Search for a standard model-like Higgs boson in the mu+ mu- and e+ e- decay channels at the LHC

A search is presented for a standard model-like Higgs boson decaying to muon pair or electron pair final states based on proton-proton collisions recorded by the CMS experiment at the CERN LHC. The data correspond to integrated luminosities of 5.0 inverse femtobarns at a centre-of-mass energy of 7 TeV and 19.7 inverse femtobarns at 8 TeV for the muon search, and of 19.7 inverse femtobarns at 8 TeV for the electron search. Upper limits on the production cross section times branching fraction at the 95% confidence level are reported for Higgs boson masses in the range from 120 to 150 GeV. For a Higgs boson with a mass of 125 GeV decaying to muons, the observed (expected) upper limit on the production rate is found to be 7.4 (6.5 +2.8 -1.9) times the standard model value. This corresponds to an upper limit on the branching fraction of 0.0016. Similarly, for electrons, an upper limit of 0.0019 is placed on the branching fraction, which is about 3.7 E5 times the standard model value. These results, together with recent evidence of the 125 GeV boson coupling to tau-leptons with a larger branching fraction consistent with the standard model, confirm that the leptonic couplings of the new boson are not flavour-universal.


Introduction 2 CMS detector and event reconstruction
boson production mode, gluon-fusion (GF), and the sub-dominant production mode, vector boson fusion (VBF). Higgs boson production in association with a vector boson (VH), while not optimized for, is taken into account in the H → µ + µ − analysis. The SM predicts Higgs boson production to be 87.3% GF, 7.1% VBF, and 5.0% VH for m H = 125 GeV at 8 TeV [17]. In addition to m , the most powerful variables for discriminating between the Higgs boson signal and the Drell-Yan and tt backgrounds are the jet multiplicity, the dilepton transversemomentum (p T ), and the invariant mass of the two largest transverse-momentum jets (m jj ). The gluon-gluon initial state of GF production tends to lead to more jet radiation than the quark-antiquark initial state of Drell-Yan production, leading to larger p T and jet multiplicity. Similarly, VBF production involves a pair of forward-backward jets with a large m jj compared to Drell-Yan plus two-jet or tt production. Events are further categorized by their m resolution and the kinematics of the jets. This paper is organized as follows. Section 2 introduces the CMS detector and event reconstruction, Section 3 describes the H → µ + µ − event selection, Section 4 the H → µ + µ − selection efficiency, Section 5 details the systematic uncertainties included in the H → µ + µ − analysis, Section 6 presents the results of the H → µ + µ − search, Section 7 describes the H → e + e − search, and Section 8 provides a summary.

CMS detector and event reconstruction
The central feature of the CMS apparatus is a superconducting solenoid of 6 m internal diameter, providing a magnetic field of 3.8 T. Within the superconducting solenoid volume are a silicon pixel and strip tracker, a lead tungstate crystal electromagnetic calorimeter (ECAL), and a brass/scintillator hadron calorimeter (HCAL), each composed of a barrel and two endcap sections. Muons are measured in gas-ionization detectors embedded in the steel flux-return yoke outside the solenoid. Extensive forward calorimetry complements the coverage provided by the barrel and endcap detectors.
The first level of the CMS trigger system, composed of custom hardware processors, uses information from the calorimeters and muon detectors to select the most interesting events in a fixed time interval of less than 4 µs. The high level trigger processor farm further decreases the event rate from at most 100 kHz to less than 1 kHz, before data storage. A more detailed description of the detector as well as the definition of the coordinate system and relevant kinematic variables can be found in Ref. [19].
The CMS offline event reconstruction creates a global event description by combining information from all subdetectors. This combined information then leads to a list of particle-flow (PF) objects [20, 21]: candidate muons, electrons, photons, and hadrons. By combining information from all subdetectors, particle identification and energy estimation performance are improved. In addition, double counting subdetector energy deposits when reconstructing different particle types is eliminated.
Due to the high instantaneous luminosity of the LHC, many proton-proton interactions occur in each bunch crossing. An average of 9 and 21 interactions occur in each bunch crossing for the 7 and 8 TeV data samples, respectively. Most interactions produce particles with relatively low transverse-momentum (p T ), compared to the particles produced in an H → + − signal event. These interactions are termed "pileup", and can interfere with the reconstruction of the high-p T interaction, whose vertex is identified as the vertex with the largest scalar sum of the squared transverse momenta of the tracks associated with it. All charged PF objects with tracks coming from another vertex are then removed.
