Search for top quark decays via Higgs-boson-mediated flavor-changing neutral currents in pp collisions at sqrt(s) = 8 TeV

A search is performed for Higgs-boson-mediated flavor-changing neutral currents in the decays of top quarks. The search is based on proton-proton collision data corresponding to an integrated luminosity of 19.7 inverse femtobarns at a center-of-mass energy of 8 TeV collected with the CMS detector at the LHC. Events in which a top quark pair is produced with one top quark decaying into a charm or up quark and a Higgs boson (H), and the other top quark decaying into a bottom quark and a W boson are selected. The Higgs boson in these events is assumed to subsequently decay into either dibosons or difermions. No significant excess is observed above the expected standard model background, and an upper limit at the 95% confidence level is set on the branching fraction B(t to Hc) of 0.40% and B(t to Hu) of 0.55%, where the expected upper limits are 0.43% and 0.40%, respectively. These results correspond to upper limits on the square of the flavor-changing Higgs boson Yukawa couplings |lambda[tc]^H|^2<6.9E-3 and |lambda[tu]^H|^2<9.8E-3.


Introduction
With the discovery of the Higgs boson (H) [1][2][3] it is possible to probe new physics by measuring its coupling to other particles. Of particular interest is the flavor-changing neutral current (FCNC) decay of the top quark to the Higgs boson. The investigation of this process at the CERN LHC is motivated by the large tt production cross section and the variety of possible decay modes of the Higgs boson.
The next-to-next-to-leading-order tt production cross section at a center-of-mass energy of 8 TeV and with a top quark mass (m t ) of 173.5 GeV [4] is 252 pb [5]. The standard model (SM) predicts that the top quark decays with a branching fraction of nearly 100% into a bottom quark and a Wboson (t → Wb).
In the SM, FCNC decays are absent at leading-order and occur only via loop-level processes that are additionally suppressed by the Glashow-Iliopoulos-Maiani mechanism [6,7]. Because the leading-order decay rate of t → Wb is also quite large, the SM branching fraction B (t → Hq), where q is an up or charm quark, is predicted to be of O(10 −15 ) [6][7][8], far below the experimental sensitivity at the LHC. However, some extensions of the SM predict an enhanced t → Hq decay rate. Thus, observation of a large branching fraction would be clear evidence for new physics. The largest enhancement in B(t → Hq) is predicted in models that incorporate a two-Higgs doublet, where the branching fraction can be of O(10 −3 ) [8].
Previous searches for FCNC in top quark decays mediated by a Higgs boson have been performed at the LHC by ATLAS [9,10] and CMS [11]. The CMS search considered both multilepton and diphoton final states and the observed upper limit of B(t → Hc) at the 95% confidence level (CL) was determined to be 0.56%. The recent ATLAS result included final states where the Higgs boson decays to b quark pairs, and measured the observed upper limits of B(t → Hc) and B(t → Hu) at the 95% CL to be 0.46% and 0.45%, respectively.
The analysis presented here uses a data sample recorded with the CMS detector and corresponding to an integrated luminosity of 19.7 fb −1 of pp collisions at √ s = 8 TeV. The data were recorded in 2012 with instantaneous luminosities of 5-8 ×10 33 cm −2 s −1 and an average of 21 interactions per bunch crossing. The inelastic collisions that occur in addition to the hardscattering process in the same beam crossing produce mainly low-p T particles that form the so-called "pileup" background.
In this paper, the FCNC decays t → Hc and t → Hu are searched for through the processes tt → Hc + Wb or Hu + Wb. Three independent analyses are perfomed and their results are then combined. The multilepton analysis considers events with two same-sign (SS) leptons or three charged leptons (electrons or muons). This channel is sensitive to the Higgs boson decaying into WW, ZZ, or ττ which have branching fractions of 21.5%, 2.6%, and 6.3%, respectively [12]. The diphoton analysis considers events with two photons, a bottom quark, and a W boson that decays either hadronically or leptonically. The two photons in this channel are used to reconstruct the Higgs boson which decays to diphotons with B (H → γγ) = 0.23% [12]. Finally, events with at least four jets, three of which result from the hadronization of bottom quarks (b jets), and a leptonically decaying W boson are considered. The b jet + lepton channel takes advantage of the large Higgs boson branching fraction into bb pairs, B(H → bb) = 57% [13]. A summary of the enumerated final states is shown in Table 1.
The CMS detector and trigger are described in Section 2, and the event selection and reconstruction in Section 3. Section 4 then discusses the Monte Carlo (MC) simulation samples. The signal selection and background estimations for each of the three analyses are given in Section 5, and the systematic uncertainties in Section 6. Finally, the individual and combined results from the 2 The CMS detector and trigger analyses are presented in Section 7.

