Search For Higgs Boson Pair Production in the gamma gamma b bbar Final State using pp Collision Data at sqrt(s)=8 TeV from the ATLAS Detector

Searches are performed for resonant and non-resonant Higgs boson pair production in the γγbb ﬁnal state using 20 fb -1 of proton–proton collisions at a center-of-mass energy of 8 TeV recorded with the ATLAS detector at the CERN Large Hadron Collider. A 95% conﬁdence level upper limit on the cross section times branching ratio of non-resonant production is set at 2.2 pb, while the expected limit is 1.0 pb. The difference derives from a modest excess of events, corresponding to 2.4 standard deviations from the background-only hypothesis. The limit observed in the search for a narrow X → hh resonance ranges between 0.7 and 3.5 pb as a function of the resonance mass.

Search for Higgs Boson Pair Production in the γγbb Final State using pp Collision Data at √ s = 8 TeV from the ATLAS Detector Searches are performed for resonant and non-resonant Higgs boson pair production in the hh → γγbb final state using 20 fb -1 of proton-proton collisions at a center-of-mass energy of 8 TeV recorded with the ATLAS detector at the CERN Large Hadron Collider. A 95% confidence level upper limit on the cross section times branching ratio of non-resonant production is set at 2.2 pb, while the expected limit is 1.0 pb. The corresponding limit observed for a narrow resonance ranges between 0.8 and 3.5 pb as a function of its mass.
PACS numbers: 12.60.Fr, 14.80. Da,14.80.Ec Within two years of discovering a new boson with a mass near 125 GeV [1,2], the ATLAS and CMS collaborations have completed a slate of measurements demonstrating that its spin and couplings conform to the predictions of the Standard Model (SM) Higgs boson within current experimental and theoretical uncertainties [3,4]. Despite the lack of deviations from SM predictions, the Higgs boson, h, offers a rich potential for new physics searches. This Letter reports on searches for non-SM physics with events consistent with either resonant (X → hh) or non-resonant pair production of Higgs bosons in the hh → γγbb channel.
The predicted rate for Higgs boson pair production in the SM is several orders of magnitude smaller than the rate for the single h process [5][6][7][8]; hh production is thus not expected to be observable with current LHC data sets. However, a variety of extensions to the SM would enhance Higgs boson pair production. In many two Higgs doublet models (2HDMs) [9][10][11] the heavier of the neutral scalar Higgs bosons, H, may decay to a pair of its lighter scalar partners, h. Depending on the parameters of the 2HDM, H → hh production cross sections may reach more than a picobarn [11]. Altering the Higgs boson self-coupling, λ hhh , could also increase the non-resonant production rate. Such deviations could be observed with future data sets [8]. Larger enhancements in the pp → hh rate could arise from a direct tthh vertex, which is natural in composite models [12,13], or from the addition of light colored scalars to the SM [14]. Resonant production of two Higgs bosons could also appear from the production and decay of gravitons, radions or stoponium [15][16][17], as well as from a hidden sector mixing with the observed Higgs boson [18].
The γγbb channel is a powerful final state in which to search for Higgs boson pair production [19] thanks to the large h → bb branching ratio, clean diphoton trigger, excellent diphoton invariant mass resolution, and low backgrounds. This channel is particularly important in the search for resonances with mass, m X , in the range 260 < m X < 500 GeV considered in this Letter, where backgrounds and combinatorics make other channels such as bbbb or bbτ + τ − challenging.
Processes that do not contain Higgs bosons are estimated from data; all other processes are simulated using Monte Carlo techniques. The standard ATLAS detector simulation [20] based on Geant4 [21] is used. The simula-tion parameters are tuned to describe soft components of hadronic final states [22,23]. Simulated minimum bias collisions are overlaid on the hard scatter process, and events are reweighted so that the average number of interactions per bunch crossing ( ∼ 20) matches the observed distribution.
Background events with a single Higgs boson produced in association with a W or Z boson or tt are simulated with Pythia8 [24] using CTEQ6L1 parton distribution functions of the proton (PDFs) [25]. Higgs boson production via gluon or vector-boson fusion is simulated using CT10 PDFs [26] with Powheg-Box [27,28] interfaced to Pythia8 for showering and hadronization. Cross sections and associated uncertainties are taken from Ref. [29].
Two benchmark signal models are defined: SM Higgs boson pair production for the non-resonance search, and a gluon-initiated, spin-0 resonant state in the narrowwidth approximation for the resonance search. Both models are generated using MadGraph5 [30,31] and CTEQ6L1 PDFs. A generator filter requires a pair of b-quarks and a pair of photons in each event. Pythia8 is used to decay the two Higgs bosons, and to shower and hadronize the events. The implementation of SM Higgs boson pair production includes the interference between diagrams with trilinear Higgs boson couplings and box diagrams. The next-to-leading-order inclusive production cross section of 9.2 fb is taken from Ref. [8]. Resonant samples are generated with a width of 10 MeV (corresponding to a narrow width approximation) at masses m X = 260, 300, 350, and 500 GeV. Production cross sections for benchmark 2HDM models are calculated with SusHi [32], and branching ratios with 2HDMC [33].
