ARC Review of B2G -16-004

Documentation

* CADI Line: http://cms.cern.ch/iCMS/analysisadmin/cadilines?line=B2G-16-004&tp=an&id=1603&ancode=B2G-16-004
* AN Note: AN-15-197 ( http://cms.cern.ch/iCMS/jsp/db_notes/noteInfo.jsp?cmsnoteid=CMS%20AN-2015/197)
* HN Forum: https://hypernews.cern.ch/HyperNews/CMS/get/B2G-16-004.html

Key to color-coded answers:

• We agree, implemented in analysis/text.
• We disagree, will not implement.
• Authors and/or ARC/conveners need to discuss. (Open item.)
• We agree, but need someone to do it. (Open item.)
• We agree, but no changes to the analysis/text are necessary.

ARC review: comments from Andrea Massironi on 26.2.2016

congratulation for the excellent analysis and documentation!

I still need to finish to read some details of the big AN, but I have a couple of questions and just few comments on the documentation, some physics related questions and just one main question:

Text and figure related (PAS): - it would be great to mention in the introduction the fact that W>tau (leptonic decay) nu sample is considered as well, since it is mentioned in the conclusion chapter.

Done

- l.101: "id changed to tight", I think it's CMS jargon, you may say "changed to tighter selections" of "the selections have been tightened"

Fixed. Actually, we rewrote as "the selections have been loosened", since the current IDs (tight) are looser than the high-mass ones (HeepID/highPt muon)

- general comment: there are references to the previous PAS, that is good, but maybe 2 lines for explanation when a reference is give may help. For example l.263 a one sentence summarizing the procedure to estimate the mass shift and resolution would be good to have.

added.

- (minor issue) the plots in the PAS have no consistent style. Why in some the uncertainty bars compare and in other not?

This is because some of the plots (e.g: the ttbar region control plots, used to extract wtagger scale factors) were made for the high-mass analysis, and since we used the same control region and scale factors of that analysis, the plots haven't changed. (We can fix them later in case it is explicitly required)

- Figure 1: what is the difference between the first and second row? Not mentioned in the caption.

Fixed. (top row is muon channel, bottom row is electron channel)

- Figure 1: what are the uncertainty bars? Only MC statistics uncertainty?

Yes, the control plots made at this level include only the MC statistical uncertainty.

- Table 1: related to the third comment, without an explanation the table looses a lot of communicative power. For example: why are there errors on the MC?

Fixed in the previous comment. The error is due to the fact that the values are extracted from a fit, so this is the error from the fit.

- Table 2: it is mentioned "migration" effects between categories. Between which categories? At the end you have only 2 regions (electron and muon), right?

Fixed to "signal yield". That was a feature of the high mass analysis, where we had two separate categories (W- and Z- enriched). Since in this case we only have one category, the uncertainty affects the signal yield. We also changed the caption accordingly.

Content related: - l106-111: if I understood correctly, the selections applied offline are exactly the same cuts as the online, then you are working also on the turn on curve. Is there a reason for this? Do you have a pt spectrum of your signal? How much is the uncertainty related to the use of the turn on, not only in terms of global normalization, but in terms of shaping the mVV distribution. Maybe the effect is negligible, because the pt spectrum for the signal is shifted to much higher pT ranges.

The offline selections are different from the online one and they are explained in sections 4.4 and 4.5. For instance, the pT offline threshold for muon (electron) is 40 (45) GeV, so we are working far from the trigger turn-on (for reference, see trigger plots in the AN).

- general remark: mVV is not defined. It could be trivial, but since you have a neutrino in the final state, it's not really trivial. I suppose you perform a kinematic fit (do you? I could not find a reference to it)

The explanation on how the neutrino's 4-momentum is reconstructed is given in sec.4.7. The longitudinal momentum of the neutrino is reconstructed imposing a constraint on the l-nu system, imposing M(lv)=M(W). A sentence (at the end of the same section) has been also added in order to explicitly define mVV.