The CMS detector and trigger
A detailed description of the CMS detector, together with a definition of the coordinate system used and the relevant kinematic variables, can be found in Ref. [14]. The central feature of the CMS apparatus is a superconducting solenoid, 13 m in length and 6 m in diameter, which provides an axial magnetic field of 3.8 T. Within the field volume there are several particle detection systems. Charged particle trajectories are measured by silicon pixel and strip trackers, covering 0 ≤ φ ≤ 2π in azimuth and |η| < 2.5 in pseudorapidity. A lead tungstate crystal electromagnetic calorimeter (ECAL) surrounds the tracking volume. It is comprised of a barrel region |η| < 1.48 and two endcaps that extend up to |η| = 3. A brass and scintillator hadron calorimeter (HCAL) surrounds the ECAL and also covers the region |η| < 3. The forward hadron calorimeter (HF) uses steel as the absorber and quartz fibers as the sensitive material. The HF extends the calorimeter coverage to the range 3.0 < |η| < 5.2. A lead and silicon-strip preshower detector is located in front of the ECAL endcaps. Muons are identified and measured in gas-ionization detectors embedded in the steel flux-return yoke outside the solenoid. The detector is nearly hermetic, allowing momentum balance measurements in the plane transverse to the beam direction.
Depending on the final state under consideration, events are selected at the trigger level by either requiring at least two leptons, (ee, µµ or eµ), at least two photons, or a single lepton (e or µ) to be within the detector acceptance and to pass loose identification and kinematic requirements.
The dilepton triggers used in the multilepton selection require one lepton with p T > 17 GeV and one lepton with p T > 8 GeV. At the trigger level and during the offline selection, electrons are required to be within |η| < 2.5, and muons are required to be within |η| < 2.4. All leptons must be isolated, as described in Section 3, and have p T > 20 GeV for the highest-p T lepton, and p T > 10 GeV for all subsequent leptons in the event. For events satisfying the full multilepton selection, the dimuon, dielectron, and electron-muon trigger efficiencies are measured to be 98%, 91%, and 94%, respectively, for the SS dilepton selection, and 100% for the trilepton selection.
The diphoton trigger requires the presence of one photon with p T > 36 GeV and a second photon with p T > 22 GeV. Loose isolation and shower shape requirements are applied to both photons [15]. The average diphoton trigger efficiency is measured to be 99.4% after applying the full event selection for photons within |η| < 2.5, excluding the barrel-endcap transition region 1.44 < |η| < 1.57.
The b jet + lepton selection uses the single-lepton triggers. The single-muon trigger requires at least one isolated muon with p T > 24 GeV and |η| < 2.1 to be reconstructed online. The singleelectron trigger requires at least one isolated electron with p T > 27 GeV and |η| < 2.5. The offline selection further requires that electrons have p T > 30 GeV and muons have p T > 26 GeV. This results in an average trigger efficiency of 84% for the single-electron triggers and 92% for the single-muon trigger after the b jet + lepton selection.

Event selection and reconstruction
Events are required to have a primary vertex with a reconstructed longitudinal position within 24 cm of the geometric center of the detector and a transverse position within 2 cm from the nominal interaction point. To distinguish the hard-scattering vertex from vertices arising from pileup interactions, the reconstructed vertex with the highest scalar sum of the p 2 T of its associated tracks is chosen as the primary vertex. To ensure that leptons originate from the same primary vertex, a loose requirement is applied to their longitudinal and transverse impact parameters with respect to the primary vertex.
The particle-flow event algorithm [16, 17] is used to reconstruct and identify individual particles using an optimized combination of information from the elements of the detector. Prompt electrons and muons arising from W and Z decays are typically more isolated than nonprompt leptons arising from the decay of hadrons within jets. In order to distinguish between prompt and nonprompt lepton candidates, a relative isolation parameter is defined for each lepton candidate. This is calculated by summing the p T of all charged and neutral particles reconstructed using the particle-flow algorithm within a cone of angular radius ∆R ≡ √ (∆η) 2 + (∆φ) 2 = 0.4 around the lepton candidate momentum, where ∆η and ∆φ are the pseudorapidity and azimuthal angle (in radians) differences, respectively, between the directions of the lepton and the other particle [18,19]. This cone excludes the lepton candidate and the charged particles associated with the pileup vertices. The resulting quantity is corrected for additional underlyingevent activity owing to neutral particles [3], and then divided by the lepton candidate's p T . The relative isolation parameter is required to be less than 0.15 for electrons and 0.12 for muons.