Hadronic jets are clustered from reconstructed PF objects with the infrared-and collinear-safe anti-k T algorithm [22,23], operated with a size parameter of 0.5. The jet momentum is determined as the vectorial sum of the momenta of all PF objects in the jet, and is found in the simulation to be within 5% to 10% of the true momentum over the whole p T spectrum of interest and detector acceptance. An offset correction is applied to take into account the extra neutral energy clustered in jets due to pileup. Jet energy corrections are derived from the simulation, and are confirmed by in-situ measurements of the energy balance in dijet, photon plus jet, and Z plus jet (where the Z-boson decays to µ + µ − or e + e − ) events [24]. The jet energy resolution is 15% at 10 GeV, 8% at 100 GeV, and 4% at 1 TeV [25]. Additional selection criteria are applied to each event to remove spurious jet-like objects originating from isolated noise patterns in certain HCAL regions.
Matching muons to tracks measured in the silicon tracker results in a relative p T resolution for muons with 20 < p T < 100 GeV of 1.3-2.0% in the barrel and better than 6% in the endcaps. The p T resolution in the barrel is better than 10% for muons with p T up to 1 TeV [26]. The mass resolution for Z → µµ decays is between 1.1% and 1.9% depending on the pseudorapidity of each muon, for |η| < 2.1. The mass resolution for Z → ee decays when both electrons are in the ECAL barrel (endcaps) is 1.6% (2.6%) [27].

H → µ + µ − event selection
Online collection of events is performed with a trigger that requires at least one isolated muon candidate with p T above 24 GeV in the pseudorapidity range |η| ≤ 2.1. In the offline selection, muon candidates are required to pass the "Tight muon selection" [26] and each muon trajectory is required to have an impact parameter with respect to the primary vertex smaller than 5 mm and 2 mm in the longitudinal and transverse directions, respectively. They must also have p T > 15 GeV and |η| ≤ 2.1.
For each muon candidate, an isolation variable is constructed using the scalar sum of the transverse-momentum of particles, reconstructed as PF objects, within a cone centered on the muon. The boundary of the cone is ∆R = √ (∆η) 2 + (∆φ) 2 = 0.4 away from the muon, and the p T of the muon is not included in the sum. While only charged particles associated with the primary vertex are taken into account, a correction must be applied for contamination from neutral particles coming from pileup interactions. On average, in inelastic proton-proton collisions, neutral pileup particles deposit half as much energy as charged pileup particles. The amount of energy coming from charged pileup particles is estimated as the sum of the transverse momenta of charged tracks originating from vertices other than the primary vertex, but still entering the isolation cone. The neutral pileup energy in the isolation cone is then estimated to be 50% of this value and subtracted from the muon isolation variable. A muon candidate is accepted if the corrected isolation variable is less than 12% of the muon p T .
To pass the offline selection, events must contain a pair of opposite-sign muon candidates passing the above selection, and the muon which triggered the event is required to have p T > 25 GeV. All combinations of opposite-sign pairs, where one of the muons triggers the event, are considered as dimuon candidates in the dimuon invariant mass distribution analysis. Each pair is effectively treated as a separate event, and referred to as such for the remainder of this paper. Less than 0.1% of the SM Higgs boson events and 0.005% of the background events in each category contain more than one pair of muons.
After selecting events with a pair of isolated opposite-sign muons, events are categorized according to the properties of jets. Jets reconstructed from PF objects are only considered if their p T is greater than 30 GeV and |η| < 4.7. A multivariate analysis (MVA) technique is used to discriminate between jets originating from hard interactions and jets originating from pileup [28].
Dimuon events are classified into two general categories: a 2-jet category and a 0,1-jet category. The 2-jet category requires at least two jets, with p T > 40 GeV for the leading jet and p T > 30 GeV for the subleading jet. A 2-jet event must also have p miss T < 40 GeV, where p miss T is the magnitude of the vector sum of the transverse momenta of the dimuon and dijet systems. The p miss T requirement reduces the tt contamination in the 2-jet category, since tt decays also include missing transverse momentum due to neutrinos. All dimuon events not selected for the 2-jet category are placed into the 0,1-jet category where the signal is produced dominantly by GF.