- l.117: do you perform any jet cleaning, removing overlap with the leptons?

yes, we do. Explained at L135: "The AK8 (AK4) jets are required to be separated from any well-identified electron or muon by ∆R > 0.8 (0.3)"

-l.152: you mention the trimming algorithm at trigger level, but in the AN Table 16 you only mention single lepton triggers. Am I missing something?

indeed, we use only the pruning algorithm, not the trimming. The text has been fixed.

- l.220: why do you cut on DR and not on Dphi? By cutting on DR you may remove events that are boosted along Z.

We decided to keep the analysis as similar as possible to the high-mass one, so changing the selections as little as possible. Besides that, we checked the impact of this cut on the signal after the full analysis selection, and we found that the impact is negligible (<0.01%) on the signal efficiency for the two extreme mass points (M=600 GeV and M=1 TeV).

- l.329: tt and tW: how much is the relative contribution of the two samples? I'm asking because there is a theoretical uncertainty related to the relative contribution of the order of ~8%. If the shape of the tt and tW is different, this becomes a shape uncertainty

The contribution of single top is negligible wrt the tt one. As it can be observed from fig.2, tW is roughly 1/10 of tt. Moreover, the shapes of the two backgrounds (in Mj and Mlvj) are similar, look for instance at fig. 62 and 67 of the AN (v12).

Main concern: my main concern is about the bias and the bias test results. Looking at the bias results from "simulating background function A and fitting with parametric function B", like Figure 60 right of AN 2015/197_v12, I see there is a bias of 10%. How much is the usual "allowed" level of bias in parametric analyses?

For reference, the "allowed" level of bias in some Higgs analysis (e.g. H -> GamGam or HH -> bbGamGam) is 14%. As far as we know, there is no "official" recommandation anywhere, unfortunately.

My doubt, rather than 5% or 10% is on the shape of the bias: it has a distinct "excess" structure between 650 to 900 GeV, all in positive direction, that is very similar to Figure 6 of the PAS. Unfortunately, being the analysis un-blinded, I have to admit I could be "biased" by what we see in data, but the fact that the structure is the same, triggers some doubts. Since you have already the toys used to create the bias study, could you plot the distribution of the expected limit? That is: - for each toy simulated with A, you calculate the limit as a function of MG, by fitting with B - you have then a line, the observed limit - since you have many toys, you create a "band" If the bias has no effect on the final observed result, this band from toys should match the error band from figure 6 in the PAS: it could be a bit smaller or larger, but the important thing is that it should not be shifted.

We made the following additional check: we generate the toys with two functions (simple Exp and Power law), without signal injection, and then we fit the toys with the default function (ExpN). Then we compared the observed with the expected in both cases. The results are in the slide attached to the twiki (https://twiki.cern.ch/twiki/pub/Sandbox/B2G16004ARCReview/ARCreview1_290216.pdf). Basically we observed that the line of the observed is very close to the expected. Moreover, the two expected limits are very similar to the one from the default function, shown in the PAS.

Further explanation: both the expected (and the 1,2 sigma band) and the observed lines are built from the mean of the 500 toys. For instance, this means that the point of the observed at 750 GeV is the mean of the observed limit over the 500 toys at 750 GeV. Same story for the expected.

Thanks, Andrea

ARC review: comments from Justin Pilot on 29.2.2016

Dear Authors,

Thanks for the very complete documentation. I have gone through the PAS (v1) and AN (v12), and have a first set of questions and comments, as listed below.

Thanks, Justin

===

Physics Questions

- Why is the same jet correction used for jet pT and jet mass? These are not a priori expected to be the same, are they? Has this been studied?

Actually it is not: we apply L1-L2-L3 (L23) corrections to the jet pT, while for the mass we do not apply L1 pile-up correction since we are already using the pruning algorithm.

