Observation and measurement of forward proton scattering in association with lepton pairs produced via the photon fusion mechanism at ATLAS

The observation of forward proton scattering in association with lepton pairs ( e þ e − þ p or μ þ μ − þ p ) produced via photon fusion is presented. The scattered proton is detected by the ATLAS Forward Proton spectrometer, while the leptons are reconstructed by the central ATLAS detector. Proton-proton collision data recorded in 2017 at a center-of-mass energy of ﬃﬃﬃ s p ¼ 13 TeV are analyzed, corresponding to an integrated luminosity of 14 . 6 fb − 1 . A total of 57 (123) candidates in the ee þ p ( μμ þ p ) final state are selected, allowing the background-only hypothesis to be rejected with a significance exceeding 5 standard deviations in each channel. Proton-tagging techniques are introduced for cross-section measurements in the fiducial detector acceptance, corresponding to σ ee þ p ¼ 11 . 0 (cid:2) 2 . 6 ð stat Þ (cid:2) 1 . 2 ð syst Þ (cid:2) 0 . 3 ð lumi Þ and σ μμ þ p ¼ 7 . 2 (cid:2) 1 . 6 ð stat Þ (cid:2) 0 . 9 ð syst Þ (cid:2) 0 . 2 ð lumi Þ fb in the dielectron and dimuon channel, respectively. DOI: 10.1103/PhysRevLett.125.261801

Measuring proton-tagged dilepton production, pp → p(γγ → + − )p ( * ) , where p ( * ) denotes a proton that remains intact or dissociates following electromagnetic excitation, is important for several reasons.Predictions of photon fusion processes have significant uncertainties associated with modeling strong-force interactions between scattered protons, which suppress cross-sections by factors known as soft-survival probabilities [29][30][31][32].This suppression is poorly constrained, especially at high γγ invariant masses important for new physics searches, as existing probes indirectly infer dissociation rates using only centraldetector information [4][5][6][7].Proton-tagging overcomes this longstanding experimental ambiguity by directly detecting the scattered protons.Detecting a proton also directly suppresses background processes and events involving proton dissociation, while enabling reconstruction of the initial γγ system independently of central-detector information.The successful demonstration of proton-tagging techniques for cross-section measurements accomplishes the crucial first step towards a diverse program using proton-tagging in measurements of Standard Model processes [33][34][35][36][37][38] and searches for new phenomena [39][40][41][42][43].
This Letter introduces proton-tagging for cross-section measurements of pp → p(γγ → + − )p ( * ) .The ATLAS Forward Proton (AFP) spectrometer detects one of the intact protons and the central ATLAS detector reconstructs the leptons.The dataset was collected in 2017 and corresponds to 14.6 fb −1 of √ s = 13 TeV proton-proton (pp) collisions.The average number of interactions per bunch crossing was 36.Several methods specific to proton-tagging are introduced: in situ calibration of proton kinematics using the dimuon system, a novel data-mixing background estimation method, and tag-and-probe determination of the AFP reconstruction efficiency.
The AFP spectrometer [53,54] consists of four tracking units located along the beampipe at z = ±205 m and ±217 m, referred to as Near and Far stations, respectively.The +z (−z) direction is labeled side A (C).Each station houses a silicon tracker comprising four planes of edgeless silicon pixel sensors [55][56][57][58].The sensors have 336 × 80 pixels with area 50 × 250 µm 2 .The direction normal to each sensor is tilted 14 • relative to the beam to improve hit efficiency and x-position resolution, resulting in an overall spatial resolution of σ x = 6 µm [59].Movable near-beam devices at each station, known as Roman Pots, insert the tracker along the x-direction in the beampipe.Data-taking with the AFP commences once the trackers are at a position where the innermost silicon edge is within 2 mm of the beam center during stable beams.
Data quality for this analysis requires that every AFP station has at least three silicon planes operational at high voltage, and the AFP data acquisition system [60] must report no problems.Simulated events of the exclusive signal, pp → p(γγ → + − )p, were produced using the H 7 Monte Carlo (MC) generator [61,62].The single-dissociative signal, pp → p(γγ → + − )p * , was generated using L 4.0 [63], with proton dissociation modeled using the Brasse [64] and Suri-Yennie [65] structure functions interfaced with J S 7.408 [66,67].Simulation of these processes is detailed in Ref. [5].To model the central detector response, the exclusive signal underwent full detector simulation based on G 4 [68].The single-dissociative samples employed a fast simulation [69], which uses a parameterization of the calorimeter response [70].The response of the AFP spectrometer is modeled by a fast simulation, where a Gaussian smearing is applied to track positions based on the AFP spatial resolution.Simulated samples include the effect on the central detector of multiple pp interactions in the same and neighboring bunch crossing (pileup), as detailed in Ref. [5].