The 2-jet category is further divided into VBF Tight, GF Tight, and Loose subcategories. The VBF Tight category has a large S/B ratio for VBF produced events. It requires m jj > 650 GeV and |∆η(jj)| > 3.5, where |∆η(jj)| is the absolute value of the difference in pseudorapidity between the two leading jets. For a SM Higgs boson with m H = 125 GeV, 79% of the signal events in this category are from VBF production. Signal events in the 2-jet category that do not pass the VBF Tight criteria mainly arise from GF events, which contain two jets from initial-state radiation. The GF Tight category captures these events by requiring the dimuon transverse momentum (p µµ T ) to be greater than 50 GeV and m jj > 250 GeV. To further increase the sensitivity of this search, 2-jet events that fail the VBF Tight and GF Tight criteria are still retained in a third subcategory called 2-jet Loose.
In the 0,1-jet category, events are split into two subcategories based on the value of p µµ T . The most sensitive subcategory is 0,1-jet Tight which requires p µµ T greater than 10 GeV, while the events with p µµ T less than 10 GeV are placed in the 0,1-jet Loose subcategory. The S/B ratio is further improved by categorizing events based on the dimuon invariant mass resolution as follows. Given the narrow Higgs boson decay width, the mass resolution fully determines the shape of the signal peak. The dimuon mass resolution is dominated by the muon p T resolution, which worsens with increasing |η| [26]. Hence, events are further sorted into subcategories based on the |η| of each muon and are labeled as "barrel" muons (B) for |η| < 0.8, "overlap" muons (O) for 0.8 ≤ |η| < 1.6, and "endcap" muons (E) for 1.6 ≤ |η| < 2.1. The 0,1-jet dimuon events are then assigned, within the corresponding Tight and Loose categories, to all possible dimuon |η| combinations. The dimuon mass resolution for each category is shown in Table 1. Due to the limited size of the data samples, the 2-jet subcategories are not split into further subcategories according to the muon p T resolution. This leads to a total of fifteen subcategories: three 2-jet subcategories, six 0,1-jet Tight subcategories, and six 0,1-jet Loose subcategories.

H → µ + µ − event selection efficiency
While the background shape and normalization are obtained from data, the selection efficiency for signal events has to be determined using MC simulation. For the GF and VBF production modes, signal samples are produced using the POWHEG-BOX next-to-leading-order (NLO) generator [29][30][31] interfaced with PYTHIA 6.4.26 [32] for parton showering. VH samples are produced using HERWIG++ [33] and its integrated implementation of the NLO POWHEG method.
These samples are then passed through a simulation of the CMS detector, based on GEANT4 [34], that has been extensively validated on both 7 and 8 TeV data. This validation includes a comparison of data with MC simulations of the Drell-Yan plus jets and tt plus jets backgrounds produced using MADGRAPH [35] interfaced with PYTHIA 6.4.26 for parton showering. In all categories, the simulated m µµ spectra agree well with the data, for 110 < m µµ < 160 GeV. Scale factors related to muon identification, isolation, and trigger efficiency are applied to each Table 1: Details regarding each of the H → µ + µ − categories. The top half of the table refers to the 5.0 ± 0.1 fb −1 at 7 TeV, while the bottom half refers to the 19.7 ± 0.5 fb −1 at 8 TeV. Each row lists the category name, FWHM of the signal peak, acceptance times selection efficiency (A ) for GF, A for VBF, A for VH, expected number of SM signal events in the category for m H = 125 GeV (N S ), number of background events within a FWHM-wide window centered on 125 GeV estimated by a signal plus background fit to the data (N B ), number of observed events within a FWHM-wide window centered on 125 GeV (N Data ), systematic uncertainty to account for the parameterization of the background (N P ), and N P divided by the statistical uncertainty on the fitted number of signal events (N P /σ Stat ). The expected number of SM signal events is where L is the integrated luminosity and σB is the SM cross section times branching fraction. --simulated signal sample to correct for discrepancies between the detector simulation and data. These scale factors are estimated using the "tag-and-probe" technique [26]. The detector simulation and data typically agree to within 1% on the muon identification efficiency, to within 2% on the muon isolation efficiency, and to within 5% on the muon trigger efficiency.
The overall acceptance times selection efficiency for the H → µ + µ − signal depends on the mass of the Higgs boson. For a Higgs boson mass of 125 GeV, the acceptance times selection efficiencies are shown in Table 1.

