Update on the Diphoton + MET Analysis Basckground Bruce Schumm, Susan Fowler (Penn), Osamu Jinnouchi (Tokyo Tech), Khilesh Mistry (Penn), Tobias Orthen (Tokyo Tech), Ryan Reese (SCIPP), Sheena Schier (SCIPP), Brig Williams (Penn) 25 November 2014 SUSY Group Analysis Meeting

25 November Conference Note public in early January: Search for Supersymmetry in Diphoton Events with Large Missing Transverse Momentum in 8 TeV pp Collision Data with the ATLAS Detector ATLAS-CONF Since then…

25 November  Complete retooling after loss of Ben Auerbach  Updated software/tools releases  “Harmonization” with other Photon + X + MET analyses  Re-thinking of approach to QCD background (big conceptual change, small numerical change) FAR talk delivered 30 October presented new background numbers; based on Version 2.0 of backup note (ATL-COM-PHYS )  Two sets of comments to address (Nikola from V2.0 reading, groups comments from FAR) to finalize background  Evaluate signal systematics  Set limits

25 November QCD Backgrounds Primer Estimate with  MC Estimate via control sample Sum of contributions scaled to  data for 0<MET<60 Primary systematic from varying “  ” contribution between 0 and 50%

25 November New (vs. Old) Background Results Note that QCD component uncertainties are >100%. OLD NEW “OLD” is from CONF Note

25 November I: Responses to Questions Raised in FAR Talk 1.Add uncertainty on the gamma-gamma MC modeling of Etmiss (e.g. add JES, photon ES, etc) Not yet done. However, given the large uncertainties on the QCD background component estimate, it is very unlikely to change things.

25 November Validate (e.g. using MC) that pseudo photon data (1 tight + 1 loose-not-tight) can model MET in gamma+jet and jet+jet data We find that the diphoton MET distribution has somewhat smaller tails than the control sample MET distribution, suggesting that the QCD background estimate may be a bit high. However, since the lower end of our systematic range for all 5 SR's QCD background estimate is 0 events, we feel that we should add no additional systematic uncertainty based on this comparison.  -jet MC: control sample selection  -jet MC: diphoton sample selection

25 November Document well how the systematics are extracted on the QCD background (in particular the question was on the upper limit of the SP SRs where the central value of the expectation is 0). SP1 & SP2 1)77%  MC + 23% Control Sample  N QCD < 68% CL 2)50%  MC + 50% Control Sample  N QCD < 68% CL 3)Since (0.11) 2 + (0.21) 2 = (0.23) 2 we say N QCD = 0.0 ( ) ( ) WP1 & WP2 Central value and stat error (actual stat error used is asymmetric Poisson) Sys error range

25 November Look at the diphoton cross-section analysis and check if you can use the gam-gam vs gam+jet fraction from that paper as a function of e.g. pt(gg) instead of using the inclusive H->gg analysis. From that paper, the fraction found for their selection is about 68%. For our result, we use 77%, with a systematic range between 50% and 100%. In addition, the MET and dphi distributions we observe for our selection suggest a larger, not a smaller, diphoton contribution than that suggested by the Higgs analysis. So we believe that the central value and range we have chosen make sense.

25 November Compare in one place the e->gamma fake factors and systematics of all photon+X analyses. (Related to this: Check that the systematics are evaluated consistently for all photon+X analyses when it should be the case (e.g. e->gamma fakes). The goal would be to be able to make a consistent description of the various analyses in the paper.) We compared our fake factors to those of the photon+bjet+MET analysis. The latter are higher than the former by 20-25%. However, our photon isolation requirement is tighter than that of the photon+b analysis because we have found this beneficial in reducing backgrounds. In addition, it should be noted that the background levels are an order of magnitude higher in the photon+b analysis than in the diphoton analysis. Thus, the degree of scrutiny placed on the estimation of the e->g fake rates might be expected to be rather different for the two analyses. To us it doesn't make sense for the diphoton analysis to throw effort into reducing e->g fake rate uncertainties given the small contribution and the large statistical errors arising from the control sample. Thus, upon thinking about it a bit more, we don't find it warranted to expend effort in "harmonizing" the fake-rate determinations and errors between the diphoton group and the remaining photon+X+MET groups.