Reconstructed events must contain at least one interaction vertex with two or more associated inner-detector tracks that satisfy p T > 500 MeV, |η| < 2.5, and the Loose criterion [71,72] Selected events must have exactly two same-flavor leptons with opposite electric charge (e + e − or µ + µ − ) and be matched to the leptons that triggered the event.To suppress quarkonia and Z boson resonances, the dilepton invariant mass must satisfy m > 20 GeV and m [70,105] GeV.To select events compatible with pp → p(γγ → + − )p ( * ) processes based on the simulated signals, the dilepton transverse momentum must satisfy p T < 5 GeV.This set of criteria is referred to as the preselection.Signal event candidates must additionally have small acoplanarity A φ = 1 − |∆φ |/π < 0.01.These events must have no inner-detector tracks (N 0.5 mm tracks = 0) that satisfy ∆R(track, ) > 0.01 for both leptons and |z track 0 − z 0 | < 0.5 mm, where z track 0 is the track z 0 position and z 0 = (z 1 0 + z 2 0 )/2 with 1,2 denoting the two leptons.The expected proton energy loss based on lepton kinematics, ξ , is determined from m and the dilepton rapidity y by momentum conservation ξ ± = (m / √ s)e ±y , where + (−) corresponds to the proton on side A (C).
Reconstruction of scattered protons combines information from the AFP tracker and LHC magnet lattice [78].
Protons transported to the AFP leave hits in the silicon tracker, which are processed by clustering and track-finding algorithms detailed in Ref. [56].Tracks are reconstructed from clusters in at least two planes.Small corrections of around 0.1 mm are applied to the cluster positions to account for misalignment between planes.The proton transport function x AFP = T(ξ AFP ) relates the track x-position, x AFP , to the fractional energy loss of the scattered proton, ξ AFP = 1 − E scattered /E beam , where E scattered (E beam ) is the scattered (beam) proton energy.The LHC magnets and beam optics [79] govern the form of T(ξ AFP ) [80], which is simulated in the MAD-X package [81,82] with further details discussed in Refs.[53,83,84].Determination of ξ AFP uses both the Near and Far stations if tracks are within their common acceptance, otherwise only the Far station is used.
The absolute scale of E scattered depends on the closest separation, x s 0 , between each AFP station s and the beam center [84].The beam positions relative to the detectors were determined in dedicated runs with beam-based alignment procedures [85] using Beam Loss Monitors [86], and cross-checked with Beam Position Monitor measurements [87].In early 2017, the x s 0 values were initially set to −4.0 (−3.0) mm on side A and −3.8 (−2.9) mm on side C for the Near (Far) stations; during a second data-taking period, all stations were moved 0.5 mm closer to the beam to improve acceptance.This first (second) data-taking period corresponds to 5% (17%) of the analyzed dataset.For the remaining dataset, the Far stations were moved a further 0.2 mm towards the beam.The initially measured x AFP values relative to x s 0 are calibrated in situ using the dimuon data sample passing the signal event selection.The x s − x s AFP distribution is peaked for signal processes due to the kinematic correlation between x s and x s AFP , where x = T(ξ ) is the expected position calculated using the transport function.Additive corrections are applied to x s AFP in data to center the maximum of the peak at zero.These corrections are found to be −0.28 (−0.34) mm on side A and −0.17 (−0.36) mm on side C for the Near (Far) stations.Selected dielectron events are used to verify that the signal is centered at zero.After applying these corrections, the lower value of the acceptance corresponds to ξ A AFP > 0.028 (0.018) on side A and ξ C AFP > 0.026 (0.019) on side C for the Near (Far) stations.The upper value of the acceptance is bounded by ξ AFP < 0.12 due to the presence of beam collimators [53].
To select events with one or more proton candidates, the ξ and ξ AFP values for at least one AFP side are required to be within the range [0.02, 0.12].If there is more than one proton candidate on the same AFP side, which occurs in 35% of selected events, the proton with ξ AFP closest to ξ is chosen.Proton-tagged dilepton candidates, denoted + p, are selected by requiring kinematic matching on at least one AFP side, |ξ AFP − ξ | < 0.005, which retains (rejects) more than 95% (85%) of the signal (background).
The dominant source of background after this selection arises from lepton pairs produced in a pp interaction different from that of the detected proton.In this case, the lepton pairs are produced via the Drell-Yan mechanism as well as γγ → + − processes in which any outgoing protons are either outside the AFP acceptance or not reconstructed in AFP due to detector inefficiency.These events are collectively referred to as combinatorial backgrounds and are estimated using a data-driven method.A mixed-data sample is constructed by randomly pairing each measured ξ value, passing AFP acceptance ξ AFP ∈ [0.02, 0.12], with 100 values of ξ AFP from a large control sample of > 10 6 events.This control sample is constructed from the preselected events and requiring A φ > 0.01.The 123 selected data events failing kinematic matching, |ξ AFP − ξ | > 0.005, result mostly from combinatorial background processes, which are used to normalize the mixed-data sample using a background-only profile-likelihood fit [88,89].