H → µ + µ − systematic uncertainties
Since the statistical analysis is performed on the dimuon invariant mass spectrum, it is necessary to categorize the sources of systematic uncertainties into "shape" uncertainties that change the shape of the dimuon invariant mass distribution, and "rate" uncertainties that affect the overall signal yield in each category.
The only relevant shape uncertainties for the signal are related to the knowledge of the muon momentum scale and resolution and they affect the width of the signal peak by 3%. The signal shape is parameterized by a double-Gaussian (see Section 6) and this uncertainty is applied by constraining the width of the narrower Gaussian. The probability density function used to constrain this nuisance parameter in the limit setting procedure is itself Gaussian with its mean set to the nominal value and its width set to 3% of the nominal value.
Rate uncertainties in the signal yield are evaluated separately for each Higgs boson production process and each centre-of-mass energy. These uncertainties are applied using log-normal probability density functions as described in Ref. [36].
To estimate the theoretical uncertainty in the signal production processes due to neglected higher-order quantum corrections, the renormalization and factorization scales are varied simultaneously by a factor of two up and down from their nominal values. This leads to an uncertainty in the cross section and acceptance times efficiency which depends on the mass of the Higgs boson. For 125 GeV GF production, the uncertainty in the signal yield ranges from 1% in the 0,1-jet Tight category to 25% in the 2-jet VBF Tight and GF Tight categories [15][16][17]. For VBF production, the uncertainties range from 1% to 7% [15][16][17].
Uncertainty in the knowledge of the parton distribution functions (PDFs) also leads to uncertainty in the signal production process. This uncertainty is estimated using the PDF4LHC prescription [37,38] and the CT10 [39], MSTW2008 [40], and NNPDF 2.3 [41] PDF sets provided by the LHAPDF package version 5.8.9 [42]. The value of the uncertainty depends on the mass of the Higgs boson. For m H = 125 GeV, the uncertainties on the signal yields for GF and VBF are 11% and 5%, respectively [15-17].
For VH production, only rate uncertainties in the production cross section due to quantum corrections and PDFs are considered. They are 3% or less.
Uncertainty in the modeling of the parton showers and underlying event activity may affect the kinematics of selected jets. This uncertainty is estimated by comparing various tunes of the relevant PYTHIA parameters. The D6T [43], P0 [44], ProPT0, and ProQ20 [45] tunes are compared with the Z2* [43] tune, which is the nominal choice. Signal yield uncertainties range from 6% to 60% (the largest uncertainty is in the 2-jet VBF Tight category) for the GF production mode and 2% to 15% for the VBF production mode in the various categories. Large uncertainties in the 2-jet categories are expected for the GF production mode, since two-jet events are simulated solely by parton showering in the POWHEG-PYTHIA NLO samples.
The theoretical uncertainty in the branching fraction to µ + µ − is taken from Refs. [15][16][17], and depends on the Higgs boson mass. For m H = 125 GeV, it is 6%, while it shrinks to 3% for m H = 150 GeV.
The uncertainty in the luminosity, 2.2% at 7 TeV [46] and 2.6% at 8 TeV [47], is directly applied to the signal yield in all categories. The signal yield uncertainty due to the limited size of the simulated event samples varies from 1% to 8% depending on the category.
The small uncertainty associated with the "tag-and-probe" technique used to determine the data to simulation muon efficiency scale factors leads to an uncertainty in the signal yield of 1.6% [26].
A systematic uncertainty in the knowledge of the pileup multiplicity is evaluated by varying the total cross section for inelastic proton-proton collisions. The resulting uncertainty in the signal yield varies, for GF production, from less than 1% to 5% in the various categories, while for VBF production it varies from less than 1% to 2%.
The acceptance and selection efficiency of the jet-based selections are affected by uncertainty in the jet energy resolution and absolute jet energy scale calibration [24]. The energy resolution related uncertainty in the signal yield is 1% to 3% for GF and 1% to 2% for VBF, while the absolute energy scale calibration uncertainty in the signal yield is 1% to 8% for GF, and 2% to 6% for VBF in the various categories.
Misidentification of "hard jets" (jets originating from the hard interaction) as "pileup jets" (jets originating from pileup interactions) can lead to migration of signal events from the 2-jet category to the 0,1-jet category. Events containing a Z-boson, tagged by its dilepton decay, recoiling against a jet provide a pure source of hard jets similar to the Higgs boson signal. Data events may then be used to estimate the misidentification rate of the MVA technique used to discriminate between hard jets and pileup jets using data [28]. A pure source of hard jets is found by selecting events with p Z T > 30 GeV and jets where |∆φ(Z, j)| > 2.5 and 0.5 < p j T /p Z T < 1.5 The misidentification rate of these jets as pileup jets is compared in data and simulation, and the difference taken as a systematic uncertainty. The resulting uncertainty in the signal yield varies from 1% to 4% for the various categories.