The electron selection criteria are optimized using a multivariate approach that combined information from both the tracks and ECAL clusters, and have a combined identification and isolation efficiency of approximately 60% at low p T (10 GeV) and 90% at high p T (50 GeV) for electrons from W or Z boson decays [20]. The training of the multivariate electron reconstruction is performed using simulated events, while the performance is validated using data.
Muon candidates are reconstructed with a global trajectory fit using hits in the tracker and the muon system. The efficiency for muons to pass both the identification and isolation criteria is measured from data to be larger than 95% [3,21].
For events in which there is an overlap between a muon and an electron, i.e., an electron within ∆R < 0.1 of a muon, precedence is given to the muon by vetoing the electron. In the multilepton selection, events in which there are more than three isolated leptons (electron or muon) with p T > 10 GeV are rejected to reduce diboson contamination. The invariant mass of dilepton pairs in the SS channel is required to be greater than 30 GeV in order to reject low-mass resonances and reduce poorly modeled backgrounds (e.g., QCD). In the b jet + lepton selection, events in which there are additional isolated electrons with p T > 20 GeV and |η| < 2.5 or isolated muons with p T > 10 GeV and |η| < 2.4 are rejected.
The photon energy is reconstructed from the sum of signals in the ECAL crystals [15]. The ECAL signals are calibrated [22], and a multivariate regression, developed for a previous H → γγ analysis [23], is used to estimate the energy of the photon. Clusters are formed from the neighboring ECAL crystals seeded around local maxima of energy deposits, and the collection of clusters that contain the energy of a photon or an electron is called a supercluster. Identification criteria are applied to distinguish photons from jets and electrons. The observables used in the photon identification criteria are the isolation variables, the ratio of the energy in the HCAL towers behind the supercluster to the electromagnetic energy in the supercluster, the transverse width in η of the electromagnetic shower, and the number of charged tracks matched to the supercluster. The photon efficiency identification is measured using Z → e + e − events in data by reconstructing the electron showers as photons [24], taking into account the shower shape and whether the electron probe is located in the barrel or endcap. The two highest p T photons must exceed 33 and 25 GeV, respectively.
Jets are reconstructed from the candidates produced by the particle-flow algorithm. An anti-k T clustering algorithm [25] with a distance parameter of 0.5 is used for jet reconstruction. Jets with a significant fraction of energy coming from pileup interactions or not associated with the primary vertex are rejected. Remaining pileup energy in jets is subtracted using a technique that relies on information about the jet area [26][27][28]. Reconstructed jets are calibrated to take into account differences in detector response [29]. The jets in the multilepton and b jet + lepton selections are required to have p T > 30 GeV, |η| < 2.5, and to be separated from leptons such that ∆R(lepton, jet) > 0.3. The selection of jets in the diphoton events differs by requiring the jet E T > 20 GeV and the jets be separated from both photons such that ∆R(photon, jet) > 0.3.
To characterize the amount of hadronic activity in an event, the scalar sum of the transverse energy of jets passing all of these requirements (H T ) is calculated. The missing transverse energy (E miss T ) is calculated as the magnitude of the vector sum of the transverse momenta of all reconstructed particle-flow candidates in the event.
Jets originating from the hadronization of b quarks are identified by the combined secondary vertex (CSV) b tagging algorithm [30]. The selection criteria that are used have an identification efficiency of 66%, and a misidentification rate of 18% for charm quarks and 1% for light-quark and gluon jets. The diphoton and b jet + lepton selections require b-tagged jets. Although the identification of b jets is not used to select signal events in the multilepton selection, it is used for the purpose of defining control samples to check the normalization of simulated background processes. No additional tagging is used to discriminate between jets originating from c quarks.
The inclusion of b jets in the diphoton and b jet + lepton selections results in a difference in the sensitivity to the t → Hu and t → Hc decay modes. This is caused by the larger likelihood of b tagging a jet originating from a charm quark than from an up quark. The multilepton analyses do not include b tagging to enhance the signal sensitivity so the two FCNC top quark decay modes are indistinguishable.

