Diphoton+MET 2015: Overview of Path towards First Results A living document… Bruce Schumm SCIPP 18 May 2015
May Timelines I Is this still the best guess? 50 nsec running Few pb -1 by end of May (but trigger in commissioning) As much as 1 fb -1 by end of June 2-3 fb -1 by end of August can use for CONF NOTE 25 nsec running 10 fb -1 end of October For journal publication Proposal: Push/optimize for 2-3 fb -1 result
May Timelines II If we push for “August” result… Analysis walkthrough beginning of June ~2 Hr process, with much discussion Expected to present unblinding case during walkthrough, up to necessary lacunae associated with data-driven studies Editorial Board formed at that point Draft of support note expected at that point Seems unlikely. Grids probably not event ready. Instead just present status
May Tasks Overview Code/Infrastructure [Some done] xAOD Derivations Higher-level infrastructure (Ryan’s package) Events variables (MET with photons, etc.) Event selection Preliminary studies Optimization Backgrounds QCD Electroweak Irreducible Overlap (?) Models [Largely done] SM samples Strong & EW signal Full vs. fast sim?
May Models SM samples defined; generation underway. Much overlap with other groups; much hard work by Milano group! Gluino, wino grids defined All BF and decay length issues resolved Fast Sim sufficiently validated Of order ~2m events at 10K/point, more or less approved Final validation step (generator-level filter to ensure two binos in each event) underway Need this soon! (optimization)
May Gluino-Bino Grid (not quite final version)
May Wino-Bino Grid (also not quite final)
May Backgrounds - QCD Prior approach was to assume real diphotons are 75 25% of low-MET background Diphoton MC used to estimate high-MET contribution Pseudophoton control sample scaled to remainder of low-MET events used to estimate -jet contribution Using 8 TeV data to explore new approach (ABCD method with pseudophotons and relaxed isolation); preliminary results expected soon If this doesn’t work, will need to fall back to old pseudophoton control-sample technique
May Backgrounds - EW Estimate with e control sample scaled by e fake rate Need to select e control sample Tag&probe study of e fake rate underway [Giacomo] MAYBE: W MC suggests that ~25% of EW background doesn’t arise from e fakes Some of this may be accounted for in QCD background Some of QCD background may include e fake events Prior approach was to include 25% systematic error on the EW background Perform QCD/EW background overlap study?
May Backgrounds - Irreducible W contribution estimated via l control sample and simultaneous fit with SR Question about comparison w/ VBFNLO expectation Need to develop control sample and explore Z contribution from Sherpa, scaled to VBFNLO (via MadGraph) in relevant kinematic region Big difference between VBFNLO and Sherpa not understood (Sherpa much larger) Need to revisit
May TASKS POTENTIALLY COMMON TO ALL GROUPS SM MC samples xAOD framework issues (not completely sure what I mean by this…) Code snippets reflecting agreed-upon object definitions Isolation variable code Pseudophoton object definition code e->gamma fake rate study dPhi studies (one-sided or two-sided?) MAYBE? Common derivation (probably not since different triggers)? Common systematics tasks?
March Code/Infrastructure xAOD-based analysis: TokyoTech, UCSC need to catch up Derivations followed through upon by Milano (status?) Higher-level statistics and plotting utility (Ryan…) Past quantities that have required study (do we need to look into these?) MET Isolation definition ???
March Event Selection: Preliminary Studies In past, formal optimization was last step, considering only M_eff (or H T ), MET Individual, preliminary studies used to establish Photon P T cut; see e.g. Δφ -MET : make use of or not; cut value. Should we also cut on (Δφ -MET - )? Δφ jet-MET : cut value. Should we also cut on (Δφ jet-MET - )? For 8 TeV, used M eff vs. MET visualization plane (see below) Will need signal grid points for this already!
March Optimization: 8 TeV Approach Last step done by inspection of M eff (or H T ) vs MET plane Can be confounded by statistics; also look at background and signal stats over same plane See 8 TeV backup note YES NO WP2 Optimization
March Optimization: The Conundrum How to estimate backgrounds when final background estimates not available? For 8 TeV analysis optimization, backgrounds estimated by QCD background estimated by scaling 1 tight + 1 non- isolated pseudophoton sample to 2 tight pseudophoton sample with no M eff (H T ) cut for 0 < 60 < MET (DATA) EW background estimated by scaling e sample by uniform 2% e scale factor (DATA) W , Z from MC SUSY group will accept leaving final data-driven step and quick reoptimization before unblinding. Or, pre-optimize as a function of one to-be-determined background value
March What SRs to Create? For 8 TeV Analysis Strong production: High M eff ; backgrounds near 0 EW production: Intermediate H T ; backgrounds 1-2 events Low mass bino, high mass bino for both SP1, SP2, WP1, WP2 Also: Model-independent SR (MIS), no M eff (H T ) cut. Based on choosing MET cut at which EW and QCD backgrounds about the same (~1 event each)
March Model-Independent SR (?) 8 TeV analysis: at MET=250, M eff = 0 backgrounds about same EW QCD Question: Should we rethink? What do we really want to do to minimize chance that we miss a signal? Hmmm…. How do we think about this?
March What Physics Could Hide Signal with Dominant BF into Photons and DM? Degenerate SUSY scenarios? No – energy has to go somewhere. We would see it in photons and/or MET. Photons will not be soft because decaying state will either be high-mass or boosted. Low photonic BF? Would need to accelerate single-photon analyses. Not really practical. Long-lived scenarios? Need to re-create non-pointing photon reconstruction. Probably no competition from CMS here anyway. Perhaps most likely scenario is lower-than-expected cross section from non-SUSY process. Probably best addressed by what was done before, or perhaps just use no M eff or H t cut and use lower MET cut of the other, model-dependent SRs. Could perhaps also maintain low photon E T cut but that could be a “can of worms”.
March Wrap UP I haven’t mentioned limit setting within HistFitter Immediate motivation is to get to unblind before or simultaneous with CMS I’m not assuming we’ll necessary be setting limits! Our work is cut out for us. Thoughts? We should start writing the skeleton of the backup note. If anyone is itching to do this, by all means. Otherwise, I’m very happy to do that.