When estimating each of the signal yield uncertainties, attention is paid to the sign of the yield variation in each category. Categories that vary in the same direction are considered fully correlated while categories that vary in opposite directions are considered anticorrelated. These correlations are considered between all categories at both beam energies for all of the signal yield uncertainties except for the luminosity uncertainty and the uncertainty caused by the limited size of the simulated event samples. The luminosity uncertainty is considered fully correlated between all categories, but uncorrelated between the two centre-of-mass energies. The MC simulation statistical uncertainty is considered uncorrelated between all categories and both centre-of-mass energies.
To account for the possibility that the nominal background parameterization may imperfectly describe the true background shape, an additional systematic uncertainty is included. This uncertainty is implemented as a floating additive contribution to the number of signal events, constrained by a Gaussian probability density function with mean set to zero and width set to the systematic uncertainty. This systematic uncertainty is estimated by checking the bias in terms of the number of signal events that are found when fitting the signal plus nominal background model (see Section 6) to pseudo-data generated from various alternative background models, including polynomials, that were fit to data. Bias estimates are performed for Higgs boson mass points from 120 to 150 GeV. The uncertainty estimate is then taken as the maximum absolute value of the bias of all of the mass points and all of the alternative background models. It is then applied uniformly to all Higgs boson masses. The estimates of the uncertainty in the parameterization of the background (N P ) are shown in Table 1 for each category. The effect of this systematic uncertainty is larger than all of the others. The expected limit (see Section 6) would be 20% lower at m H = 125 GeV without the systematic uncertainty in the parameterization of the background.

H → µ + µ − results
To estimate the signal rate, the dimuon invariant mass (m µµ ) spectrum is fit with the sum of parameterized signal and background shapes. This fit is performed simultaneously in all of the categories. Since in the mass range of interest the natural width of the Higgs boson is narrower than the detector resolution, the m µµ shape is only dependent on the detector resolution and QED final state radiation. A double-Gaussian function is chosen to parameterize the shape of the signal. The background shape, dominated by the Drell-Yan process, is modeled by a function, f (m µµ ), that is the sum of a Breit-Wigner function and a 1/m 2 µµ term, to model the Z-boson and photon contributions, both multiplied by an exponential function to approximate the effect of the PDF on the m µµ distribution. This function is shown in the following equation, and involves the parameters λ, β, m Z , and Γ: The coefficients C 1 and C 2 are set to ensure the integral of each of the two terms is normalized to unity in the m µµ fit range, 110 to 160 GeV. Before results are extracted, the mass and width of the Z-boson peak, m Z and Γ, are estimated by fitting a Breit-Wigner function to the Z-boson mass peak region (88-94 GeV) in each category. The other parameters, λ and β, are fit simultaneously with the amount of signal in the signal plus background fit. Besides the Drell-Yan process, most of the remaining background events come from tt production. The background parameterization has been shown to fit the dimuon mass spectrum well, even when it includes a large tt fraction. Fits of the background model to data (assuming no signal contribution) are presented in Fig. 1 for the most sensitive categories: the 0,1-jet Tight category with both muons reconstructed in the barrel region and the 2-jet VBF Tight category.
Results are presented in terms of the signal strength, which is the ratio of the observed (or expected) σB, to that predicted in the SM for the H → µ + µ − process. Results are also presented, for m H = 125 GeV, in terms of σB, and B. No significant excess is observed. Upper limits at the 95% CL are presented using the CL s criterion [48,49]. They are calculated using an asymptotic profile likelihood ratio method [36,50,51] involving dimuon mass shapes for each signal process and for background. Systematic uncertainties are incorporated as nuisance parameters and treated according to the frequentist paradigm [36].