- Is the ttbar/single top correction factor of 0.69 from Section 5 consistent with other analyses? It seems to be lower by a factor of 2 from some other results I've seen. Why is this different from the factor of ~0.85 that is quoted in Sec. 6.1? Shouldn't these be the same? Also the ttbar control region plots seem to be different between AN/PAS, is this expected?

Indeed, the text in the PAS was not very clear. The 0.69 refers to the W-tagger scale factor, while the top scale factor is 0.85. We improved the explanation of this section like this: "For the tt and single top contributions, we compute a scale factor based on the difference be- tween data and simulation expectation in the signal region of the control sample. We apply all the cuts used to define the tt control region; then, looking only inside the jet mass signal region and applying the N-subjettiness 21 selection, we account for the difference after the τ 21 cut, to get the data/simulation scale factor. The final computed scale factor is 0.85 ± 0.04, combining muon and electron channels. For the WW/WZ and signal contributions, we are only concerned with the efficiency for pure W-jet signal, so it is necessary to subtract background contributions to get the correct signal efficiency scale factor. A simultaneous fit to the jet mass distributions is performed in the top- quark enriched sample, in both data and simulation, to extract the efficiency of the τ 21 cut. Differences in the resulting W-tagging efficiencies will be driven by the discrepancy between data and simulation in the τ 21 distribution. The ratio of the efficiency in data and simulation yields W-tagging scale factors are used to correct the total signal efficiency predicted by the simulation. The scale factor for W tagging is 0.69 ± 0.14, combining the muon and electron channels. A more detailed explanation of the procedure for the extraction of the W-tagger scale factor is given in [2]."

- PAS Figure 4 -- it is hard to read the red band in the pull distribution plot. It seems in the tail of the distribution, you are under-predicting the background contribution? Is this covered by systematic uncertainties? Do you have this on log-scale, and additionally the post-fit plot after limit setting?

The log-scale versions of the plots in figure 4 are shown above. The data seems to be in reasonable agreement with the prediction in the tail, given the uncertainties. Though the region of interest for this study is not in the tails - [600,1000].
Obtaining the post-fit plot after limit setting could take a couple of extra days (Zijun is working on that).

- Have you updated to the new luminosity uncertainty and central value? The PAS still lists 4.6%.

It is updated in the SVN - the text; reference is added to LUM-15-001; the limits are recalculated.

- I don't understand the difference between Table 2 and Table 3. Does Table 2 apply only to background (which ones)? Table 3 seems to describe the systematic effects for signal? Can this information be combined into a single table?

The purpose was to show the systematics related to the jet energy scale and resolution in a separate table. However, we combined the two tables, to have a more compact information.

- I share Andrea's concern about the bias tests showing an excess between 600-900 GeV. It might be good to look at the pull distributions for the nuisance parameters used in the fit to see if it is isolated to a specific contribution. Why is there such a big difference between Fig. 60 and 61 when the difference between the two functions used for parametrization seem to be within a few percent of each other (Fig. 69)?

Below you can find the pull-plot for the nuisance parameters post-fit. On the left is the case in which the toys have been generated with the Exp and fitted with the default ExpN, while in the right there is the case where the toys have been generated with a Pow and fitted with the ExpN. Each point in the pull plot is the mean of the distribution of the corresponding post-fit value of that nuisance.

The reason of the difference between fig. 60-61 and 69 is that in the alpha plot, the shown alternate function is an ExpTail; in the bias tests instead we tested several functions, in particular the results shown in fig.60 and 61 are with the Exp and Pow.

Editorial/Presentational Comments

PAS v1

= Please ensure that plot styles are similar; update to recommended CMS_lumi.C script where needed

L12 - Remove () around [6] L13 - Suggest changing "This motivates" -> "These results motivate" L14 - "which is the intent of" -> "which is described in" L20 - "natural width of the resonance"

Fixed

L79 - move (MC) immediately after "Monte Carlo"

Fixed

L96 - suggest "Most part" -> "Many" L99 - "lowered down" -> "lowered" L100 - "for both muon and electron categories"

Fixed

L135 - is the DR cut for leptons close to AK4 jets really 0.3?