Simulated samples
The determination of the expected signal and background yields relies on simulated events, as well as an estimation based on control samples in data, as discussed in later sections. Samples of Drell-Yan, tt, W+jets, W + bb, diboson, tt + Z, tt + W, and triboson events are generated using the MADGRAPH event generator (v5.1.5.11) [31]. The samples of ZZ to four charged leptons and single top quark events are generated using POWHEG (v1.0 r1380) [32][33][34]. In all cases, hadronization and showering are done through PYTHIA (v6.426) [35], and τ decays are simulated using TAUOLA (v2.75) [36]. Three additional production processes are considered for the nonresonant diphoton backgrounds, where the dominant one coming from γγ + jets is simulated with SHERPA (v1.4.2) [37]. Top quark pairs with one additional photon are simulated with MADGRAPH, while those with two additional photons are simulated using the WHIZARD (v2.1.1) [38] generator interfaced with PYTHIA. The Z2 tune [39] of PYTHIA is used to model the underlying event.
Events that arise from the SM Higgs boson production are treated as a background. The gluonfusion (ggH) and vector-boson-fusion (VBF) Higgs boson production processes are generated with POWHEG at next-to-leading order (NLO) in QCD, interfaced with PYTHIA. The associated W/ZH production and ttH processes are simulated with PYTHIA at leading order. The cross sections and branching fractions of the SM Higgs boson processes are set to the values recommended by the LHC Higgs cross section working group [12].
The simulated samples for the signal process tt → Hq + Wb (q = c or u) are produced using PYTHIA for the case of the Higgs boson decaying to WW, ZZ, ττ, and γγ, and with MADGRAPH for H → bb. The use of different generators is an artifact of the various modes being analyzed separately. The Higgs boson is assumed to have a mass of 125 GeV.
The set of parton distribution functions (PDF) used is CTEQ6L [40] in all cases, except for H → bb, where CT10 [41] is used.
The CMS detector response is simulated using a GEANT4-based (v9.4) [42] model, and the events are reconstructed and analyzed using the same software used to process collision data. The effect of pileup is included in the simulation process by superimposing simulated events on the process of interest. The simulated signal events are weighted to account for the differences between data and simulation of the trigger, reconstruction, and isolation efficiencies, and the distributions of the reconstructed vertices coming from pileup. Additional corrections are applied to account for the energy scale and lepton p T resolution. The observed jet energy resolution and scale [29], top quark p T distribution [43], and b tagging efficiency and discriminator distribution [44] in data are used to correct the simulated events. Corrections accounting for the differences in lepton selection efficiencies are derived using the tag-and-probe technique [45].

Signal selection and background estimation
The sensitivity of the search is enhanced by combining the twelve exclusive channels, shown in Table 1, defined according to the expected decay modes of the Higgs and W bosons.

Multilepton channels
The multilepton analysis is conducted with the goal of enhancing the signal sensitivity in the trilepton channel: tt → Hq + Wb → ν νq + νb, and the SS dilepton channel: tt → Hq + Wb → νqqq + νb, where represents either a muon or electron. The main target of optimization is final states resulting from H → WW decays.
In the case of the trilepton channel, rejection of events containing dileptons originating from resonant Z boson production is necessary to remove backgrounds from WZ production, asymmetric internal conversions (AIC, the process in which final-state radiation in a Drell-Yan event converts to dileptons where one of the leptons carries most of the photon momentum) [46] or final-state radiation where the photon is misidentified as an electron. A comparison of the twodimensional distribution of the trilepton mass versus the opposite-sign dilepton mass is shown in Figure 1 for the estimated signal and background processes, and data. Events satisfying any of the following criteria are vetoed to reduce the contribution from resonant Z production: (1) the invariant mass of an opposite-sign, same-flavor (OSSF) lepton pair is within 15 GeV of the Z boson mass [4]; (2) the invariant mass of an OSSF lepton pair is greater than 30 GeV and the trilepton invariant mass is within 10 GeV of the Z boson mass. For the SS dielectron channel, electron pairs with an invariant mass within 15 GeV of the Z boson mass are rejected to reduce the background arising from misidentification of the electron charge. No invariant mass requirement is applied to the µ ± µ ± and e ± µ ± final states since there is a negligible contamination from resonant Z boson production.
The jet multiplicity after rejecting events containing a Z boson is shown in Figure 2. To improve the sensitivity of the search, we require at least two jets in the final state. Figure 3 shows the E miss T and H T distributions for trilepton and SS dilepton events after applying the Z veto and jet requirement. A candidate event in the trilepton channel has no additional requirements on E miss T or H T . The SS events are required to pass an E miss T -dependent H T requirement (shown in Table 2) and have E miss T greater than 30 GeV. The E miss T and H T requirements are obtained by maximizing the estimated signal significance, defined as the number of signal events over the square root of the number of background events.
