Review AN-2016/131

Comments by Ivan Mikulec Apr25,2016

cms.cern.ch/iCMS/jsp/openfile.jsp?tp=draft&files=AN2016_131_v2.pdf

Dear Lovedeep and authors!

I went through the analysis note and here are my comments/questions. It would be great if you could try to get over these before or at least during the meeting of Wed.

Cheers, Ivan

1) Could you, please, summarize somewhere in the AN (and in your presentation) all differences wrt. 13 TeV analysis. This will help all reviewers in the process.

In slides for non-had meeting on Apr 27: https://indico.cern.ch/event/523680/

2) Fig. 3: Could you, please, make the caption more verbose? Why, in the right plot there is no g+j background left and dijet increased so much?

SM processes in left plot are without cross-section normalization, but RHS plot is when each process is normalised properly according to cross-section, and then pu-weights are applied. And these plots are prepared without any event selection requirements, so QCD (dijet) dominates Gamma+Jet. For next version of AN, "left side" plot would be replaced with proper normalization of SM processes according to respective cross-sections. Caption edited accordingly.

3) Fig. 4a: Could you describe in the text how is this plot obtained (which sample which selection)? How serious is the discrepancy on the signal side?

We followed EGM POG recipes for this given at https://indico.cern.ch/event/298231/contributions/680379/attachments/560675/772427/egamma-Jan29-2014.pdf . It is same as what HIG-16-014 that search for high mass H -> Zgamma, used. We edited text in AN accordingly. Plot in Fig4a has very minimal skimming level selections i.e. events should have atleast one fat jet, one photon; pt> 150 GeV for both objects, HLT_Photon150 passed. And samples in stack are all SM processes (qcd dijet, gamma+jet, and other ewk v+jets, V+gamma), along with data and signal samples as histograms. For signal selection of BDT response in EB, we require this response to be above 0.133 in EB; zooming into this region, as can be seen in this plot, we have reasonable agreement between data and MC. * hpho_MVAEB_pre.png:

4) Fig. 4b: You seem to optimize only against efficiency. Please, could you support your id choice with some sensitivity figure of merit?

We checked the id choice w.r.t expected limits: For cut based ID; for Mx = 750, exp. limit is 23.72 fb & for Mx = 1 it is 12.43 fb. For MVA ID, exp limit for 750 GeV is 22.95 fb and for 1TeV it is 11.59 fb. thus, MVA computes slightly better expected limits.

5) Table 5: is this selection a POG WP?

Yes, this is JETMET recommended tight WP. Reference provided now in AN.

6) Fig. 5: is the walking of the pruned jet mass with signal mass understood?

For high mass signals, we have pruned jet mass distribution wider and with more low-mass tails, this is due to well known effect and observed by other EXO VV analyses as well as by 13TeV Zg.

7) Section 3.4: were the CSV WPs and corresponding SFs validated also in the subjet environment? Could you specify in the AN the size of the SFs? From Fig. 7 they look quite different from 1.

yes, CSV Loose and Medium WPs are supported for boosted event topologies (BTV-13-001). We followed the same recipe to apply btagging Sfs as prescribed by BTV group for 8TeV here: https://twiki.cern.ch/twiki/bin/viewauth/CMS/BtagRecommendation53XReReco#Boosted_event_topologies. References are in AN and text for SFs too. Following recipes from BTV-POG to apply SFs to reweight events, we find the variation in Sfs from 0.85 to 1.3, depending on jet flavor, pt and eta.

8) l. 183: is the k-factor of 1.3 used only for the relative scale between qcd and ewk processes? When you say "distributions in MC are normalized to match the yield in data", do you mean you scale all bkg MC samples with the same factor?

We scaled each dijet and gamma+jet process by 1.3 as done by public q*-analysis. In order to normalize SM backgrounds to data integral, we scale each background component according to its relevant contribution i.e. if gamma+jet contributes ~60%, so we scale it to 0.6*dataIntegral.

9) l. 185/187: please fix the Fig. reference (the one on l. 185 should point to FIg. 6 and not 8) [alsofor all other broken Fig references later]

fixed

10) ll. 204-211: It is not very clear how you make this optimization, could you, please, make it more clear in the AN: a) why is the background taken from Mj sideband? Is it because it is taken from data?

yes, background is taken from 50 < pruned jet mass < 70 GeV in data. text in AN fixed.

b) when the background is from Mj sideband, how is it scaled to the signal region?

We scale shapes of variables (like pT/M) taken from Mj sideband in data to the yields in signal region.

c) first you say you optimize against S/sqrt(B) but then you make the choice based on the expected limit, which is not unambiguous since for MX=1TeV the table 6 would point to a cut of 0.40.

We find that 0.34 to 0.40 gives us quite close sensitivity in both s/root(B) as well in expected limits. We picked pT/M > 0.34 as final selection to make this analysis close to our 13TeV one, thus to keep review easy and fast.

11) ll. 212-215: Is the technique here the same as for pt/MZ? Did you try also e.g. Medium+Medium?

yes, we computed expected limits. Following BTV-POG recommendations we found following combination as the optimal one: if dR(subjets)>0.4, select CSV Loose+Medium btag working points, else if subjets are quite close, btag fatjet only with CSV Medium WP.

12) Tables 10-11: could you please, give also the uncertainties here?

Done

13) Table 11: Which mass range does this correspond to? Is the data entry the total background after all cuts in the full mass range?

yes, here we used full mass range as it is just total integral of data, not shape.

14) Section 5.2: Did you perform background bias studies? They will be needed.

Work in progress, using same methods as for 13TeV

15 Figs. 12-13: could you, please, make more clear what does it mean unweighted sideband?

Since we are blinded, so we use sideband in data for background shape. For two categories in analysis, b-tagged and anti-b-tagged, we scale this sideband according to yields in signal region in data. so, figure 21a is when Mzg sideband distribution is scaled to signal region yields of anti-b-tagged categories, 21b is when sideband scaled to b-tagged yields in signal region and 21c is directly Mzg in sideband i.e. no scaling.

16 Section 6.1: The JES is usually applied to data and the uncertainty is evaluated by propagating it to the whole analysis. It seems that here it is applied in an opposite way (only on signal MC). Could you, please, clarify this?

The JES uncertainty is only considered for the simulation of the resonance, as the background component comes directly from the data from the S+B fit and therefore has the same JES as the data.

17) Section 6.4: Are the POG photon id SFs valid in this energy range?

We have assigned extra 2% uncertainty to take this into account. We will confirm this with EGM POG also.

18) Section 6.4: What is the time scale for adding photon energy scale and resolution systematics?

now in AN, done.

19) Section 7.1: Is the search range 700-3000 or 700-3500 GeV? The table says the latter and the Figs. the former...

We provide limits upto 3TeV in mass. Table edited for this.

20) Section 7.1: are the limits calculated with asymptotic approximation or with toys?

We use asymptotic approximation with combine tool.

-- LovedeepKaurSaini - 2016-04-26