Yes

L138 - remove "the W-jet candidates from". (You don't define W-jet until the end of this paragraph.

Fixed

L218-219 - "semileptonic and fully hadronic analyses" - it's not really a separate analysis. Maybe "semileptonic and fully hadronic W decay reconstruction" ?

Fixed

L224 - "is one or more" -> "are one or more"

Fixed

L227 - grammar problem - "Events where...have m_jet". Suggest "Events having m_jet... after the final selection are not considered..."

Fixed

L228 - remove "experimenter's"

Fixed

L230/1 - suggest rewording "The ... distributions ... are taken from simulation, and shown along with the data in Fig. 1 for both..."

Fixed

Figure 4 - Axis label should be M_VW? M_VV? You use several different terms throughout the PAS, please be consistent.

Changed to M_WW

L319 - "systematics" -> "systematic uncertainties" L319 - "is largely" -> "are largely the same as those described in [3]"

Fixed

L367 - "so it is considered" -> "and so are considered"

Fixed

L369 - "Tables"

Fixed

L371/371 - consistency of "Standard Model" vs "standard model"

Fixed

Figure 6 - include both electron- and muon-only plots

The per-channel limits are added as separated figure.

L384 - missing space in "WWin" L384 - should it be "VW"?

Fixed

ARC review: comments from Caterina Vernieri on 2.3.16:

Dear authors,

Congratulations for the nice result. Here few comments I have on the documentation (PAS v1)

thanks, Caterina

Abstract: specify the whole resonance mass range l102;

Fixed

155 You say that you changed the mass window from EXO-15-002, but you don't motivate the choice in the PAS.

We have added the following sentence: "The reason of this change is due to the fact that we are searching for neutral signals, therefore the splitting in WW- and WZ- categorization done in the high mass search is here sub-optimal. Then we have also performed some checks on the expected limit, varying the signal region window, and we found that the $65$-$95${\GeV} is the one which gives the best results;"

l112: I would rather move the primary vertex identification in sec2, around l51

We think it is better to keep the primary vertex identification in sec.4. In fact, this is a selection that is strictly related to this particular analysis, while sec.2 describes general things of the CMS detector, which are analysis-indipendent.

l211: How much is the W->taunu contributing to your acceptance? It would be good to add this information here.

The contribution is ~9% of the total number of signal events. Added the following sentence: "The contribution from W → τν events to the analysis is around 9% of the total number of signal events, measured on the two extreme mass points (600 and 1000 GeV)."

l226: It would be nice to add here the rate of ttbar rejection.

Information added to the PAS. After the full analysis selections, the rate of ttbar rejection due to the bveto only is 54% (55%) in the muon (electron) channel.

l262: the measured SF is the same as in EXO-15-002. Why don't just use it as reference for the plots in Fig2 and for Table1?

The entire section has been reformuled, following one of the comments from Justin (see above); moreover, one of the comments from Andrea was to add also some lines of explanation whenever there is a reference to the previous PAS. For these reasons we think it is better to keep the plots here, but maybe these can be further discussed.

Background modeling. I wonder if you have considered to use also the upper mass sideband to compute the transfer factor. I would guess it would give extra information to further constrain the shape. In general using more than one sideband would help to deal with any residual jet mass/pt dependency in the extrapolation of the dijet mass shape.

We tried this kind of test in the past. The main problem of using the high sideband is the correlation between Mj and MlvJ. In fact, the Mlvj shape in the high Mj sideband region is different from the one in the low-mass sideband or the one in the signal region, with a turn-on in the region of interest (see plots below: left, W+jets high sideband; right: W+jets low sideband), which makes more difficult to compute a transfer function (which is currently a "simple" ratio of two exponential function)

l 314: It's stated here Gauss+doubleCB is the model, but from the pre-approval slides it looks like you're using CB only. Which is the one that is actually used?

