow whatever guideline the editor suggests.
(ie we keep no hyphen and ask editor guideline, is it ok?)

3) Perhaps this one should be with an hyphen. To be checked with CMS publication guidelines.

4) We have changed the text as suggested.

5) It is our convention to not use a hyphen for "top squark". However, we will follow whatever guideline the editor suggests.
(ie we keep no hyphen and ask editor guideline, is it ok?)

Page 4:

1) We have changed the text as suggested.

2) We have changed the text as suggested.

3) We have now explicitly added "charm quarks". The decay products of the W from the chargino are already covered in the "light-flavor quarks".

4) It is our convention to not use a hyphen for "t tagging". However, we will follow whatever guideline the editor suggests.
(ie we keep no hyphen and ask editor guideline, is it ok?)

Page 5:

1) We have changed the text as suggested.

2) We have added some extra text to clarify this point.

3) Add explanation.

Relocated the sentence "Assuming massless input jets and trijet mass $m_{\text{3-jet}} = m_{\PQt}$" to the end of the paragraph for the equation and add one sentence there as: "Assuming massless input jets and trijet mass $m_{\text{3-jet}} = m_{\PQt}$, each of the three criteria can be deduced to be that a ratio of $m_{23}/m_{3-jet}$, $m_{12}/m_{3-jet}$ or $m_{13}/m_{3-jet}$ is within the range of $[R_{\text{min}}, R_{\text{max}}]$.

4) We have changed the text as suggested.

5) We prefer to keep the notation since R_min and R_max are the notations used in the reference for these particular requirements. Since R_min and R_max never occur without their subscripts, we think there is no confusion.

6) 7) It is our convention to not use a x symbol where multiplication is the default operation. However, we will follow whatever guideline the editor suggests.
(is it ok?)

8) The word "good" has been replaced by "high".

9) The 30 and 85% numbers in the text are approximate and are meant to give the reader an idea of the size of the top tagger efficiency. These efficiency numbers are also not used directly in the analysis, and so we do not compute a systematic uncertainty based on this plot. (The error bars on the data points in Fig. 3 depict only the statistical uncertainty.) What we do include as a systematic uncertainty for top tagging comes from the data/MC and Fastsim/Fullsim differences in the tagging efficiency. These differences, and therefore the systematic uncertainties, are small, around 5--10%.

10) We measured the tagging efficiency using different signals and ttbar backgrounds. We found the measured efficiency curves agree within their statistical uncertainty. We have added this clarification to the text.

We have updated the text to make it more clear that this is the selection used to measure the top tagging efficiency. The 4 jets requirement is indeed for the consistency with the signal region requirements.

12) 13) We have changed the text as suggested.

14) We have changed the text as suggested.

15) We have changed the text as suggested.

16) This is intended to be slightly looser than the mass range used to identify a single AK4 jet top candidate since for trijet candidate the mass resolution is expected to be worse than the single boosted top candidate.

17) We will provide this information in the supplementary material on the analysis public webpage. 17) Zhenbin and Koushik need to work on this

18) We have changed the text as suggested.

Page 6:

1) We will provide this plot in the supplementary material. Zhenbin need to work on this

2) In general, there could be multiple visible daughters resulting from a heavy particle decay (as was indicated in the line just above eq.5). And indeed, in our usage of MT2, the top candidates used as input to the MT2 calculation can be composed of multiple visible jets. We therefore prefer to keep the current phrasing.

3) We now indicate that the qTi are unknown, and that MT2 is a minimization with the constraint that their sum equals the missing transverse momentum (as was already stated in this paragraph).

4) In the MT2 calculation, one has to assume mass of the invisible particle. Since the MT2 variable is used for both signal and SM processes, it is convenient and simple to assume a fixed value for all scenarios. Although we can use mass of neutralino's, we do not know the exact value for sure. Since MT2 is also used as an important variable to reject the SM ttbar background, it is natural to use the neutrino mass for the invisible particle and calculate MT2 under the assumption uniformly for signals as well.

5) The sentence has been modified.

6) (Zhenbin) consistent with other groups? (Zhenbin) yes, it was changed to be consistent with other groups?

7) We have changed the text as suggested.

8) All the events in this analysis are not collected by this trigger. For example the events selected for the hadronic tau background estimation are using a different trigger, as mention in section IV B.

9) It is our convention to not use a hyphen for "top squark". However, we will follow whatever guideline the editor suggests.
(ie we keep no hyphen and ask editor guideline, is it ok?)

10) We have changed the text as suggested.

11) Move the MT2 definition earlier?