Systematic uncertainties in the background normalization arise from the limited size of the data sample satisfying |ξ AFP − ξ | > 0.005.An uncertainty in the background shape arises from kinematic changes in the control sample of protons due to the acoplanarity requirement.This uncertainty is estimated by replacing the A φ > 0.01 condition with N 0.5 mm tracks ≥ 1 and comparing the two background predictions in the region |ξ AFP − ξ | < 0.005; they are found to differ by 14%.Further shape uncertainties arise from instrumental effects, which are expected to be dominated by the sensitivity to the number of interactions per bunch crossing, µ.The background predictions for µ < 35 and µ ≥ 35 are found to differ by 8% in the |ξ AFP − ξ | < 0.005 region.These two shape differences are assigned as additional uncertainties.
The background estimation method is validated by applying it to the orthogonal m ∈ [70, 105] GeV region.The region |ξ AFP − ξ | > 0.005 is dominated by Drell-Yan events, which have no correlated protons.In this region, the data and prediction from the mixed-data sample are found to be compatible within the uncertainties across the ξ AFP − ξ range for both sides A and C.
After applying the event selection including kinematic matching, |ξ AFP − ξ | < 0.005, a total of 57 (123) candidates in the ee + p (µµ + p) final state are observed compared with a background-only expectation of 6.2 ± 1.2 (13.4 ± 2.5) events.Using the asymptotic profile-likelihood method [88,89], the background-only hypothesis is rejected with a significance exceeding 5σ in each channel [90].This provides direct evidence  of forward proton scattering in association with electron and muon pairs produced via photon fusion.The ξ AFP − ξ distributions of data, signal and background at detector-level before kinematic matching are shown in Figure 1.To illustrate the expected composition of the signal, the simulated samples are normalized to data with sides A and C combined and fit separately in the ee and µµ channels.Figure 2 displays positions in the y -m plane of data candidates satisfying |ξ AFP − ξ | < 0.005 on at least one side and the corresponding acceptance regions of the four AFP stations.The highest-mass ee candidate has an invariant mass m = 717 GeV and rapidity y = 0.252, so the scattered protons would be within the acceptance of both AFP sides if this were an exclusive process.However, it is found that the proton on side A fails kinematic matching |ξ AFP − ξ | < 0.005, so this event is likely a single-dissociative process where the side A proton candidate originates from a pileup interaction.The corresponding quantities for the highest-mass µµ candidate are m = 319 GeV and y = 0.255.Figure 3 illustrates detector-level distributions of dilepton acoplanarity, mass and rapidity after kinematic matching with the signal samples normalized to N obs − N bkg .
Cross-sections are measured in a fiducial region defined at particle-level with an event selection similar to that applied at detector level [91].To reliably estimate AFP reconstruction efficiencies using tag-and-probe techniques, the ξ AFP and ξ values are restricted to a tighter range [0.035, 0.08] and each proton candidate is required to have an associated track in both Near and Far stations.The measured cross-sections are defined by Here, N obs (N bkg ) is the number of observed data (expected background) events passing event selection, and C cent (C AFP ) is an overall correction factor accounting for the central-detector (AFP) efficiency.The integrated luminosity, L = 14.6 fb −1 , is measured using the LUCID-2 detector [92] and the uncertainty is determined to be 2.4% [93].In this tighter region, N obs is found to be 19 (23) for the ee (µµ) channel and N bkg = 1.7 ± 0.3 (2.3 ± 0.5).The event rate between the two channels differs more for the ξ ∈ [0.02, 0.12] than ξ ∈ [0.035, 0.08] region because µµ events with low m and high |y | have greater selection efficiency due to trigger and reconstruction requirements.