The right version is the one in the PAS. Indeed, we are using a doubleCB shape.

ARC review: comments Loic Quertenmont on 2.3.16:

Dear authors,

Congratulations for bringing this analysis to this stage. As I am not very familiar with the previous analysis, I'll do my best to review this.

See bellow my first set of comment on the PAS.

Best, Loic

General comments: Please verify all your usage of V, W and Z in the text because it's really inconsistent right now. I understand that it's hard to distinguish boosted W-->qq from Z-->qq, but that's not explained anywhere in the text so I would suggest to use W everywhere. (and eventually add a sentence saying that it's hard to distinguish)

We agree. Since we look for WW explicitly the text is changed to reflect this.

Title: is a bit mysterious: are you really looking for VW ? or (more likely) for a narrow resonance which decays in VW?

Agree, it’s now changed to “Search for a narrow resonance decaying to WW $\to lvqq$ in the mass range from 600-1000 GeV"

Abstract: - you mention a decay to a pair of W bosons while the title mention VW. Make it consistent.

Now the title is consistent with this line

- 2.2fb-1 will need to be updated to the new lumi value (of 2.3fb-1)

Done

- "is presented" is a bit alone in that long sentence

Done, moved to the beginning of the sentence

- gravitons --> such resonance ?

Did not change. Instead removed Graviton earlier in the abstract

L17: It is a bit unfortunate that you only focus on spin2 resonance, while the main reaction of the TH community on the CMS di-photon excess was that it would be much more interesting to see the results for a scalar. But I guess it's too late for this. I just hope there is not a clone paper of this one in a different physics group targeting the scalar resonance. (can you please check and let me know)

There is no other analysis at the moment in CMS which deals with scalar interpretation in this channel (NRA). It most likely will be done by this analysis team in some foreseeable future.

L21: This is again inconsistent with the title in VW.

the title has changed

Sec.3: It might be interesting to remind the principle behind the bulk graviton model in one or two sentence.

Done, but we prefere to add this into sec.1, improving the theory introduction. We did it as follows: "There are several theory models that motivate the existence of heavy particles that decay to pairs of bosons. These models usually aim to explain open questions of the Standard Model (SM) such as the integration of gravity into the SM using extra dimensions. Whereas the aim of the prior analysis was to test a large set of these different models, this analysis fo- cuses exclusively on the bulk scenario [7–9] of the Randall-Sundrum Warped Extra Dimensions model [10, 11] The bulk graviton model is an extension of the Warped Extra Dimensions framework, which proposes that the standard model fields also propagate on the extra spatial dimension. (“bulk”). From the experimental point of view, the main characteristics of the models are: • the large branching fractions to WW, ZZ and hh channels. Unlike the original RS graviton, the bulk graviton is not amenable to searches in the e + e − , μ + μ − and γγ channels, due to its extremely low branching fractions in those channels. • the polarization of the produced W and Z bosons. In the bulk graviton decay, the produced vector gauge bosons are longitudinal polarized more than 99% of the time."

L89: I would remove "supplementary" as there are no other Minbias interaction discussed before.

Fixed.

L96: Are you really forcing the reader (me in the present case) to read again [3] in order to know what is your selection? I would HIGHLY suggest that you describe briefly your selection before describing the difference with respect to [3]. OR, if this is explained in the following sub-section, move that small paragraph to the end of section 4.

It was explicitly asked during the convener's review to describe the differences wrt the high mass at the beginning of this section. However, we moved this at the end of the section, since it probably makes more sense.

Sec 4.2: Can you describe what you do when the same jets is reconstructed in both collections AK4 and AK8 ? Is there a type of cleaning/matching procedure?