The main sources of background can be divided into two categories according to the origin of the identified leptons and the E miss T . These include (1) irreducible background processes: events with leptons originating from the decay of SM bosons and having large E miss T arising from neutrinos; (2) reducible background processes: events with misidentified leptons produced either by nonprompt leptons from hadron decays (e.g., semileptonic decays of B mesons), by misidentified hadrons, or by mismeasurement of the lepton charge.
Given that at least two isolated leptons and two jets are required in the final state, the main sources of irreducible backgrounds are tt associated with vector boson production, WZ → 3 ν, ZZ → 4 , Z → 4 , and, to a lesser extent, triboson and W ± W ± production. The contribution from all of these processes except Z → 4 production are estimated from simulated samples. The WZ cross section used in the simulation is cross-checked against a control sample from data that is enriched in WZ events by requiring that there be three leptons, with two of them forming a dilepton pair whose invariant mass is consistent with a Z boson. No correction to the WZ normalization is needed. This sample is also used to assess the systematic uncertainty in the simulation of the background.
For the presentation of the results, several of the backgrounds are grouped into a single category referred to as the rare backgrounds. The rare background contribution is estimated mainly from simulation (see the following paragraph), and the processes include ZZ → 4 , tt+Z, tt+W, triboson, W ± W ± , and tt+H. The WZ → 3 ν background contribution is presented separately.
The residual contribution in the trilepton channel from asymmetric internal conversions (AIC) arising from Drell-Yan events is estimated using a data-driven technique [46] that uses Z → + − + γ events in data to model Z → + − + e/µ events. This is because the process that gives rise to the two final states is the same (final-state radiation in Drell-Yan events), and the third lepton that is detected in the AIC event carries most of the photon momentum. The + − + γ events are scaled based on photon p T -dependent weights coming from a control sample defined as having a three-body invariant mass within 15 GeV of the Z boson mass. The average conversion probabilities for photons in dimuon and dielectron events are (0.57 ± 0.07)% and (0.7 ± 0.1)%, respectively.
There are two major types of reducible backgrounds coming from bb, Drell-Yan, W+jets, and tt processes. One source comes from events with either nonprompt leptons produced during the hadronization process of the outgoing quarks (e.g., semileptonic decays of B mesons) or  Figure 3: The E miss T (top) and H T (bottom) distributions in the trilepton (left) and SS dilepton (right) channels in data (points with bars) and predicted by the SM background simulations (filled histograms) after rejecting events containing Z bosons, requiring at least two jets, and the event selection described in Section 3. The overall background uncertainty is shown in shaded black. The expected signal assuming a B(t → Hc) of 3% is shown by the unfilled histogram.

CMS
hadrons misidentified as prompt leptons. The other source originates from the charge misidentification of a lepton in the more frequent production of opposite-sign dileptons. This background mostly contaminates the SS dielectron final states. Data-driven methods are used to estimate these two types of reducible backgrounds.
Mismeasuring the charge of a lepton can be a significant source of background in SS dilepton final states when there are one or more electrons. Even though the probability for mismeasuring the charge of an electron is relatively low (≈0.1%), the production rate of opposite-sign dileptons is very high in comparison to processes that result in genuine SS dileptons. The probability of mismeasuring the charge of a muon is negligible (<10 −6 ) and is therefore not considered here. In order to estimate the probability of misidentifying the charge of an electron from data, a control sample is selected consisting of events containing a dielectron pair with an invariant mass within 15 GeV of the Z boson mass. The rate of charge misidentification is then determined from the ratio of the number of SS events to opposite-sign events as a function of p T and η. The measured charge misidentification for electrons with |η| < 1.48 is less than 0.2% for p T < 100 GeV, while for |η| > 1.48 it is 0.1% at 10 GeV and increases with p T to 2.5% at 125 GeV. These measurements are in agreement with those obtained from simulated Drell-Yan events.
Two control samples are used to estimate the misidentification rate of prompt leptons [47][48][49]: one region is enriched in bb events; the other is enriched in Z + jet production. Both samples are used to estimate the probability of misidentifying nonprompt electrons and muons as a function of p T and η. The measured misidentification rate for electrons ranges from 2% to 8% and for muons ranges from 1% to 6%. Simulated events are used to correct for the contamination arising from prompt leptons in the nonprompt misidentification rate measurement (e.g., WZ production in the Z+jet control region). The rates are then applied to events where one or more of the lepton candidates fail the tight lepton identification requirements. The differences between the nonprompt misidentification rates in the two measurement regions and the signal region are then used to estimate the systematic uncertainty of this background. To further assess the systematic uncertainty, the misidentification rates are also measured in simulated events that reproduce the background composition of events in the signal region and compared to the rates measured from data.