25 November Check the photon pt spectrum in the el->ph fake control region and in signal regions. This would be important if the FF turn out to be pt-dependent. Rather than do this, we have instead explored the possible effect of a Pt dependence of the fake rate on the EW background estimate. Applying the Pt dependence measured by the photon+b analysis to our overall fake rate, we find a ~25% reduction in the EW background estimate. We now take this as an additional systematic uncertainty on our EW background estimate, although the effect on the overall background estimate from the inclusion of this systematic is negligible (an increase of.01 events in the systematic error for the WP2 and MIS SRs; no change for the other SRs).

25 November Plot the photon Etmiss component and the other Etmiss components separately for data vs MC. Also add plots of the diphoton system Still underway. Unlikely to impact final result. 8. Add a systematics band to data/MC comparisons (in particular because the modeling of Etmiss is not well-modeled on p.18) Still underway. Unlikely to impact final result.

25 November Add upper Etmiss cut on Wgg CR to avoid overlap with SRs The CR is now 50 < MET < 150. Since all events in the CR had MET<125, the "enhancement factor“ went up from to Lepton-  CR 50 < MET < 150

25 November Add an uncertainty (e.g. 50%) to the bkgd contamination of the Wgg contamination and check the effect We have included an overall 50% uncertainty on the total background contribution to the lepton-gg control region. The resulting uncertainty on the Wgg production rate is 23%. We conservatively add this in quadrature to the 55% uncertainty from other sources. It makes little difference in the overall background uncertainty in the SR. 11. Explain clearly why there is no extrapolation uncertainty on going from Wgg CR-> SR (question from Mario) The K factor in question has a steep dependence on one variable: the magnitude of the momentum of the l  system. However, above 100 GeV in this observable, the K factor levels out to its asymptotic value (which is about 20). On the other hand, the selection cuts for the various SRs bias the selection of W  events to very high values of the l  momentum - MC studies indicate that almost the entire contribution comes from events above 100 GeV. So as long as the control region itself has a lower bound of 100 GeV in this observable, we are measuring the correct region of phase space and there is no extrapolation to be done.

25 November II: Responses to selected CDS Questions From Nikola 1. Do you apply the BCH cleaning? We have explored the effects of BCH cleaning on the signal and background MC selection efficiencies and found the effects to be negligible. 2. How do you know that MC simulation reproduces MET tails? Why do you want to set the relative fraction of your fake met samples using the H  sample sample? This has now been addressed above and in the prior FAR talk. 3. The QCD control sample needs to include backgrounds both from  -jet and dijet events. The relative fractions of these contributions can depend on the kinematic cuts. Are the MET distributions the same for these two samples? All the same kinematic cuts are applied to the control sample as to the signal sample, for each SR independently. In addition, all control sample events include at least one jet faking a (loose) photon, with a corresponding effect on the MET tail. So within the large (> 100%) uncertainties, we believe that this is incorporated in our range of uncertainty.

25 November Why you did not consider performing a fit to extract the relative fraction of the  and  -jet subsamples of the QCD background model? In fact, we had wanted to do just that. However, in the control region, both samples (MC diphoton and the QCD control sample) overestimate the tight- isolated diphoton MET distribution for intermediate values of MET (between 50 and 100 GeV), with the MC diphoton distribution being much closer to the diphoton data than the control sample. Thus, it is clear that a fit would find that the sample is 100% diphoton. So in some sense we did do a fit, and include that result as a lower bound on the possible QCD background contribution.

25 November III: Updated Limit Results Expected vs. observed by SR Model-independent (number of event) limits from CONF Note Re-calculated model- independent limits using CONF Note inputs

25 November Question: Do we exclude low-mass winos? # events 95&CL is ~4  Yes, easily! Yield, in # events

25 November Expected and observed limits by SR; new background results, full signal systematics SP1 WP1 SP2 WP2

25 November Old/New Limits by Grid, New Background, Full Signal Systematics Gluino-bino new Gluino-bino CONF Wino-bino new Wino-bino CONF

25 November Summary/Outlook Most (all major) tasks from 30 Oct FAR talk CDS comments addressed. Remaining tasks should not take long Backgrounds finalized Backup note and written responses up-to-date with respect to CDS comment, but new version awaits remaining tasks before uploading. Signal systematics estimated and limits set Very little difference in either backgrounds or limits relative to CONF result Need to document systematics and limit setting, including a number of standard limit plots (cross section limits, expected-limit-only plots, etc.) We hope work is largely pro-forma at this point although we need to see what comments arise when updated Note is circulated.

22 April (NO) BACKUP