The AK4 jets are only used to make a b-veto. Otherwise the default collection is AK8. When b-vetoing AK4 jets, the are explicitly asked to be outside DR > 0.8 from the AK8 jets. There is no cleaning done except forf the DR matching. Since they are not used for the same physics observable, there should not be a problem in multiple R parameter interpretations of the jet.

Sec 4.2: are you performing the jet smearing in the MC?

We do not perform jet energy smearing in the MC, however the effect of possible difference in jet energy resolution between data and MC is taken into account in the jet related systematic uncertainties - "jet energy resolution" in particular.

Sec 4.3: as you are extending the results to a lower mass range, you should probably discuss here the case where the jets from the W decay are resolved. Which should be more likely to happen for the lighter resonance that you are now targeting.

For the resonance masses above ~600 GeV the dominant mode is the one with merged jets. This is the case for similar studies (semileptonic WV) in HIG an SMP. The reconstruction efficiency as a function of the W pt can be seen from fig 1 in JME-13-006-paper for example - above 200 GeV the merged jets mode is taking over. The 600 GeV Bulk gravitons have average pt of ~280 GeV ( <10% of the cases W_pt<200GeV).

L152: what does it mean at trigger level ? Your triggers are single lepton triggers, so I don't really see the connection with the jet grooming algorithms.

Agree, these are cross-checks which are done but are irrelevant for this analysis. Removed.

Swap Sec 4.3.1 and 4.3.2: As 4.3.2 is done BEFORE Pruning, it sounds more logical to present it first.

This section discusses the AK8 clustering thus occurs before 4.3.2.

L199 : the raw Etmiss is modified.... --> Corrections to account... are applied.

Done.

Fig 1 (also applies to others): - "CMS data" --> data - CMS Preliminary must be larger than this. - GeV/c² --> GeV (OR the text should be change to use GeV/c and GeV/c² everywhere - 5 --> 5 GeV (in Y axis legend) - luminsoty must be increased to 2.3fb

%PGREEN% Done

Sec5: Is there a reason why the correction factors do not depends on the boost of the W ? Was this checked?

The W tag scale factor is derived in a particular pT region ~300-500 GeV simply due to ttbar kinematics. With the 13 TeV data, we have not checked the pT dependence due to lack of statistics (the uncertainties are quite large at the moment). That said, often when extrapolating for W's out of this pT region an extra uncertainty is applied based on MC with different parton shower models. For this search it's not necessary as the pT regime of the of the masses explored are similar to the ttbar validation region.

L271: mVV --> mWW in order to be consistent with the rest of the speech.

Done

L272 V-JEts has not been defined before, so I would suggest to switch to W-Jets (which would also be more consistant)

Done

L288: mVV and V-jet

Done

Sec 6.1: how do you suppress Z+Jets ? (I don't see any description of a third lepton veto, so I would expect that it can possibly contribute too). I believe you should add an explanation somewhere.

Indeed we do suppress Z+Jets by applying (loose) lepton veto, which reject events with any number of additional leptons. In the PAS we wrote "Events must have exactly one isolated muon with pT ...."(L182) and "We require exactly one electron with pT..." (L194).

Fig5: this has clearly not the quality level of a PAS. So improve it or drop it.

Improved. See attachment.

Table2: What about QCD scale uncertainties?

Fixed. Added 11 % scale uncertainty

Fig6: the Y axis legend is incorrect. the "95%" should apply to sigma X BR. Why not writing: sigma_{95%} (pp --> G_RS --> WW) (pb) ?

L384: typo

Fixed.

L384-385: Rephrase "in which at least one of the bosons decays hadronically" sorry, but that's a way of saying that you analysis is sensitive to WW-->qqqq, which is not correct I fear. That's also inconsistent with L385.

Fixed. removed "at least"

386 : simply say: "where the tau decay leptonically"

Fixed

Summary: It would be interesting to comment on previous results (as the search is argued to be an extension, you can comment on the common region)

We added that these limits improve the results of the high mass search in the common region (800-1000 GeV)

I am not checking the reference.