The predicted numbers of background and signal events for the trilepton and SS dileptons are given in Table 3. The backgrounds are separated into nonprompt lepton, charge misidentification, WZ → 3 ν, and the rare backgrounds. The predicted number of signal events assumes B(t → Hq) = 1%. The total number of observed events, also given in Table 3, is consistent with the predicted number of background events.

Diphoton channel
The diphoton analysis is performed using both leptonic and hadronic W boson decays: tt → Hq + Wb → γγq + νb, and tt → Hq + Wb → γγq + qqb. The mass of the diphoton system m γγ is the primary variable used to search for the Higgs boson decay. The contribution of the nonresonant backgrounds is estimated by fitting the m γγ distribution from data in the mass range 100 < m γγ < 180 GeV, whereas the contribution of resonant backgrounds is taken from the simulation.
The two highest-p T photons must have p T > m γγ /3 and p T > m γγ /4, respectively. The use of p T thresholds scaled by m γγ prevents a distortion of the low end of the m γγ spectrum that would result from a fixed threshold [50]. In the rare case of multiple diphoton candidates in an event, the one with the highest p T sum is selected.
The hadronic analysis uses events with at least four jets and exactly one b jet. The b jet and the three jets with the highest p T are used to reconstruct the invariant mass of the two top quarks, m jγγ and m bjj . There are three possible (m jγγ , m bjj ) pairs per event. The combination of jets with the minimum value of m jγγ /m bjj − 1 + m bjj /m jγγ − 1 is selected. The allowed ranges for m jγγ , m bjj , and the W boson mass m W associated with m bjj are obtained by maximizing the signal significance S/ √ B in the simulation, where S is the number of signal events and B is number of the background events. The background events are assumed to come from γγ+jets and are taken from simulation. The highest signal significance is found to be 16% obtained for 142 ≤ m bjj ≤ 222 GeV, 158 ≤ m jγγ ≤ 202 GeV, and 44 ≤ m W ≤ 140 GeV.
The leptonic analysis uses events with at least three jets, exactly one b jet, and at least one lepton. The reconstructed top mass m bν is found from the b jet, the lepton, and E miss T . The longitudinal momentum of the neutrino is estimated by using the W boson mass as a constraint, which leads to a quadratic equation. If the equation has a complex solution, the real part of the solution is used. If the equation has two real solutions, the one with the smaller value of m jγγ /m b ν − 1 + m b ν /m jγγ − 1 is chosen. The mass windows for m bjj , m jγγ , and m W are the same as in the hadronic channel.
The signal region is defined using the experimental width of the Higgs boson, 1.4 GeV, around the nominal mass peak position. As in the analysis of the inclusive SM Higgs boson decaying into diphotons [50], the signal shape of the diphoton invariant mass distribution is described by the sum of three Gaussian functions. Although the contribution from the SM Higgs boson background, dominated by the ttH process, is relatively small in comparison to the contribution of the nonresonant diphoton background, the resonant diphoton background cannot be ignored because it has a very similar m γγ distribution as the signal.
To determine the shape of the nonresonant diphoton background, a function consisting of a test model and the resonant diphoton background is fitted to the data under the background-only hypothesis. The model of the resonant diphoton background is the same as the signal function. The background function is used to generate 1000 pseudo-experiment samples that are fitted with the background plus signal probability density function.
A pull is then defined as (N fit − N gen )/σ N fit , where N fit is the fitted number of signal events in the pseudo-experiments, N gen is the number of generated signal events, and σ N fit is the corresponding uncertainty. In the case under consideration, N gen = 0. The procedure is verified by injecting signal in the pseudo-experiments. Several models are tried, and the chosen function for nonresonant diphoton background is the one whose bias (offset of the pull distribution) is less than 0.15 and with the minimum number of degrees of freedom for the entire set of tested models. A third-order Bernstein polynomial is selected as the functional form of the background for both the hadronic and leptonic channels. After determining the function to describe the nonresonant diphoton background, a function given by the sum of probability density functions of the resonant and nonresonant diphoton backgrounds and signal is fitted to the data. The normalization of the resonant diphoton background is allowed to vary within its uncertainties, while the normalization of the nonresonant component is unconstrained. Table 4 gives a summary of the observed and expected event yields for the two diphoton channels and Figure 4 shows the fit result overlaid with the data.