The analysis described in this Letter uses the full √ s = 8 TeV data set of proton-proton collisions recorded by the ATLAS experiment in 2012, corresponding to an integrated luminosity of 20.3 ± 0.6 fb -1 [34]. Data quality criteria are applied to reject events with diminished detector performance [35,36]. A description of the ATLAS detector can be found elsewhere [37].
The photon and event selection for the present search largely follow those of existing ATLAS h → γγ analyses [3,38]. Events are selected using a loose diphoton trigger that is nearly 100% efficient for events passing the offline photon selection. Photons are reconstructed starting from clusters of energy deposited in the electromagnetic calorimeter. Events are required to contain two photon candidates whose calorimeter energy clusters match the expectations for photon-induced electromagnetic showers [39,40]. The pseudorapidity [41], η, of the two photons must fall within the geometric acceptance of the detector for photons, |η| < 2.37, excluding the region between the barrel and endcap calorimeters (1.37 < |η| < 1.56). The ratio of the transverse momentum of the leading (subleading) photon to the invariant mass of the pair, p T /m γγ , must exceed 0.35 (0.25). The invariant mass of the pair is calculated as in Ref. [3]. Photons are required to be isolated: the energy in the calorimeter [3,42] within a cone of size ∆R ≡ ∆η 2 + ∆φ 2 = 0.4 around the photon direction must be less than 6 GeV, and the scalar sum of the p T of the tracks in a cone of ∆R = 0.2 must be less than 2.6 GeV. In addition, the photon pair must satisfy 105 < m γγ < 160 GeV for an event to be considered.
Jets are reconstructed from clusters of energy in the electromagnetic and hadronic calorimeters using the antik t algorithm [43] with a radius parameter of 0.4, starting from energy deposits grouped into topological clusters [44]. Simulation is used to correct jets for instrumental effects [45] and to account for the average energy in the detector in the event due to additional pp interactions and the underlying event [46]. The calibration is refined using in situ measurements. Jets are required to fall within the tracker acceptance of |η| < 2.5 and satisfy p T > 35 GeV, with the leading jet in the event required to have p T > 55 GeV. Events with jets arising from noisy regions in the calorimeters, beam backgrounds, or cosmic rays are rejected [45]. Low-p T jets from additional proton-proton interactions in the same bunch crossing are rejected with a requirement on the scalar sum of the p T of tracks associated with the jet: for jets with |η| < 2.4 and p T < 50 GeV, tracks associated with the hard scatter vertex must contribute over 50% of the sum.
Jets from the decay of long-lived heavy-flavor hadrons are selected using a multivariate tagging algorithm (btagging) [47] with an efficiency of 70% for jets from bquark fragmentation in tt simulation. The four-momenta of muons [48] closer than ∆R = 0.4 to a b-tagged jet and with p T > 4 GeV are included in the jet four-momentum.
Events containing both a pair of photons and two jets with dijet mass 95 < m jj < 135 GeV, consistent with the decay of a Higgs boson, are kept for further study. The mass resolution for pairs of b-tagged jets is approximately 13 GeV; it is asymmetric since neutrinos from semileptonic b-decays are not measured.
In the non-resonance search, the background and potential signal are fit to the unbinned m γγ distribution of all events passing the dijet and diphoton selections described above. This fit has three components: the signal with a pair of Higgs bosons, the SM single Higgs boson background that is resonant in m γγ , and the continuum background. The combined acceptance and selection efficiency for the SM Higgs boson pair production signal is 7.4%. Simulation studies show that the continuum contribution in the signal region is split between events with two photons and events with a single photon in associa- (Upper plot) Diphoton invariant mass spectrum for data and the corresponding fitted signal and background in the signal region for the non-resonance search. (Lower plot) The diphoton invariant mass spectrum in the continuum background from events with fewer than two b-tags and the corresponding fitted curve, the shape of which is also used in the upper plot.
tion with a jet faking the second photon. The b-tagged jets include real heavy-flavor jets and mis-tagged lightflavor jets. The contribution from dileptonic decays of tt events where two electrons fake the two photons is roughly 10% of the total background. The contribution from other processes is negligible.
The fit is performed simultaneously in two categories. The first category is the signal region, in which at least two jets are b-tagged. The second is a control region, containing events with fewer than two b-tags. The two classes of events are kinematically identical: in the signal region, the mass and p T requirements defined above must be satisfied by the two leading tagged jets, whereas in the control region, they are met by the two leading jets.
