Source-linked AI summary
Search for resonant and nonresonant new phenomena in high-mass dilepton final states at $\sqrt{s} = $ 13 TeV
CMS Collaboration
TL;DR
The paper searches high-mass dilepton spectra for resonant and nonresonant phenomena that could indicate physics beyond the Standard Model. Using CMS proton-proton collision data at 13 TeV, it finds agreement with background expectations and sets stringent limits on resonance masses and ultraviolet cutoff parameters.
Problem
The search addresses whether high-mass dilepton events contain resonant or nonresonant phenomena predicted by models extending the Standard Model, including possible lepton flavor universality violation.
Source: §1 · p. 3Method
CMS analyzes high-mass dielectron and dimuon invariant-mass spectra using 13 TeV collision data, modeling Standard Model backgrounds and testing resonant, nonresonant, and flavor-ratio signatures.
Source: §1 · p. 4, §3.2 · p. 7, §10 · p. 31Results
No significant deviation from background expectations is observed; limits include graviton masses of 2.47–4.78 TeV and ultraviolet cutoffs of 23.8–36.4 TeV for contact interactions.
Source: §8 · p. 17, §10 · pp. 31–32Takeaways & Limitations
The results constrain models with new resonances, contact interactions, and large extra dimensions through lower limits on their masses or ultraviolet cutoff parameters.
Source: §10 · pp. 31–32For the largest considered resonance width, 10%, background-shape parameterization can significantly bias limits at low masses, so results below 700 GeV are not shown.
Source: §9.1 · p. 21
Abstract
from arXiv · showhide
A search is presented for physics beyond the standard model (SM) using electron or muon pairs with high invariant mass. A data set of proton-proton collisions collected by the CMS experiment at the LHC at $\sqrt{s} =$ 13 TeV from 2016 to 2018 corresponding to a total integrated luminosity of up to 140 fb$^{-1}$ is analyzed. No significant deviation is observed with respect to the SM background expectations. Upper limits are presented on the ratio of the product of the production cross section and the branching fraction to dileptons of a new narrow resonance to that of the Z boson. These provide the most stringent lower limits to date on the masses for various spin-1 particles, spin-2 gravitons in the Randall--Sundrum model, as well as spin-1 mediators between the SM and dark matter particles. Lower limits on the ultraviolet cutoff parameter are set both for four-fermion contact interactions and for the Arkani-Hamed, Dimopoulos, and Dvali model with large extra dimensions. Lepton flavor universality is tested at the TeV scale for the first time by comparing the dimuon and dielectron mass spectra. No significant deviation from the SM expectation of unity is observed.
1 Introduction
The paper searches the high-mass dilepton spectrum for resonant and nonresonant phenomena predicted by extensions of the SM, including spin-1 and spin-2 resonances, dark-matter mediators, contact interactions, and lepton-flavor-universality violation. It uses the full 13 TeV CMS data set from 2016–2018 and combines resonant fits, nonresonant binned analyses, and a dimuon-to-dielectron ratio test.
Motivation
- The search targets resonant spin-1 or spin-2 states and nonresonant high-mass dilepton excesses predicted by several extensions of the SM.The considered motivations include extra-dimensional gravitons, dark-matter mediators, and preon-inspired contact interactions.Source: §1 · p. 3
Lepton flavor universality
- Lepton flavor universality is tested for the first time in these final states by measuring the dimuon-to-dielectron ratio and comparing it with the SM expectation of unity.The comparison includes corrections for detector effects, lepton acceptances, and lepton efficiencies.Source: §1 · pp. 3–5
Data and analysis scope
- The analysis uses the full 2016–2018 CMS data set at √s = 13 TeV, corresponding to 137 fb−1 in dielectron and 140 fb−1 in dimuon events.It searches deviations from SM predictions in both dilepton invariant-mass spectra, with backgrounds primarily estimated from simulation and normalized near the Z peak.Source: §1 · p. 4
Resonant search
- Resonant searches scan masses and widths with unbinned maximum-likelihood fits, setting limits on the new-resonance cross-section times branching-fraction ratio to the SM Z boson.The ratio reduces common experimental and theoretical uncertainties between the resonance and Z-boson measurements.Source: §1 · p. 4
Nonresonant search
- Nonresonant searches combine counting experiments binned in dilepton mass and Collins–Soper cos θ∗, retaining lower-mass information on interference while gaining sensitivity at high mass.Unlike the resonant analysis, this strategy does not normalize the signal to the Z-boson cross section and is therefore more dependent on signal and background modeling.Source: §1 · p. 4
2 The CMS detector
The CMS detector combines a 3.8 T superconducting solenoid with tracking, calorimetry, and muon systems covering complementary pseudorapidity ranges. A two-tier trigger selects events online, reducing the rate from around 100 kHz to around 1 kHz before storage.
