H → γγ Workshop on Photon Physics and Simulation at Hadron Colliders 2019 Ruggero Turra on behalf of the ATLAS and CMS Collaboration INFN Milano 7 May 2019
Table of contents 1 Higgs boson and its properties 2 ATLAS and CMS analyses 3 Conclusions R.Turra (INFN Milano) H → γγ 7 May 2019 2 / 39
Recent H → γγ analyses Date 1 Collaboration Topic Luminosity link 36 fb − 1 2018/6 ATLAS Mass Phys. Lett. B 784 (2018) 345 36 fb − 1 2017/11 ATLAS EFT from STXS ATL-PHYS-PUB-2017-018 36 fb − 1 2018/2 ATLAS STXS + fiducial/diff Phys. Rev. D 98 (2018) 052005 80 fb − 1 2018/7 ATLAS STXS + fiducial/diff ATLAS-CONF-2018-028 36 fb − 1 2018/4 CMS coupling JHEP 11 (2018) 185 36 fb − 1 2018/7 CMS fiducial/diff JHEP 01 (2019) 183 77 fb − 1 2019/3 CMS STXS (ggF and VBF) CMS-PAS-HIG-18-029 140 fb − 1 2019/4 ATLAS ttH ATLAS-CONF-2019-004 77 fb − 1 2018/10 CMS ttH CMS-PAS-HIG-18-018 1 of the preprint R.Turra (INFN Milano) H → γγ 7 May 2019 3 / 39
Section 1 Higgs boson and its properties
Higgs properties Excess compatible with Higgs boson firmly established by ATLAS+CMS in 2012. Measurements Mass: m H known at 0.2% (single experiment) σ × Br : inclusive, for each production-mode, fiducial region (STXS) (very optimized on the SM, acceptance extrapolations, model dependent) Fiducial cross sections or differential cross sections in fiducial regions (minimal model dependence) Interpretations Spin and parity: 0 + , other models excluded in Run 1. Signal strengths: µ i = σ i /σ SM (inclusive, per-production-mode, . . . ) i Coupling modifiers to SM particles (k-framework) EFT interpretations, CP, . . . R.Turra (INFN Milano) H → γγ 7 May 2019 5 / 39
Higgs properties Excess compatible with Higgs boson firmly established by ATLAS+CMS in 2012. Measurements Mass: m H known at 0.2% (single experiment) σ × Br : inclusive, for each production-mode, fiducial region (STXS) (very optimized on the SM, acceptance extrapolations, model dependent) Fiducial cross sections or differential cross sections in fiducial regions (minimal model dependence) Interpretations Spin and parity: 0 + , other models excluded in Run 1. Signal strengths: µ i = σ i /σ SM (inclusive, per-production-mode, . . . ) i Coupling modifiers to SM particles (k-framework) EFT interpretations, CP, . . . R.Turra (INFN Milano) H → γγ 7 May 2019 5 / 39
Simplified template cross sections (STXS) stage 1 Yellow Report 4 - CERN-2017-002-M Exclusive fiducial regions defined by production mode, p H T , N j , VBF-topology, p j 1 T , p Hjj T , p V T STXS bins interpr. Design measurement to split events according to STXS R.Turra (INFN Milano) H → γγ 7 May 2019 6 / 39
H → γγ Small Br: 2 . 27 × 10 − 3 Loop decay: sensitive to BSM Simple final state: good resolution (1 . 4–2 . 1 GeV), good efficiency ( ≃ 40%) Very large background q ¯ q / gg → γγ and fakes Falling background, can be modelled fitting data m γγ sidebands R.Turra (INFN Milano) H → γγ 7 May 2019 7 / 39
Section 2 ATLAS and CMS analyses
General strategy for H → γγ Shape ( m γγ ) analysis in categories of selected events Use fit to m γγ to extract signal and bkg yields in selected sample Modeling background m γγ distribution with analytical functions Signal shape from MC, as double sided Crystall Ball (ATLAS) or sum of Gaussians (CMS) Photons are selected using shower shapes and isolation: ATLAS: rectangular cuts (tight selection) CMS: BDT, usually used as a continous variable Extract signal in different categories pure of events under study (production mode, STXS, kinematic bin, . . . ) Categories defined from properies of selected objects (kinematic, identification, quality, . . . ) CMS has a predictor of the expected resolution R.Turra (INFN Milano) H → γγ 7 May 2019 9 / 39
Coupling categorization in ATLAS (STXS) ATLAS-CONF-2018-028 To measure many cross-sections with small correlation split events in pure categories ttH VH VBF 29 reco-categories inspired by STXS ggF prod modes R.Turra (INFN Milano) H → γγ 7 May 2019 10 / 39
Categorization in CMS (ggF and VBF STXS) CMS-PAS-HIG-18-029 Diphoton-BDT using photon kinematic, BDT-id score, mass resolution, vertex probability VBF categorization: dijet-BDT trained with signal: VBF, background: ggF+jets and non-Higgs (from control region inverting photon id score) 6 categories using dijet ⊗ diphoton-BDT and p Hjj T , m jj , p j 1 T mimic STXS (2J, 3J, BSM, rest) ggF categorization: mimic STXS using p γγ and number of jets. Split also by diphoton BDT T (“Tag”). All BDT validated on Z → ee R.Turra (INFN Milano) H → γγ 7 May 2019 11 / 39