Page 7:

1) This is indeed the end of the baseline selection. Some modifications in the text have been made to make this easier to understand (cf other related questions). After the baseline cuts, 422pm6 events are expected from standard model monte carlo prediction, 8.39pm0.04 for T2tt(850,100). The motivation for MT2 cut at 200 GeV is primarily to reduce the SM semi-leptonic ttbar background. MT2 should have a kinematic edge around top quark mass for the SM semi-leptonic ttbar background. While for the direct stop production signals, we have most of top squark mass larger than top quark mass. And for the gluino mediated stop production signals, generally we have large MET and/or hadronic activity. Therefore the MT2 cut at 200 GeV is a good choice to reduce a great amount of ttbar background while keep high efficiency for most of the signal points. (removing the MT2 cut: 615pm7 SM events and 8.56pm0.04 T2tt(850,100) events)

2) Each background uses a different control sample to predict the number of events in the signal region. It is why the control sample definitions are not described in a separated section, but with the background estimation description. The signal regions were chosen after a sensitivity study performed with MC simulation to obtain the best sensitivity for some benchmark signal points. The CS for lost lepton is a muon+jets control sample. We have chosen to use the same cuts as the signal region (to have very similar events in term of MET,MT2, jet multiplicity, etc), except that the muon veto is replaced by asking for exactly 1 isolated muon (to get an orthogonal sample). To reduce any possible signal contamination, an additional mtW cut is added, as explain in the text. The CS for had tau is indeed similar (also muon+jet events) but not exactly the same because the fact that we replace the muon pt by a tau pt using a template can give larger MET to a given event, so we are using a different trigger selection and met requirement to be able to use events with lower met values. For QCD, the control sample is very different, because we are using inverted Dphi events. It is also very different for the Zinv background which is using Zmumu events.

3) These plots use direct MC simulation to estimate the various backgrounds, and are meant as introduction to which backgrounds are the most important for the analysis. We have added "simulation" to the text to clarify this point.

4) The word "preselection" has been removed from the trigger section to avoid any confusion. What we call "preselection" is all the analysis cuts defined in this "event selection" section, before splitting the events in terms of tops, b jets multiplicities, and MET, MT2 values. The text in this section has also been modified to make this easier to understand for the reader.

5) Perhaps this one should be with an hyphen. To be checked with CMS publication guidelines.

6) We have changed the text as suggested.

7) The sentence in the text has been modified to explain that it was to improve the sensitivity of all signal topologies. Different signals will have high signal/background ratio in different bins, depending of the MET and MT2 spectrum of the signals. The cuts values have been determined to maximize the sensitivity for some benchmark signals (eg high stop mass, low LSP or high stop mass, high LSP mass, etc) and taking also into account that the corresponding muon+jet control sample needs to have enough events to allow a proper background estimate.

8) This is an introduction to the various backgrounds of the analysis. This number comes from MC and is just an estimate for the reader to see the importance of each background. This number is not used in the analysis. All backgrounds are estimated with the data-driven technics described in this section.

9) The closure test is made using only MC simulation. No date are involved here. No data-MC scale factor is applied here.

10) We are talking about statistical fluctuations in MC because there is no data involved in this closure test. This is a purely MC-based test.

Page 8:

1) We think it’s clear to see the lines. And to some extent, we think this is probably the best we can have now.

2) The signal curves are not normalized to the cross-section here because we want to show the shape of these distributions. If we normalized them the cross-section, each signal needs to be multiplied by a different number to make them readable and we will end up with the same figure but with a more complicated legend.

3) We don't explain here in detail how all these systematic uncertainties are computed because most of these systematics are quite small with respect to the closure uncertainty and with respect to the statistical uncertainty. For the lepton uncertainty, the systematic is computed by adding in quadrature the uncertainty coming from the limited Monte Carlo statistic used to compute these efficiencies and from the uncertainty covering data/MC difference. The data/MC difference is estimated using tag&probe technique in Z resonance, comparing the results obtained in data and MC. Since no systematic bias becomes visible, no correction based on data/MC scaling factors is introduced, we are only taking this into account with the systematic uncertainty obtained by propagating the scale-factor. The uncertainty of the acceptance efficiency consists of the uncertainty in the parton distribution functions (PDF), the MC renormalization/factorization scale and the uncertainty arising from the statistical precision of the efficiency maps. The PDF uncertainties are studied by varying PDF sets used to produce the MC samples, according to their uncertainties for the baseline selection. The purity is expected to be very high (> 99%) so this only leads to a minor systematic uncertainty and a conservative uncertainty of 20% on the impurity is assigned. Furthermore, the statistical uncertainties from MC are propagated. Both dileptonic corrections are only minor compared to the remaining ones so a conservative systematic uncertainty is assigned here. An uncertainty of 50% is assigned on the number of di-leptonic events. Furthermore, the statistical uncertainties from MC are propagated. The uncertainty associated with the MT cut consists also of two parts: the statistical uncertainty of the efficiency map from MC and a systematic uncertainty. For the latter, the uncertainty on the jet energy corrections is propagated to pmiss and the efficiency of the MT cut is recalculated. The isolated-track vetoes lead to a reduction of about 40% of the lost-lepton background. Due to the size of this reduction, it is important to study the validity of the efficiency maps. A detailed study has been performed comparing Tag-and-Probe efficiencies on the Z resonance from data and MC. This study has shown that a systematic uncertainty of 10% is a conservative estimate for the systematic uncertainty on the number of events removed by the isolated track veto.

