Diphoton + MET Analysis Update

1. Diphoton + MET Analysis Update Bruce Schumm UC Santa Cruz / SCIPP 03 July 2013 Editorial Board Meeting

2. [From Philip’s 17 June email summarizing the 14 June Ed Board meeting]:  • Still missing: test uniformity of acceptance of the chosen cuts by fixing the optimized cuts, and then plotting eff and A over all grids. Conclusion on optimization: If there are no artificial features visible in the abovementioned plots, we treat the optimization as final. • FOLLOW UP: Plots had been made some time ago; shown in 28 June Ed Board meeting. • Response from Ed Board: • Questions to look at: • convince yourselves that even for the lowest acceptance you still have sensitivity in at least one of the signal regions (based on modified asimov, or real limit machine, or whatever) • Will happen as we study expected limits over parameter space • add a few explanatory plots to the note showing what exactly reduces your acceptance for low bino mass in SP, and low Wino mass in WP • Still need to do • make sure that the artificial effects in the grid, like the missing triangle for mbino=450 to 600 in the WP grid, does not cause a gap in the exclusion • Again, should become clear as we study expected limits over parameter space

3. [From Philip’s 17 June email summarizing the 14 June Ed Board meeting]:     Question: Is it OK to leave DeltaPhi^jet_min = 0.0 ? Looks like, is included in BG control Conclusion: Look at DeltaPhi plot in QCD control region whether this is well described, explain how it is controlled in the control regions, if no strange features: OK FOLLOW UP: Plots made; agreement not so good for  between MET and jet. • Response from Ed Board • make sure that the bin at deltaPhi=100 is included in the normalization • Needs to be done • study the effect of making the cut.not making the cut on the BG estimate, assign systematics if necessary (show that again, I guess) • Removing jetMET cut, WP2 QCD background estimate increases from 0.90 to 1.70 (1.52 for LocHadTopo). But note that some increase is expected since a cut is removed

4. [From 7 June email]      • Why document QCDg if not usebale acc. to page 3 with MetRefFinal? Conclusion: streamine the way it is presented in the note. Include QCDg everywhere in the plots examining the MET performance, but *just* remove the QCDg samples from the BG control tables which are argued to be unusable based on the first set of plots. With clear arguments of what to use and what not, based on the MET performance plots • Note updated. • [Ed Board should] Review the argument concerning the QCD background estimates in the note. Led to Ed-board follow up: • add a very conservative systematics on the final QCD estimate in the following way: • shape: relative difference between g+Iso and tg in the signal region • normalizaton: largest relative difference between g+Iso and tg or gg and QCDtg (whatever is larger) in the control region (MET<100) • Some further discussion (next pages…)

5. add a very conservative systematics on the final QCD estimate in the following way: • shape: relative difference between g+Iso and tg in the signal region • normalizaton: largest relative difference between g+Iso and tg or gg and QCDtg (whatever is larger) in the control region (MET<100) Ratios relative to QCDtg+iso After 10 July discussion: Systematic error should not come from comparison with LocHadTopo, but instead from looking at range of estimates that arise from varying the control region. Propose to compare estimates with control regions of 0 < MET < 60 (standard), 20 < MET < 80 and 40 < MET < 100. Underway

6. More plots: QCDxx/gg for MET < 100 ~50% changes in ratios as control region is chosen (any slice of MET as long as MET < 100)

7. [From Philip’s summary of the 26 June Ed Board meeting] • sideband studies: Very important study, I think! • the high-HT region generally looks fine. Maybe have a closer look at the one single event at HT>1800 on page 22, where you would expect none. • Not yet done. • For MetRefFinal, the chosen MET estimator, the low HT region looks strange. This is probably highly correlated with the overshoot at low MET in the plot on page 4. Since the plot on page 4 clearly shows that the MET agreement for the QCD estimate gets better above the region of 60<MET<100 shown here, this is probably nothing to worry about too much. But we should see the sideband study including MET systematics to be sure (see 'Ratio plots on QCD estimators' above). • See update below • It also could be that the QCD estimate from data is polluted by EWK in this very low MET area (not expected at higher MET in the signal region). Try to roughly estimate that (from MC?) and put it in the sideband estimate. Including these updates, if the sideband comparison loks good, I would have a lot of faith in the total BG estimate. • See update below

9. Note: will also do study cutting out events with _met_jet within 0.5 of 0,

11.    I've taken a look at the probably effect of performing the HT sideband study with QCDg+iso rather than QCDtg. The information I've used can be found in Ben's talk at the prior Ed Board meetinghttps://indico.cern.ch/getFile.py/access?contribId=1&resId=0&materialId=slides&confId=259792P4 shows the MetRefFinal ratios of QCGtg and QCDg+iso to tight-tight as a function of MET. Pages 23 and 24 show the QCDtg-based sideband studies for MetRefFinal. What I did was to look at the relative difference between QCDtg and QCDg+iso on P5 and use this to scale the QCD contribution to the expected background on pp 23-24, as follows.The QCD background estimate are based on the ratio of the given control sample between 60 < MET < 100 and 0 < MET < 60. This will be dominated by the events at the low end of the spectrum since the spectrum falls rapidly. For low MET, QCGtg and QCDg+iso are about the same, but for 60 < MET < 100, QCDg+iso is about 75% of QCDtg. Looking at the HT trends on pp 24-25, I guesstimate that this would lead to a 30% reduction of the QCD background estimate (QCDg+iso relative to QCDtg) for HT < 300, and a 20% reduction for 300 < HT < 600 and a 10% reduction for 600 < HT < 900. Making this correction guesses, I arrive at the followingNO DPHI CUT (NOTE: In parentheses is after muon veto in QCDtg control sample)HT RANGE                    DATA                  PRIOR EXPECTED        CORRECTED EXPECTEDHT < 300                      108                          257 (253)                          193300 < HT < 600            105                         160  (157)                          132600 < HT < 900            29                             37   (37)                            33WITH DPHI CUT (NOTE: In parentheses is after muon veto in QCDtg control sample)HT RANGE                    DATA                  PRIOR EXPECTED        CORRECTED EXPECTEDHT < 300                      65                          160 (158)                           122300 < HT < 600           58                           94    (92)                             78600 < HT < 900           12                            19    (19)                               17

12. Explain calculation of the 'observed' K factor and its                uncertainty in the note and check the result with SM                group and their Wgg analysis Had already been done. SM analysis apparently not yet mature. Separate limit for ttbar? Simplified models?                The suggestion is to take the GMSB inspired signal MC                grid and separate the MC at each point into pure qq, qt                and tt samples, then make simplified model limit result                plots using exactly the same selections, BG systematics                and BG estimation as for the combined result. - Other Action items:     - answer all remaining question on CDS     - upload new version of the note     - ratio plots for plots on page 3 and 4 (Done)    - check ctau reach of the analysis?     - look into MET>100 pure MI selection?     - look into SPS8 point Other than the ratio plots, none of these has been attacked.