b jet + lepton channel
The basic event selection requirements for the b jet + lepton channel are a single-lepton trigger, one isolated lepton, a minimum E miss T of 30 GeV, and at least four jets, with at least three of them tagged as b jets. The background is dominated by tt → bbW + W − production. Figure 5 shows the distributions of E miss T and the W boson transverse mass (M T ) for data and simulation after the basic event selection criteria are applied. The transverse mass is defined as where p T is the p T of the lepton, E miss T is used in place of the p T of the neutrino, and ∆φ( , ν)) is the azimuthal angular difference between the directions of the lepton and neutrino. quarks. Therefore, all possible combinations are examined and a multivariate analysis (MVA) technique [51] is used to select the best candidate for each event. Several variables based on event kinematics and event topology are examined. Considering their signal-to-background separation power, the following variables are used to form a boosted decision tree (BDT) classifier [51]: • the invariant masses m Hq and m Hb of the reconstructed top quarks, • the energy of the u or c jet from the t → qH in the rest frame of its parent top quark, • the azimuthal angle between the reconstructed top quarks directions, • the azimuthal angle between the reconstructed W boson and the associated b jet directions, • the azimuthal angle between the Higgs boson and the associated jet directions, • the azimuthal angle between the directions of the b jets resulting from the Higgs boson decay.
The BDT classifier is trained with the correct and wrong combinations of simulated FCNC events determined from the generator-level parton matching. Because only event kinematics and topological variables are used, the Hu and Hc channels share the same BDT classifier. The jet-parton assignment in each event is determined by choosing the combination with the largest BDT classifier score, resulting in the correct assignment in 54% of events, as determined from simulation. The signal is determined using a template fit of the output of an artificial neural network (ANN) [51]. The ANN takes its inputs from the invariant mass of the reconstructed Higgs boson candidate and the CSV discriminator variables of the three b jets from the hadronic top quark and Higgs boson daughters. The training of the ANN is done separately for the t → Hu and t → Hc channels. A control sample dominated by tt is selected to validate the simulation used in the training. The sample is constructed by requiring one lepton and four jets, of which exactly two are b jets.
Neural network discriminator Hu fit → t Figure 6: The output distributions from the ANN discriminator for data (points) and simulated background (lines) where the ANN was trained to discriminate the backgrounds from either t → Hc (left) or t → Hu (right) decays. The solid line shows the result of the fit of the signal and background templates to data. The dotted line gives the predicted signal distribution from simulation for B(t → Hc) = 3% and the filled histogram shows the proportion of signal estimated from the fit. Figure 6 show the results of the fit performed with the 6840 observed events. The observed number of events and the expected yields of the signal and the main backgrounds estimated from simulation are shown in Table 5. The estimated background and signal based on the fit of the ANN discriminator output is shown in Table 6. The number of signal and background events from the fit result for the Hc channel are 74 ± 109 (stat) ± 24 (syst) and 6770 ± 130 (stat) ± 950 (syst), respectively. The corresponding yields for the Hu channel are 197 ± 87 (stat) ± 59 (syst) and 6640 ± 120 (stat) ± 800 (syst), respectively. Table 5: The expected number of background and signal events for the b jet + lepton selection from simulation. The signal yields from the simulation of the signal assume B(t → Hq) = 1%. Uncertainties combine both statistical and systematic components in quadrature.

Systematic uncertainties
In the fit to the data, systematic uncertainties are treated as nuisance parameters. Each of them is assigned a log-normal or Gaussian pdf, which is included into the likelihood in a frequentist manner by interpreting it as signal arising from pseudo-measurement distributions. Nuisance parameters can affect either the signal yield, the shape of kinematic variable distributions, or both. If a specific source of uncertainty is not included for a given channel, it indicates that the uncertainty is either not applicable to that channel or is found to have negligible impact on the result.
The sources of uncertainties common to all analysis channels are: the uncertainty in the total integrated luminosity (2.6%) [52]; the effects of the event pileup modeling for the signal samples (0.2-3%), which is particularly important for the b jet + lepton channel; the uncertainty in the Higgs boson branching fractions (5%) [13]; the uncertainty in the tt cross section (7.5%) [53]; the uncertainty in the jet energy scale (1-15%) [29] and resolution (0.4-8%), where the larger uncertainty is for the b jet + lepton selection; the uncertainty in the PDF used in the event generators (< 9%) [54]; the assumed top quark p T distribution (1-4%) [43]; the E miss T resolution (0.2-4%) [29]; the uncertainty in the trigger efficiency (<2%); and the corrections applied to the simulation to account for the differences in lepton identification and isolation efficiencies in data and simulation (0.01-6%), where the larger uncertainty is for the selection of events with a three-electron final state.