4) We have changed the text as suggested.

Page 9:

1) "as defined in the text" has been removed.

2) The cuts in MET and MT2 were chosen after a sensitivity study performed with MC simulation. One limitation considered in this sensitivity study is the need of enough statistic in the corresponding muon+jet control sample used for the lost lepton background estimation.

3) We have changed the text as suggested.

4) We have changed the text as suggested.

5) It is our convention to not use a hyphen for "single top quark". However, we will follow whatever guideline the editor suggests.
(ie we keep no hyphen and ask editor guideline, is it ok?)

Page 10:

1) No data-MC scale factor is applied here, because the closure test is a purely MC based test.

2) In MC, because the closure test is purely MC based.

3) These systematic uncertainties are computed using the same methods than the ones used for the lost lepton systematics.

4) "control region" has been changed to "control sample".

5) We have clarified here in the text that this is the shape of the N_j distribution.

6) We have modified the opening sentence of the Znunu section to make the process more clear from the onset of the section.

7) "as defined in the text" has been removed.

8) This has been clarified in the area referenced in question 5.

9) We have changed the text as suggested.

10) Yes we do.

Page 11:

2) We have changed the text as suggested.

3) We now explicitly mention that this plot is for the dimuon control region. So there is no contribution from Z->nunu, only from DY->mumu. We have also updated the caption to reduce confusion.

4) We derive a normalization factor from data (Rnorm) for the DY process. The assumption of the method is that this normalization factor should be the same for DY->mumu (with dimuon mass on the Z peak) and the Z->nunu process. Therefore, we do not directly use the luminosity or theoretical cross section, and do not assign any uncertainty based on them. Instead we apply the statistical uncertainty in Rnorm as an uncertainty on our prediction.

5) hyphen again...

Page 12:

1) The aim of CS is to find a multijet-rich region. The events in the CS are by definition orthogonal to the search region. The translation factor is here to translate the number of events in the CS to the number of qcd events in the search region. The text has been updated to make this easier to understand.

2) We have changed the text as suggested.

3) The translation factor is the number of qcd events in the search region divided by the number of events in the CS. This ratio is measured in data, using a low-met sideband, and the dependence as a function of MET and MT2 is taken into account by using a linear approximation, measured from the simulation. The text has been modified to make this easier to understand.

4) Caption is updated to mention what is included in "rare" processes.

5) We have changed the text as suggested.

6) None of the processes are Z->nunu, since this plot is for the dimuon control region.

Page 13:

1) Indeed, it is search bin. We have added "search bin" explicitly to the end of this sentence.

2) We use cross-section for signals for NLO-NLL calculation. The systematics do not include the cross-section uncertainty as they are separately plotted as bands for observed limits.

3) We have now introduced the term "Rare" in the beginning of Section 4. We prefer to keep Fig. 4 where it is because it serves as introduction to the background composition and thus the background estimation section.

4) Yes, every time we calculate the systematics, we vary the uncertainty of the source and then allow the calculated events migrate between signal regions.

5) We are using data-driven method to predict the backgrounds. These methods are validated with closure tests on MC, as described in the previous sections.

Page 14:

1) fix legend?

The legend is moved a bit so that the line divides between nt=1 and nt=2 can be seen.

2) We will provide this information in the supplementary material that will be available on the public analysis webpage.

3) The statistical uncertainties on data/MC control samples are modeled by the gamma distribution as the generalization of the Poisson distributions to regions with non-unity event weights (due to translation/correction factors and/or MC cross-section weights and the like). Therefore, if the control sample has a sufficiently small effective number of events (say the control sample has 4 events) then the uncertainty on the weighted background estimate will be noticeably asymmetric after applying the appropriate weights for the final prediction.

Page 16:

1) We will provide this information in the supplementary material that will be available on the public analysis webpage.

2) The statistical uncertainties on data/MC control samples are modeled by the gamma distribution as the generalization of the Poisson distributions to regions with non-unity event weights (due to translation/correction factors and/or MC cross-section weights and the like). Therefore, if the control sample has a sufficiently small effective number of events (say the control sample has 4 events) then the uncertainty on the weighted background estimate will be noticeably asymmetric after applying the appropriate weights for the final prediction.

3) We have changed the text as suggested.

4) We have changed the text as suggested.

5) 6) We indeed mean both CMS and ATLAS. The references cited in the introduction already include relevant analyses from both ATLAS and CMS.

7) We have changed the text as suggested.

8) We have changed the text as suggested.

Page 17:

1) We have changed the text as suggested.

Page 19:

1) We replaced "right" by "bottom".

-- FlorentLacroix - 2017-02-24