Exclusion limits for Higgs boson masses from 120 to 150 GeV are shown in Fig. 2. The observed 95% CL upper limits on the signal strength at 125 GeV are 22.4 using the 7 TeV data and 7.0 using the 8 TeV data. The corresponding background-only expected limits are 16.6 +7.3 −4.9 using the 7 TeV data and 7.2 +3.2 −2.1 using the 8 TeV data. Accordingly, the combined observed limit for 7 and 8 TeV is 7.4, while the background-only expected limit is 6.5 +2.8 −1.9 . The best fit value of the signal strength for a Higgs boson mass of 125 GeV is 0.8 +3.5 −3.4 . We did not restrict the fit to positive values, to preserve the generality of the result. Exclusion limits in terms of σ(8 TeV)B using only 8 TeV data are shown in Fig. 3 (left). The relative contributions of GF, VBF, and VH are assumed to be as predicted in the SM, and theoretical uncertainties on the cross sections and branching fractions are omitted. At 125 GeV, the observed 95% CL upper limit on σ(7 TeV)B using only 7 TeV data is 0.084 pb, while the background-only expected limit is 0.062 +0.026 −0.018 pb. Using only 8 TeV data, the observed limit on σ(8 TeV)B is 0.033 pb, while the background-only expected limit is 0.034 +0.014 −0.010 pb. To combine 7 and 8 TeV data, the SM ratio of production cross sections is assumed and the result is quoted at a reference energy of 8 TeV. This leads to an observed limit of 0.035 pb on σ(8 TeV)B and a background-only expected limit of 0.031 +0.013 −0.010 pb. Assuming the SM production cross section, this corresponds to an upper limit on B(H → µ + µ − ) of 0.0016.   Exclusion limits on individual production modes may also be useful to constrain BSM models that predict H → µ + µ − production dominated by a single mode. Limits are presented on the signal strength and σB using a combination of 7 and 8 TeV data. The limits on σB are at a reference energy of 8 TeV assuming the SM ratio of production cross sections. The observed 95% CL upper limit on the GF signal strength, assuming the VBF and VH rates are zero, is 13.2, while the background-only expected limit is 9.8 +4.4 −2.9 . This corresponds to an observed upper limit on σ GF (8 TeV)B of 0.054 pb and an expected limit of 0.041 +0.017 −0.012 pb. Similarly, the observed upper limit on the VBF signal strength, assuming the GF and VH rates are zero, is 11.2, while the background-only expected limit is 13.4 +6.6 −4.2 . This corresponds to an observed upper limit on σ VBF (8 TeV)B of 0.0038 pb and an expected limit of 0.0046 +0.0023 −0.0015 pb. For m H = 125 GeV, an alternative H → µ + µ − analysis was performed to check the results of the main analysis. It uses an alternative muon isolation variable based only on tracker information, an alternative jet reconstruction algorithm (the jet-plus-track algorithm [52]), and an alternative event categorization. The event categorization contains similar 2-jet categories to the main analysis, while separate categories are utilized for 0-jet and 1-jet events. Dimuon mass resolution-based categories are not used, but the 0-jet category does contain two subcategories separated by p µµ T . As in the main analysis, results are extracted by fitting signal and background shapes to the m µµ spectra in each category, but unlike the main analysis, f (m µµ ) = exp(p 1 m µµ )/(m µµ − p 2 ) 2 is used as the background shape. The systematic uncertainty on the parameterization of the background is estimated and applied in the same way as in the main analysis. For the alternative analysis, the observed (expected) 95% CL upper limit on the signal strength is 7.8 (6.5 +2.8

Search for Higgs boson decays to e + e −
In the SM, the branching fraction of the Higgs boson into e + e − is tiny, because the fermionic decay width is proportional to the mass of the fermion squared. This leads to poor sensitivity to SM production for this search when compared to the search for H → µ + µ − . On the other hand, the sensitivity in terms of σB is similar to H → µ + µ − , because dielectrons and dimuons share similar invariant mass resolutions, selection efficiencies, and backgrounds. Since the sensitivity to the SM rate of H → e + e − is so poor, an observation of the newly discovered particle decaying to e + e − with the current integrated luminosity would be evidence of physics beyond the standard model.
In a similar way to the H → µ + µ − analysis, a search in the m ee spectrum is performed for a narrow peak over a smoothly falling background. The irreducible background is dominated by Drell-Yan production, with smaller contributions from tt and diboson production. Misidentified electrons make up a reducible background that is highly suppressed by the electron identification criteria. The reducible H → γγ background is estimated from simulation to be negligible compared to other backgrounds, although large compared to the SM H → e + e − signal. The overall background shape and normalization are estimated by fitting the observed m ee spectrum in data, assuming a smooth functional form, while the signal acceptance times selection efficiency is estimated from simulation. The analysis is performed only on proton-proton collision data collected at 8 TeV, corresponding to an integrated luminosity of 19.7 ± 0.5 fb −1 .