The uncertainties specific to the signal description and background estimation for the multilepton analysis come from the 11-13% uncertainty in the ttW and ttZ theoretical cross sections [55]; the 15% uncertainty in the WZ normalization (determined from a control region); the uncertainty in the lepton misidentification rate (40% for electrons, 30% for muons); and the 20% uncertainty in the electron charge mismeasurement probability. The uncertainties specific to the signal description and background estimation for the diphoton channels are the corrections applied to the simulation to account for differences of the photon identification efficiency in data and simulation (0.1-5%); and the uncertainty in the jet and b jet identification efficiency (2-3.5%) [30]. The resonant background from the SM Higgs boson production has an uncertainty of 8.1% from the PDF uncertainty and 9.3% from the QCD scale [56].
The uncertainties specific to the signal description and background estimation for the b jet + lepton channel are dominated by the b jet identification. The uncertainty in the b tagging correction has two components: one is from the sample purity (4%) [30] and the other from the sample statistical uncertainty (24%). The uncertainty in the tt+jets cross section, determined using a leading-order event generator, is 1%. The uncertainty in the modeling of the heavyflavor daughters of the W decay in the tt simulated sample is estimated to be 3%. Additional uncertainties arise from the event generator parameters such as the renormalization and factorization scales (5%) [41], the parton-jet matching threshold (1-9%), and the top quark mass (4%).
The uncertainties owing to the integrated luminosity, jet energy scale and resolution, pileup, reconstruction of physics objects, signal PDFs, and top quark related uncertainties are assumed to be fully correlated, while all others are treated as uncorrelated.
The systematic uncertainties are summarized in Table 7.

Results
The expected number of events from the SM background processes and the expected number of signal events in data assuming a branching fraction B(t → Hq) = 1% are shown in Tables 3, 4,  and 6 for the multilepton, diphoton, and b jet + lepton selections, respectively. The final results are based on the combination of 12 channels: three SS dilepton, four trilepton, one diphoton + hadrons, two diphoton + lepton, and two b jet + lepton. The combination requires the simultaneous fit of the data selected by all the individual analyses, accounting for all statistical and systematic uncertainties, and their correlations. As B(t → Hq) is expected to be small, the possibility of both top quarks decaying via FCNC is not considered.
No excess beyond the expected SM background is observed and upper limits at the 95% CL on the branching fractions of t → Hc and t → Hu are determined using the modified frequentist approach (asymptotic CLs method [57][58][59]). The observed 95% CL upper limits on the branching fractions B(t → Hc) and B(t → Hu) are 0.40% and 0.55%, respectively, obtained from the combined multilepton, diphoton, and b jet + lepton channels. A summary of the observed and expected limits is presented in Table 8. The diphoton channels are significantly more sensitive than the other channels, largely because of the lower uncertainty in the background model. The multilepton and b jet + lepton channels provide a 15% (37%) improvement on the observed (expected) upper limit when combined with the diphoton channel. A previous search for FCNC mediated by Higgs boson interactions via the t → Hc decay at the LHC made use of trilepton and diphoton final states [11]. The inclusion of new channels (SS dilepton, diphoton, and b jet + lepton final states) in addition to refinements in the trilepton and diphoton channels results in an improvement of 30% (34%) in the observed (expected) upper limit on B(t → Hc).
The partial width of the t → Hq process is related to the square of the Yukawa coupling λ tq by the formula [60,61]: where y = m H /m t and y q = m q /m t . (Note that a convention where the parity of the coupling is ignored is adopted here: this introduces a factor of two when comparing to the ATLAS result.) Assuming the t → Wb partial width to be dominant, the upper limit on the t → Hq branching fractions can be translated into an upper limit on the square of the couplings using the relations:

Summary
A search for flavor-changing neutral currents in the decay of a top quark to a charm or up quark and a Higgs boson based on √ s = 8 TeV proton-proton collisions has been presented. Samples of multilepton, diphoton, and b jet + lepton events were selected from data recorded with the CMS detector, corresponding to an integrated luminosity of 19.7 fb −1 . The topologies pp → tt → Hq + Wb events, where q = u, c and H is allowed to decay into WW, ZZ, ττ, γγ, and bb. No excess of events above the SM background is observed, and branching fractions of B(t → Hc) larger than 0.40% and B(t → Hu) larger than 0.55% are excluded at the 95% confidence level. These observed upper limits on B(t → Hq) and the corresponding constraints on the top quark flavor-changing Higgs boson Yukawa couplings are amongst the most stringent measured to date.   [23] CMS Collaboration, "Observation of the diphoton decay of the Higgs boson and measurement of its properties", Eur. Phys. J. C 74 (2014) 3076, doi:10.1140/epjc/s10052-014-3076-z, arXiv:1407.0558.