Detector components and coverage
- A 3.8 T superconducting solenoid encloses the silicon tracker, ECAL, and HCAL, while the muon system is installed outside the solenoid within the steel flux-return yoke.The tracker covers |η| < 2.5, calorimeters extend to |η| < 5.0 with forward hadron calorimeters, and muon detection covers |η| < 2.4.Source: §2 · p. 5
- The silicon tracker measures charged-particle trajectories over |η| < 2.5, and its pixel detector was upgraded before 2017 with an added barrel layer and forward disks.The ECAL contains nearly 76 000 lead tungstate crystals, while the HCAL uses alternating brass and scintillator layers.Source: §2 · p. 5
Trigger system
- A two-tier trigger system uses custom L1 hardware and an HLT processor farm to reduce the event rate from around 100 kHz to around 1 kHz before storage.The L1 decision is made within less than 4 µs using calorimeter and muon information, while the HLT runs optimized full-event reconstruction software.Source: §2 · p. 5
3 Signal models
The search models cover resonant high-mass dilepton production from new gauge bosons, spin-2 gravitons, and dark-matter mediators, as well as nonresonant effects from ADD virtual gravitons and four-fermion contact interactions.
3.1 Models with resonant signatures
- Resonant models include Z′ bosons from U′(1) extensions, RS spin-2 gravitons, and spin-1 dark-matter mediators.The Z′ benchmarks include GSM [42], SSM, LRS [42], and E6 models [1].Source: §3.1 · pp. 5–7
- Z′ production depends on up- and down-quark couplings and parton distributions, while narrow-width limits remain applicable despite widths up to 12%.Interference with Standard Model backgrounds is negligible in a sufficiently narrow window around the Z′ mass.Source: §3.1 · p. 6
- RS gravitons are characterized by k/M_Pl, with first-excitation width-to-mass ratios of 0.01%, 0.36%, and 1.42% for 0.01, 0.05, and 0.10.The model predicts spin-2 Kaluza–Klein graviton resonances produced at the LHC and decaying to lepton pairs.Source: §3.1 · p. 6
3.2 Models with nonresonant signatures
- Nonresonant models comprise ADD virtual spin-2 gravitons and four-fermion contact interactions, whose dilepton cross sections include interference and pure-signal terms.Interference can be constructive or destructive according to the sign of the model-specific parameter η_X.Source: §3.2 · p. 7
- ADD predicts an unresolved quasi-continuous KK-graviton spectrum, parameterized by ultraviolet cutoffs Λ_T or M_S, while interference with Drell–Yan is limited and positive interference is assumed.The effective-field-theory description is reliable only below a theory-dependent energy scale.Source: §3.2 · pp. 7–8
- Contact-interaction limits are evaluated separately for LL, RR, LR, and RL helicity currents, whose angular distributions favor positive or negative cos θ* depending on the current.Each helicity scenario sets all other currents to zero and uses its own compositeness scale.Source: §3.2 · p. 8
4 Simulated event samples
Simulated samples model the year-dependent backgrounds and signal processes, with higher-order corrections, reweighting, interference treatment, and detector-related adjustments applied to improve consistency with the data-taking conditions.