Following earlier ATLAS analyses, the shape of the m γγ resonance is described by the sum of a Crystal Ball function and a wide Gaussian component that models the tails of the distribution [3]. An exponential function describes the continuum backgrounds that fall with m γγ . The slope of the exponential is shared in the fit between the two categories so that the control region constrains the background shape in the signal region. Figure 1 shows the separate diphoton mass distributions for events with ≥ 2 b-tags and events with ≤ 1 b-tag.
The search for resonant production of pairs of Higgs bosons starts with the same signal selection as above but imposes an additional requirement on m γγbb . Due to the small number of expected events after this additional requirement, the resonance analysis proceeds as a counting experiment instead of a simultaneous fit. The m γγ resolution, σ mγγ , is set to the expected value of 1.6 GeV, and the diphoton mass is required to be within ±2σ mγγ of the Higgs boson mass, m h = 125.5GeV [3]. The acceptance of this requirement on background events without Higgs bosons, mγγ , is measured by fitting an exponential function to the m γγ sidebands for events with fewer than two b-tagged jets. For this fit, the m γγ region of m h ±5 GeV is excluded to eliminate any potential contamination from resonant Higgs boson production. For N observed events with two b-tags in the sideband (|m γγ − m h | > 2σ mγγ ), the number of expected non-Higgs boson background events (N mγγ ) within 2σ mγγ around m h is given by: where the denominator compensates for the fact that mγγ = 0.13 is derived relative to the full m γγ spectrum while N contains only those events in the sidebands. Before reconstructing the four-object mass, m γγbb , a scaling factor of m h /m bb is applied to the fourmomentum of the bb system, where m h is set to the value of 125 GeV used in simulation. This improves the m γγbb resolution by 30-60% depending on the mass hypothesis, without biasing or significantly altering the shape of the background. Requirements are then made on m γγbb , to select the smallest window containing 95% of the signal events in the narrow-width simulation. These requirements vary linearly with the mass, m X , of the resonance considered. The width of the signal window varies from 17 GeV at m X = 260 GeV to 60 GeV at m X = 500 GeV. The acceptance for the continuum background to pass this requirement, m γγbb , also varies with m X . It is measured using events in data with |m γγ − m h | < 2σ mγγ and fewer than two b-tags. Studies in both data sidebands and simulation show that the shapes of m γγbb and m γγjj agree within statistical uncertainties. The distribution of m γγjj in data is fitted with a Landau function, which is integrated in the signal window to obtain m γγbb for each mass hypothesis. The bottom panel of Fig. 2 shows this fit. The value of m γγbb is small (< 8%) at low and high m X , and peaks at 18% for m X = 300 GeV. The combined acceptance and selection efficiency for a resonance signal to pass all requirements varies from 3.8% at m X = 260 GeV to 8.2% at m X = 500 GeV.
The total background from sources without Higgs boson decays in the resonance analysis N B is given by: where N B and m γγbb are functions of m X . Uncertainties on this extrapolation are described below. Because they are not accounted for by the above m γγ sideband techniques, contributions from single Higgs bosons produced in association with jets (particularly with cc or bb pairs) are estimated using simulation. In the resonance analysis, the yield from the non-resonant SM hh processes is similarly included. SM cross sections and branching fractions are assumed in all cases [29]. (Upper plot) The constrained four-object invariant mass, mγγjj, for data events in the resonance signal region. The expected backgrounds are also shown. A narrow width resonance at 300 GeV is displayed for comparison only. (Lower plot) The diphoton invariant mass spectrum in the continuum background from events with fewer than two btags and the corresponding fitted curve, the shape of which is also used in the upper plot.
Most systematic uncertainties are small when compared to statistical uncertainties, in particular for the resonance search.
The evaluation of experimental uncertainties on photon identification (2.4%) and isolation efficiencies (2%) follow the methods used in the inclusive ATLAS h → γγ analyses [3,29,38]. The theoretical uncertainties on the single Higgs boson backgrounds are similarly adopted. Uncertainties from the modeling of Higgs bosons produced in association with extra heavy-flavor quarks are estimated from comparisons of data to simulation predictions for both tt [49] and W boson [50] production with extra heavy-flavor jets. The tt (W boson) study provides the uncertainty on extra heavy-flavor content in gluon-(quark-)initiated final states, leading to a 14% uncertainty on the combined single Higgs boson contribution in the signal region. PDF and scale uncertainties on SM hh production are taken from Ref. [8].