The trigger selection requires two electrons, one with transverse energy, E T , greater than 17 GeV and the other with E T greater than 8 GeV. These electrons are required to be isolated with respect to additional energy deposits in the ECAL, and to pass selections on the ECAL cluster shape. In the offline selection, electrons are required to be inside the ECAL fiducial region: |η| < 1.44 (barrel) or 1.57 < |η| < 2.5 (endcaps). Their energy is estimated by the same multivariate regression technique used in the CMS H → ZZ analysis [53], and their E T is required to be greater than 25 GeV. Electrons are also required to satisfy standard CMS identification and isolation requirements, which correspond to a single electron efficiency of around 90% in the barrel and 80% in the endcaps [54].
To improve the sensitivity of the search we separate the sample into four distinct categories: two 0,1-jet categories and two for which a pair of jets is required. The two 2-jet categories are designed to select events produced via the VBF process. The two jets are required to have an invariant mass greater than 500 (250) GeV for the 2-jet Tight (Loose) category, p T > 30 (20) GeV, |∆η(jj)| > 3.0, |∆φ(jj, e + e − )| > 2.6, and |z| = |η(e + e − ) − [η(j 1 ) + η(j 2 )]/2| < 2.5 [55]. The cut on z ensures that the dielectron is produced centrally in the dijet reference frame, which helps to enhance the VBF signal over the Drell-Yan background. More details on the selection can be found in Ref. [56]. The rest of the events are classified into two 0,1-jet categories. To exploit the better energy resolution of electrons in the barrel region, these categories are defined as: both electrons in the ECAL barrel (0,1-jet BB) or at least one of them in the endcap (0,1-jet Not BB). For each category, the FWHM of the expected signal peak, expected number of SM signal events for m H = 125 GeV, acceptance times selection efficiency, number of background events near 125 GeV, and number of data events near 125 GeV are shown in Table 2.
Data have been compared to the simulated Drell-Yan and tt background samples described in Section 4. In all categories, the dielectron invariant mass spectra from 110 to 160 GeV agree well, and the normalizations agree within 4.5%. Using simulation, the reducible background of H → γγ events has also been estimated. For m H = 125 GeV, 0.23 SM H → γγ events are expected to pass the dielectron selection compared to about 10 −3 events for the SM H → e + e − signal. While this background is much larger than the SM H → e + e − signal, it is negligible compared to the Drell-Yan and tt backgrounds in each category.
Results are extracted from the data for m H values between 120 and 150 GeV by fitting the mass spectra of the four categories in the range 110 < m ee < 160 GeV. The parameterization used for the background is the same as used in the µ + µ − search, Equation (1). Background-only m ee fits to data are shown in Fig. 4 for the 0,1-jet BB and 2-jet Tight categories.
The uncertainty related to the choice of background parameterization is estimated and incorporated into the results using the same method as in the µ + µ − search (see Section 5). This uncertainty in terms of the number of signal events (N P ) is shown in Table 2. This systematic uncertainty is larger than all of the others, and removing it would lower the expected limit by 28%, for m H = 125 GeV.
Systematic uncertainties on the electron trigger and identification efficiencies lead to signal yield uncertainties of 1% and 2%, respectively. The uncertainty in the jet energy resolution and jet energy calibration result in an uncertainty in the signal yield of <1% (2%) for GF (VBF) in the 0,1-jet categories and 11% (3%) for GF (VBF) in the 2-jet categories. Uncertainties in the knowledge of the integrated luminosity, modeling of parton-showers, modeling of underlying event activity, neglected higher-order quantum corrections, and knowledge of the PDFs are identical to that of H → µ + µ − described in Section 5.
The uncertainties on the electron trigger efficiency, electron identification efficiency, integrated luminosity, neglected higher-order quantum corrections, and PDFs are considered fully correlated between the four categories. The signal yield uncertainties due to the modeling of parton showers, underlying event activity, jet energy resolution, and jet energy calibration primarily affect the jet selection and lead to event migration between the 0,1-jet and 2-jet categories. As a result, they are considered fully anticorrelated between the 0,1-jet categories and the 2-jet categories.