Sample production
- Dedicated simulated samples describe background and signal processes separately for each of the three data-taking years, accounting for changing beam and detector conditions.Different generators are used for individual processes.Source: §4 · p. 8
Background simulation
- The dominant DY background uses POWHEG at NLO, year-specific PDFs and PYTHIA configurations, and mass-dependent FEWZ corrections including γγ-initiated contributions.The 2016 samples use NNPDF3.0 and CUETP8M1, while 2017–2018 use NNPDF3.1 and CP5.Source: §4 · pp. 8–9
- Other backgrounds include top-quark, single-top, diboson, ττ, and W+jets processes simulated with POWHEG, PYTHIA, or MADGRAPH5 aMC@NLO and normalized to higher-order cross sections.The tt, qtW, and WW samples use POWHEG with PYTHIA showering, while WZ, ZZ, ττ, and W+jets use process-dependent generator configurations.Source: §4 · p. 9
- Because NNPDF3.1 predicts unphysical cross sections above 4 TeV, DY and tt samples are reweighted to the NNPDF3.0 NLO mℓℓ prediction for year consistency.For DY, this reweighting precedes the FEWZ higher-order correction; other affected processes are not reweighted because their contributions are small.Source: §4 · p. 9
Resonant signals
- Resonant signals use illustrative 5 TeV Z′ and 250–4000 GeV RS graviton samples, generated for electron and muon final states with year-dependent PYTHIA and PDF settings.Only one SSM Z′ sample is generated because spin-1 resonance properties can be studied with DY samples.Source: §4 · p. 9
Nonresonant signals
- CI and ADD samples model nonresonant signals with mass reweighting, interference with DY, helicity-state treatment, and a minimal ADD correction factor of 1.3.CI samples separately address LL, LR+RL, and RR states, while ADD and CI generation thresholds differ by year and model.Source: §4 · pp. 9–10
Detector simulation
- Detector effects are simulated with GEANT4, while dielectron samples receive reweighting for ECAL trigger prefiring and pileup, unlike dimuon samples without significant pileup dependence.Additional pp interactions are generated with PYTHIA to model pileup.Source: §4 · p. 10
5 Lepton reconstruction and event selection
The analysis updates lepton reconstruction, triggers, and event selection for high-mass dielectron and dimuon candidates. Electron and muon efficiencies are measured and modeled across detector regions and masses, with dedicated disambiguation and background-rejection requirements.
5.1 Electron reconstruction and selection
- Electron candidates require pT > 35 GeV and |ηC| < 1.44 or 1.57 < |ηC| < 2.50, excluding the transition region.The selection also uses calorimeter-shower, track-matching, hadronic-energy, and isolation requirements optimized for high-energy electrons.Source: §5.1 · p. 11
- Electron selection requires at least one barrel electron to suppress jet backgrounds, while both-endcap events provide a control region for estimating that background.No electron charge-sign requirement is imposed because charge misidentification rises to approximately 10% at several-TeV masses.Source: §5.1 · p. 11
5.2 Muon reconstruction and selection
- Muon candidates require pT > 53 GeV, |η| < 2.4, isolation, opposite charge, a common vertex, and an opening angle below π − 0.02.Dedicated reconstruction accounts for radiative losses above 200 GeV; cosmic-ray backgrounds remain below 0.2% after full selection for events containing muons with pT > 300 GeV.Source: §5.2 · p. 12
- 93% is the measured overall efficiency to trigger on, reconstruct, and identify dimuon pairs within detector acceptance, with little dependence on dimuon mass.The dimuon trigger efficiency exceeds 99.3% for barrel-barrel events and 99.0% when at least one muon is in an endcap.Source: §5.2 · p. 13
- For mℓℓ > 200 GeV, reconstruction and selection efficiencies are nearly mass-independent, while acceptance changes reflect lepton angular distributions and production modes.The product of acceptance and efficiency is obtained from simulation for dielectron and dimuon pairs and parameterized versus mass; the high-mass RS graviton decrease reflects changing gg and qq contributions.Source: §5.2 · p. 13, Figure 1 · p. 13
5.3 Event disambiguation
- When multiple dilepton pairs occur, the pair nearest the Z mass is selected if within 20 GeV; otherwise, the two highest-pT leptons are used.For Collins–Soper angle reconstruction, same-sign dielectron events are treated as charge-misidentified, yielding the correct lepton assignment in 76% of simulated cases.Source: §5.3 · pp. 13–14
6 Background estimation
The dominant background is Drell–Yan production, supplemented by simulated top-quark, electroweak, diboson, and jet-misidentification processes. Background estimates use simulation and data control regions, with combined normalization to the Z-boson region and channel-specific corrections.
Background composition
- Drell–Yan production is the dominant irreducible background, while tt, tW, diboson, QCD multijet, and W+jets processes provide additional high-mass dilepton events.Jets or nonisolated leptons reconstructed as isolated leptons constitute the jet-misidentification background.Source: §6 · p. 14
Simulated backgrounds
- The Drell–Yan and other non-jet backgrounds are estimated from simulation, with mass-dependent higher-order corrections for Drell–Yan and validation of tt modeling in an eμ control region.The tt simulation describes both the mass-distribution shape and event yield within uncertainties.Source: §6 · p. 14
Data-driven backgrounds
- Jet-misidentification backgrounds are estimated from data control regions, carry uncertainties up to 50%, but contribute only about 1–3% of the total event yield.Their estimate is independent of cos θ* and is scaled by 0.5 when splitting the nonresonant sample at cos θ* = 0.Source: §6 · p. 14
Control-region normalization
- The combined background is normalized in the 60 < m_ll < 120 GeV Z-boson control region, with DY pT reweighting and year- and channel-dependent normalization factors.For dimuons, the factors are 0.97 in 2016, 1.03 in 2017, and 1.00 in 2018; the electron factor is 0.95.Source: §6 · pp. 14–15
7 Systematic uncertainties
Systematic uncertainties affect efficiencies, mass measurements, resolutions, backgrounds, and signal modeling across the dilepton search. Their treatment differs between resonant and nonresonant interpretations, with a mass-dependent PDF-reweighting uncertainty reaching 30% for nonresonant signals.