Because of the cuts on the ratio p T /m γγ , photon energy scale uncertainties are negligible. The uncertainty of 13% on the diphoton mass resolution in the nonresonance search is implemented in the resonance analysis as a 1.6% uncertainty on the number of events migrating into and out of the signal region. This represents the fraction of events where an upward variation of the photon resolution causes the diphoton mass to leave the m h ± 2σ mγγ window required for the signal region. The uncertainty on m h impacts the peak position in m γγ in the signal plus background fit of the non-resonance analysis, and in the resonance search it is transformed into a 1.7% uncertainty on the number of signal events in the mass window. The uncertainty for the acceptance of the m γγ cuts on non-Higgs boson backgrounds is estimated by comparing fits of m γγ to data in control regions with reversed photon identification or b-tagging requirements, and using different functional forms. The largest deviation observed from these fits (11%) is used for all searches.
Three components contribute to the uncertainty on m γγbb , and are combined in quadrature. (1) The limited number of events in the control region with fewer than two b-tags used for the Landau fit lead to a relative statistical uncertainty of 3-18% that varies as a function of m X . (2) The m γγjj shape for untagged jets might not exactly mirror the one for tagged jets. The tagged and untagged samples are compared in simulation and the relative difference in m γγbb is taken as the uncertainty. This value varies with m X and is always less than 30%. (3) Finally, an uncertainty of 16-30%, depending on m X , is included to cover the choice of the analytic function. This was evaluated via comparisons of Landau shapes to alternate functions in simulation, including Landau shapes where the width varies with m γγbb , as well as Crystal Ball functions. Potential contamination from single Higgs boson processes in the control region is estimated to be less than 4% and is subtracted with negligible impact on the shape. Uncertainties due to the b-tagging calibration are typically 2-4% for both the single Higgs boson and signal processes. Uncertainties due to the jet energy scale are 7% (22%) for single Higgs boson backgrounds in the nonresonance (resonant) analysis, and 1.4% (4.4%) for signal processes. Uncertainties due to the jet energy resolution are 4.8% (21%) for single Higgs boson backgrounds, and 6.3% (9.3%) for signal processes. The uncertainty on the integrated luminosity is 2.8%. It is derived, following the same methodology as that detailed in Ref. [34], from a preliminary calibration of the luminosity scale derived from beam-separation scans performed in November 2012.
The combined signal plus background fit for the nonresonance analysis is shown in Fig. 1. Within a ±2σ mγγ window around the Higgs boson mass, 1.5 events are expected, with 1.3 from the continuum background and 0.2 from single Higgs boson production, which is dominated by tth events. About 0.04 events are expected from SM Higgs boson pair production. Five events are observed, corresponding to 2.4σ from the backgroundonly hypothesis. The 95% confidence level (CL) upper limit on the Higgs boson pair production cross section is calculated using the frequentist CL S method [51]. Exclusions and significances are evaluated using pseudoexperiments. Assuming SM branching ratios for the light Higgs boson decays, the expected upper limit is 1.0 +0.6 −0.3 pb; the observed limit is 2.2 pb.
For the resonance analysis, as before, SM branching fractions for the light Higgs boson are assumed. The expected exclusion improves from 1.8 to 0.8 pb as a function of m X from 260 to 500 GeV, as shown in Fig. 3. This behavior derives from increased event-level acceptance at larger masses. The observed exclusion ranges from 3.5 to 0.8 pb. The five events selected in the m γγ signal region are shown in m γγbb , in Fig. 2. The local probability of compatibility to the background-only hypothesis, p 0 , reaches a minimum of 0.002 at m X = 300 GeV, corresponding to 3.0σ. After accounting for the lookelsewhere effect [52,53], the global probability of such an excess occurring at any mass in the range studied is 0.019, corresponding to 2.1σ. The number of events lying within the m γγbb window of each mass hypothesis is readily apparent in 'steps' in the exclusion plot.
The limits derived are juxtaposed in Fig. 3 with the expectation from a sample type I 2HDM [32,33,54] not excluded by current data with cos(β − α) = −0.05 and tan(β) = 1. The heavy Higgs bosons are taken to be degenerate in mass, and the mass of the lightest CPeven Higgs boson is set to 125 GeV. All major production mechanisms of H → hh are considered. Cross sections and branching ratios were calculated as described in Ref. [55].
In conclusion, this Letter presents searches for resonant and non-resonant Higgs boson pair production using 20.3 fb -1 of proton-proton collision data at √ s = 8 TeV generated by the Large Hadron Collider and recorded by the ATLAS detector in 2012. A 95% confidence level upper limit is placed on the non-resonant production cross section at 2.2 pb, while the expected limit is 1.0 +0.6 −0.3 pb. The difference derives from a small excess of events, corresponding to 2.4σ.
In the search for a narrow resonance decaying to a pair of Higgs bosons, the expected exclusion on the production cross section falls from 1.8 pb for a resonance at 260 GeV to 0.8 pb at 500 GeV. The observed exclusion ranges from 0.8-3.5 pb. It is weaker than expected for resonances below 350 GeV.
We thank CERN for the very successful operation of the LHC, as well as the support staff from our institutions without whom ATLAS could not be operated efficiently. We