ARC review: comments Loic Quertenmont on 2.3.16 AN:

Dear authors,

Although the quality of the PAS was pretty high, the AN is not at all at the same level. It's more a big collection of plot than a readable documentation of what you've done. Sorry, but for a newcomer it's practically unreadable. And the fact that the same AN is used for different public results make the entire things even more confusing. Probably that for someone who is reviewing the analysis since the beginning it is easier, but for others it's very hard.

However, you can find bellow some comment and questions on the latest version of the AN. (but I wouldn't call that a review)

Best, Loic

L80: Trigger... What I read in this section is very very different than what is written in the PAS. I hope that the PAS is the correct version. If so, Correct the AN.

The triggers for the low mass extension are described in the section 10.3 - L598-600. They are different than those for the high mass analysis described earlier in the AN (L80). The original set of triggers used for high mass analysis do not provide optimal efficiency for the low mass searches, thus we had to use different set with lower pT thresholds.

L141-143 THis also sounds to be a remnant of the previous analysis, and is at least not consistent with the description made in the PAS.

The "low mass extension" is documented in the section 10. As it is an extension, many of the analysis components, techniques, studies, etc. are the same or similar to the original "high mass" analysis, thus the "low mass extension" is appended to the original(this) Analysis Note (as ch10).

L148-149: Idem same as above.

Fig2&3 are not updated (we have 5 times more lumi than this)

Figure 2 and 3 present results from a studies on the efficiency of the noise filters. These studies were performed during the data taking, thus the lower integrated luminosity quoted in the plots. Nevertheless, the used ~0.6/fb provide enough statistical precision for these checks and it was not necessary to repeat the study later on the entire 2015 dataset.

Fig 29: Can you comment on the large disagreement between your predictions and the points at large MWV ?

Figure 29 shows a fit to the W+jets MC to extract the functional form of the mWV distribution in WZ category for the "high mass" version of the analysis. The high mWV tails is where the MC statistics is very low. ... Sorry to say it, but for someone who hasn't been in the ARC since the beginning this AN is impossible to read, and actually point less. I am realizing now, that the only part of interest for B2G-16-004 starts at Sec.10. Before that is just an endless collection of plots with minimal description. Really impossible to follow. ...

Fig 71 and 74: I do not understand how you get a 1sigma excess for mWW=700--800 in the muon channel, while when we look on Fig 71 there seems more to have a deficit. It would be interesting to see the event yields for the backgrounds, signal and data around the mass of interest. For instance for mWW in [700, 800], or better for mWW=750+- 2*resolution.

First of all, it has to be kept in mind that the peak of the signal is slightly shifted on the right; this can be seen in Fig.72 of the AN. For instance, the fitted mean for the M=750 GeV signal is ~780 GeV. (We think this is due to the jet energy correction). Keeping in mind that, we can compare for example the data vs. the expectation in the range [peak-1sigma,peak+1sigma]. In this range, for the muon channel we observe 206 events vs. ~200 for the predicted background. For the electron channel, we observe 144 events in data, vs. 138 expected events obtained from the analysis. So data are a bit over the expectation. This is hard to see from the plots, but especially in the muon channel it can be seen that there are some bins with ~1 sigma over the prediction. However, this is not a relevant excess from the statistical point of view. ( and in fact, in the limit we observe only ~1 sigma of excess)

Still on Fig71: how did you choose the binning of this plot? Naively, I would say that a binning comparable to the detector resolution would be more appropriate. Can you perform a few teston the binning ? See how your limits changes when you multiply the #bins by 2 and when you divide the #bins by 2.

The upper limits are produced via an unbinned shape analysis, using the Higgs combination tool, thus the binning on Fig71 would not affect the limits. We have also synchronized the bin's width with the instrumental resolution, which for mWW would mean ~50 GeV (or twice wider than the previous 25GeV).

-- LucaBrianza - 2016-02-26

• plot_signal3.pdf: plot_signal3.pdf