Detector and reconstruction
- Event-selection efficiency uncertainties are 6% for dielectron barrel-barrel, 8% for barrel-endcap, and 1–2% throughout the dimuon channel.An additional one-sided dimuon efficiency uncertainty reaches −1.0 (−6.5)% at mµµ = 4 TeV in 2016 barrel-barrel (barrel-endcap plus endcap-endcap) events.Source: §7 · p. 15
Mass calibration
- Mass-scale uncertainties are 2 (1)% for dielectron barrel-barrel (barrel-endcap) events, while dimuon mass-resolution uncertainties are 15% in 2016 and 8.5% in 2017 and 2018.In 2017 and 2018, the dimuon mass-scale uncertainty reaches 1–2% at mµµ = 5 TeV; dielectron mass-resolution uncertainty is negligible.Source: §7 · p. 15
Background modeling
- High-mass Drell–Yan theoretical uncertainties reach 20% at mℓℓ = 6 TeV, while simulated subdominant backgrounds receive a 7% uncertainty.Data-driven jet-misidentification backgrounds receive 50% uncertainty in both channels and years, with negligible impact from further high-mass extrapolation uncertainties.Source: §7 · p. 16
Signal modeling
- Nonresonant PDF reweighting contributes a mass-dependent uncertainty reaching 30% at 5 TeV and changes limits on Λ and ΛT by up to 5%.This uncertainty applies to reweighted 2017 and 2018 samples and does not affect resonant signals because no signal simulation is used there.Source: §7 · p. 16
Statistical interpretation
- Resonant interpretations retain only mass-dependent signal-modeling uncertainties after data normalization, whereas nonresonant limits and Rµ+µ−/e+e− use the full uncertainty set.The full set also enters data-to-simulation uncertainty bands, with nonresonant background impacts summarized for 1 and 3 TeV thresholds in Table 3.Source: §7 · pp. 16–17, Table 2 · p. 16, Table 3 · p. 17
8 Results
The 2016–2018 dielectron and dimuon invariant-mass distributions agree with the expected SM backgrounds overall, with a slight dielectron excess above 1.8 TeV but no identified pathologies. Four events are observed above 3 TeV, while angularly split distributions illustrate the improved signal-to-background ratio for contact-interaction signals at cos θ* < 0.
Invariant-mass results
- Four events are observed above 3 TeV—two dielectron events at 3.35 and 3.47 TeV and two dimuon events at 3.07 and 3.34 TeV—amid overall agreement with SM backgrounds.A slight excess is seen in the dielectron channel above 1.8 TeV, but inspection reveals no pathologies.Source: §8 · p. 17
Event yields
- Observed and expected yields are reported for multiple dielectron and dimuon mass ranges, with correlated background uncertainties included in the totals.The dimuon jet-background estimate is unavailable in the 60 < mℓℓ < 120 GeV control region but is negligible there because of the large Z-boson cross section.Source: §8 · p. 17, Table 4 · p. 20
Nonresonant signals
- Splitting the spectra by cos θ* illustrates an improved signal-to-background ratio for contact-interaction signals in the cos θ* < 0 bin.The distributions use the mass bins adopted for the contact-interaction statistical interpretation, with a constructive LL signal shown for illustration.Source: §8 · p. 17, Figure 3 · p. 19
Invariant-mass results
- The dielectron and dimuon invariant-mass spectra agree with SM expectations across the combined 2016–2018 data sets, with simulated GKK and Z′ SSM signals shown for illustration.Figure 2 includes the trigger selections and background uncertainty bands used for the normalization and signal regions.Source: §8 · p. 17, Figure 2 · p. 18
9 Statistical interpretation
The analysis uses combined Bayesian likelihood fits to interpret resonant and nonresonant dilepton signals at 95% confidence, while testing lepton flavor universality through an unfolded muon-to-electron cross-section ratio. No statistically significant departure from the Standard Model expectation is established, and stringent limits are obtained across resonance, contact-interaction, extra-dimension, dark-matter, and flavor-universality models.