Categorization in CMS (ggF and VBF STXS) → γ γ CMS Simulation Preliminary H 13 TeV (2017) 100 Category signal composition (%) Event category VBF BSM 2 2 37 2 2 5 3 1 39 7 VBF rest 1 2 3 2 8 16 18 8 1 2 2 1 26 7 VBF 3J-like Tag1 1 3 5 2 3 7 4 3 6 21 13 16 8 1 6 VBF 3J-like Tag0 1 2 1 1 1 3 7 5 15 26 26 7 2 3 VBF 2J-like Tag1 80 2 3 4 1 2 4 2 1 18 5 40 4 13 2 VBF 2J-like Tag0 1 3 1 1 1 1 1 13 3 58 7 9 1 1 2J BSM Tag1 3 2 56 1 1 2 5 1 30 2J BSM Tag0 4 1 62 1 2 5 1 24 2J high Tag1 7 1 56 1 1 3 1 1 5 1 1 22 2J high Tag0 60 7 59 2 2 1 1 5 1 1 20 2J med Tag1 1 16 1 54 1 2 1 1 6 1 16 2J med Tag0 14 59 2 2 1 1 5 1 15 2J low Tag1 15 19 39 1 1 2 1 1 6 1 14 2J low Tag0 16 19 41 1 1 2 1 1 5 13 40 1J BSM 1 48 18 2 1 2 14 12 1J high Tag1 2 49 17 2 2 5 2 10 9 1J high Tag0 51 1 15 1 2 6 2 10 1 9 1J med Tag1 2 2 65 12 1 1 2 1 8 6 1J med Tag0 1 1 63 1 14 1 1 3 1 8 6 20 1J low Tag1 27 54 2 5 1 1 1 5 5 1J low Tag0 27 55 1 6 1 1 5 5 0J Tag2 83 8 2 1 2 3 0J Tag1 78 10 4 1 1 1 2 3 0J Tag0 73 11 5 1 1 1 3 4 0 ggH 0J ggH 1J med ggH 1J high ggH 1J BSM ggH 2J med ggH 2J high ggH 2J BSM ggH VBF-like 2J ggH VBF-like 3J VBF 2J-like VBF 3J-like VBF rest VBF BSM VBF VH-like Other ggH 1J low ggH 2J low STXS process R.Turra (INFN Milano) H → γγ 7 May 2019 12 / 39
ttH -categories ATLAS-CONF-2019-004 ATLAS, great improvement using agressive optimization: Leptonic ( ≥ 1 ℓ , ≥ 1 b ), hadronic ( ≥ 1 b , ≥ 2 j , 0 ℓ ) regions Train two BDT with low-level variables: p T / m γγ , η, φ of photons, p µ of up to two leptons, p µ of up to four/six jets (lep/had, p T sorted), magnitude and φ of E T -miss Categories defined from the BDT output Trained on MC ttH signal and control regions (non-tight non-isolated) for background CMS: Leptonic ( ≥ 1 ℓ , ≥ 1 b ), hadronic ( ≥ 2 j , 0 ℓ ) regions Train two BDT with: p T / m γγ , η , BDT-id of photons, ∆ φ γγ or φ , p T and η of diphoton (had only), number of (b)-jets, p T and η of the first three (four) jets ( p T ordering), � all jets p T (had), b -discriminant, p T and η of the lepton, E T -miss Categories defined from the numbers of leptons (1/2) and BDT output Trained on MC ttH signal, bkg MC, ggF+VBF MC (only had) R.Turra (INFN Milano) H → γγ 7 May 2019 13 / 39
ttH -categories Main problem: theoretical systematic for ggF contamination (ggF+HF) ATLAS: 100% uncertainty on ggF, VBF, VH production (supported by H → 4 ℓ , ttbb , Vb ). Impact 3-4% CMS (only ggF): parton shower (from difference in jet multiplicity aMCNLO/data tt+j), gluon splitting (scaling the fraction of events from ggF+b in simulation by the measured difference data/simulation of σ ( ttbb ) /σ ( ttjj )). Impact 2%. Better recipe? R.Turra (INFN Milano) H → γγ 7 May 2019 14 / 39
Background modeling in ATLAS Background m γγ distributions are fitted directly on data with simple analytical functions both in ATLAS and CMS Different approach how the functional form (exponential, . . . ) are selected In ATLAS, for each category, one functional form is selected by dedicated studies on MC or control regions Due to the little s / b , small mismodeling on the shape can bias the signal yield Quantify bias using closure tests injecting 0 signal events: spurious signal Select functional form which pass criteria based on the spurious signal R.Turra (INFN Milano) H → γγ 7 May 2019 15 / 39
Spurious signal (ATLAS) Build a background-only template, usually mixing γγ from Sherpa NLO and γ j from control region Run a signal + background fit The spurious signal (the bias) is the number of fitted events (positive or negative) Assume the systematic on the signal yield to be the max m H | bias | changing m H in a window Problems: MC is not data: need to try different variations (e.g. γγ purity) MC is limited: statistical fluctuation in the MC can increase the estimated spurious signal Limiting factor of the procedure: not scalable Produce faster MC (e.g. smeared truth MC, LO generators): bilions of events Use more flexible function (even not analytic) so that you expect better modeling (how to validate?) Remove statistical fluctuation from your template R.Turra (INFN Milano) H → γγ 7 May 2019 16 / 39
Recommend
More recommend