No significant excess of events is observed. Upper limits on σ(8 TeV)B and B are reported. The observed 95% CL upper limit on σ(8 TeV)B at 125 GeV is 0.041 pb while the background-only expected limit is 0.052 +0.022 −0.015 pb. Assuming the SM production cross-section, this corresponds to an observed upper limit on B(H → e + e − ) of 0.0019, which is approximately 3.7 × 10 5 times the SM prediction. Upper limits on σ(8 TeV)B are shown for Higgs boson masses from 120 to 150 GeV at the 95% CL in Fig. 3 ( right).

Summary
Results are presented from a search for a SM-like Higgs boson decaying to µ + µ − and for the first time to e + e − . For the search in µ + µ − , the analyzed CMS data correspond to integrated luminosities of 5.0 ± 0.1 fb −1 collected at 7 TeV and 19.7 ± 0.5 fb −1 collected at 8 TeV, while only the 8 TeV data are used for the search in the e + e − channel. The Higgs boson signal is sought as a narrow peak in the dilepton invariant mass spectrum on top of a smoothly falling background dominated by the Drell-Yan, tt, and vector boson pair-production processes. Events are split into categories corresponding to different production topologies and dilepton invariant mass resolutions. The signal strength is then extracted using a simultaneous fit to the dilepton invariant mass spectra in all of the categories.
No significant H → µ + µ − signal is observed. Upper limits are set on the signal strength at the 95% CL. Results are presented for Higgs boson masses between 120 and 150 GeV. The com- Table 2: Details regarding each category of the H → e + e − analysis, for 19.7 ± 0.5 fb −1 at 8 TeV. Each row lists the category name, FWHM of the signal peak, acceptance times selection efficiency (A ) for GF, A for VBF, expected number of SM signal events in the category times 10 5 for m H = 125 GeV (N S ), number of background events within a FWHM-wide window centered on 125 GeV estimated by a signal plus background fit to the data (N B ), number of observed events within a FWHM-wide window centered on 125 GeV (N Data ), systematic uncertainty to account for the parameterization of the background (N P ), and N P divided by the statistical uncertainty on the fitted number of signal events (N P /σ Stat ). The expected number of SM signal events is N S = L × (σB A ) GF + L × (σB A ) VBF , where L is the integrated luminosity and σB is the SM cross section times branching fraction. bined observed limit on the signal strength, for a Higgs boson with a mass of 125 GeV, is 7.4, while the expected limit is 6.5 +2.8 −1.9 . Assuming the SM production cross section, this corresponds to an upper limit of 0.0016 on B(H → µ + µ − ). For a Higgs boson mass of 125 GeV, the best fit signal strength is 0.8 +3.5 −3.4 . In the H → e + e − channel, SM Higgs boson decays are far too rare to detect, and no signal is observed. For a Higgs boson mass of 125 GeV, a 95% CL upper limit of 0.041 pb is set on σB(H → e + e − ) at 8 TeV. Assuming the SM production cross section, this corresponds to an upper limit on B(H → e + e − ) of 0.0019, which is approximately 3.7 × 10 5 times the SM prediction. For comparison, the H → µ + µ − observed 95% CL upper limit on σB(H → µ + µ − ) is 0.033 pb (using only 8 TeV data), which is 7.0 times the expected SM Higgs boson cross section.
These results, together with recent evidence for the 125 GeV boson's coupling to τ-leptons [6] with a larger B consistent with the SM value of 0.0632 ± 0.0036 [5], show for the first time that the leptonic couplings of the new boson are not flavour-universal, confirming the SM prediction.

Acknowledgments
We congratulate our colleagues in the CERN accelerator departments for the excellent performance of the LHC and thank the technical and administrative staffs at CERN and at other CMS institutes for their contributions to the success of the CMS effort. In addition, we gratefully acknowledge the computing centres and personnel of the Worldwide LHC Computing Grid for delivering so effectively the computing infrastructure essential to our analyses. Finally, we acknowledge the enduring support for the construction and operation of the LHC and the CMS detector provided by the following funding agencies: BMWFW and FWF (Austria); FNRS and [10] N. Vignaroli, "Searching for a dilaton decaying to muon pairs at the LHC", Phys. Rev  [15] LHC Higgs Cross Section Working Group, "Handbook of LHC Higgs Cross Sections: 1. Inclusive Observables", (2011). arXiv:1101.0593.
[18] CMS Collaboration, "Constraints on the Higgs boson width from off-shell production and decay to Z-boson pairs", Phys.