9 Statistical interpretation
- The statistical interpretation uses 95% confidence Bayesian limits with unbinned resonance fits and binned nonresonance fits simultaneously combining channels, years, η categories, and relevant angular bins.Signal cross sections use positive uniform priors, while efficiency, resolution, scale, and other nuisance parameters use log-normal priors.Source: §9 · p. 20
9.1 Search for resonant signals
- The combined limits are consistent with Standard Model backgrounds, excluding Z′_SSM and Z′_ψ masses below 5.15 and 4.56 TeV, respectively.These observed limits apply to narrow resonances with intrinsic width 0.6%.Source: §9.1 · p. 22
- Spin-2 resonance mass limits range from slightly above 2 to nearly 5 TeV, improving previous CMS limits by approximately 370–530 GeV, while mediator exclusions reach 1.92 and 4.64 TeV for vector and axial-vector models.For m_DM = 0, the axial-vector and vector exclusions are 3.41 and 1.04 TeV, respectively.Source: §9.1 · pp. 24–25, Table 6 · p. 24, Figure 9 · p. 28
9.2 Search for nonresonant signals
- ADD ultraviolet-cutoff limits reach 7.5 TeV in the combined GRW convention, while the considered conventions yield 5.9–8.9 TeV and improve previous limits by 0.5–1.0 TeV.The combined GRW observed limit is 7.5 TeV, compared with an expected 7.8 TeV, after applying the NNLO correction factor.Source: §9.2 · p. 27, Table 7 · p. 30
- CI limits on Λ range from 23.9 to 36.4 TeV, improve previous CMS results by 3.5–4.5 TeV, and gain approximately 1–3 TeV from cos θ∗ splitting.The improvement from angular splitting is approximately 1 TeV for destructive interference and approximately 3 TeV for constructive interference.Source: §9.2 · p. 28, Figure 11 · p. 31
9.3 Search for lepton flavor universality violation
- The unfolded R_μ+μ−/e+e− ratio agrees with unity up to 1.5 TeV; above 400 GeV, the combined χ2/dof is 17.9/7 with one-sided p-value 0.012.The measurement provides a particle-level basis for testing models that predict lepton flavor universality violation.Source: §9.3 · p. 30
10 Summary
The CMS search of high-mass dielectron and dimuon spectra at 13 TeV using up to 140 fb^-1 found no significant deviation from Standard Model expectations. It constrained resonant and nonresonant new phenomena, including spin-1 and spin-2 resonances, dark-matter mediators, contact interactions, extra dimensions, and lepton flavor universality.
Overall search
- No significant deviation from Standard Model expectations is observed in the high-mass dilepton spectrum.The analysis used up to 140 fb^-1 of proton-proton collision data, with simulation-based backgrounds and data-driven normalization near resonance sidebands or the Z-boson peak.Source: §10 · p. 31
Resonant searches
- Lower mass limits of 2.47–4.78 TeV are set for Randall–Sundrum spin-2 gravitons, and the spin-1 and spin-2 resonance limits are the most stringent to date.The resonance limits are interpreted for narrow signals, SSM and superstring-inspired spin-1 models, and generalized Z′ quark couplings.Source: §10 · pp. 31–32
Dark-matter mediators
- Mediator masses below 1.92 (4.64) TeV are excluded for vector (axial-vector) dark-matter mediators at large mDM, decreasing to 1.04 and 3.41 TeV for mDM = 0.The limits are set in the mediator–dark-matter mass plane, with the stated values corresponding to small and large lepton couplings, respectively.Source: §10 · p. 32
Nonresonant searches
- Lower limits on the contact-interaction cutoff Λ range from 23.8 to 36.4 TeV, while ADD-model cutoff limits range from 5.9 to 8.9 TeV.The contact-interaction range depends on helicity structure and interference sign with the Standard Model Drell–Yan background; the ADD range depends on parameter convention.Source: §10 · p. 32
Lepton flavor universality
- The dimuon-to-dielectron differential cross-section ratio is compared at the TeV scale after detector corrections, with no significant deviation from lepton flavor universality.The comparison includes barrel, endcap, and combined event categories, with statistical and systematic uncertainties.Source: §10 · p. 32, Figure 12 · p. 32