The C cent factor is defined as the ratio of the number of MC events passing detector-level selection to the number passing the particle-level fiducial requirements.Uncertainties in C cent are estimated by varying the electron (muon) energy (momentum) scale and resolution, and data/MC correction factors described in Refs.[73,74], together with corrections applied to account for pileup modeling.The dominant uncertainties   for ee events arise from pileup modeling (2%) and identification (1%), while for µµ events, these correspond to pileup modeling (3%), resolution (3%), and scale (2%); other sources such as trigger and isolation efficiencies contribute 1% or less.Using data-driven methods described in Ref. [5], a further correction of 0.89 ± 0.04 is applied to C cent to account for data/MC differences in modeling the luminous region at the interaction point.The 5% uncertainty in this correction is evaluated as the difference between either applying this data-driven method to simulated signal samples or imposing the N 0.5 mm tracks = 0 requirement on these samples.Overall, this results in C ee cent = 0.12 ± 0.01 (C µµ cent = 0.22 ± 0.02) for the ee (µµ) channel.The C AFP factor is defined by the product track • smear .The track reconstruction efficiency, track , is found to be 0.92 ± 0.02 for sides A and C. The Near station efficiency is estimated using a tag-and-probe method by first selecting events with exactly one track in the Far (tag) station in the acceptance common to both stations, −12 < x AFP < −5 mm.The efficiency is the fraction of these events that also have one or more tracks in the Near (probe) station satisfying |x Near − x Far | < 2 mm.The tag and probe stations are inverted to measure the Far station efficiency.It is found that track varies with ξ AFP by 2%, which is assigned as an additional   uncertainty.The proton resolution correction, smear , is found to be 0.98 ± 0.02 (0.96 ± 0.04) for the ee (µµ) channel.This is evaluated as the fraction of simulated signal events passing ξ AFP , ξ ∈ [0.035, 0.08] and |ξ AFP − ξ | < 0.005 out of those satisfying ξ ∈ [0.035, 0.08].Uncertainties in C AFP are dominated by global alignment (6%) evaluated by ±0.3 mm variations of x AFP , and beam optics (5%) evaluated by varying the beam crossing angle by 50 µrad in the MAD-X package.Uncertainties involving track and cluster reconstruction are found to be less than 1%.The overall uncertainty in C AFP is 9%.
The measured fiducial cross-sections in the ee and µµ channels are σ fid.ee+p = 11.0 ± 2.6 (stat.)± 1.2 (syst.)± 0.3 (lumi.)fb and σ fid.µµ+p = 7.2 ± 1.6 (stat.)± 0.9 (syst.)± 0.2 (lumi.)fb, respectively.Table 1 compares these with the combined H and L predictions assuming unit soft-survival factors S surv = 1.Soft-survival effects are included using an m -dependent reweighting of these predictions to S surv calculated for exclusive processes from Ref. [31]; L predictions are additionally scaled down by 15% to account for S surv being lower for single-dissociative processes [30].S C 4 [94] predictions include full kinematic dependence on S surv for exclusive, single-and double-dissociative processes.The predictions for ee are higher than for µµ due to the looser η(e) requirement [91].
In summary, forward proton scattering in association with lepton pairs produced via photon fusion, pp → p(γγ → + − )p ( * ) , is observed with a significance exceeding 5σ in both the ee + p and µµ + p final states using 14.6 fb −1 of √ s = 13 TeV pp collisions at the LHC.These results demonstrate that the ATLAS Forward Proton spectrometer performs well in high-luminosity data-taking.Furthermore, proton-tagging is introduced for cross-section measurements of photon fusion processes at the electroweak scale.[75] z 0 is the longitudinal impact parameter relative to the primary vertex, where the primary vertex is defined as the vertex with the largest p 2 T of associated tracks.[78] L. Evans and P. Bryant, LHC Machine, JINST 3 (2008) S08001.[95] Uncertainties on predicted soft-survival factors are estimated in accord with Ref. [30].For the exclusive process, the uncertainty on S surv is estimated by the m variations, while for the singledissociative process, the uncertainty on S surv is estimated by taking the difference in S surv between the exclusive and single-dissociative processes.
States of America.

Figure 1 :
Figure 1: Distributions of ξ AFP − ξ with ξ and ξ AFP satisfying [0.02, 0.12] for side A (left) and side C (right).The total prediction comprises the signal and combinatorial background processes, where p * denotes a dissociated proton.The simulated predictions are normalized to data to illustrate the expected signal composition.The first (last) bin includes underflow (overflow).The hatched band indicates the combined statistical and systematic uncertainties of the prediction.Error bars denote statistical uncertainties of the data.

Figure 2 :
Figure 2: Data event candidates in the dilepton rapidity y vs. m plane satisfying event selection and kinematic matching, |ξ AFP − ξ | < 0.005, on at least one side.Shaded (hatched) areas denote the acceptance (no acceptance) for the AFP stations indicated in the legend.Areas neither shaded nor hatched correspond to ξ [0, 1].

Figure 3 :
Figure 3: Distributions of dilepton acoplanarity A φ (left), invariant mass m (center), rapidity y (right) satisfying ξ , ξ AFP ∈ [0.02, 0.12] and |ξ AFP − ξ | < 0.005 for at least one AFP side.Events with 70 < m < 105 GeV are vetoed.The total prediction comprises the signal and combinatorial background processes, where p * denotes a dissociated proton.The simulated predictions are normalized to data to illustrate the expected signal composition.The first (last) bin includes underflow (overflow).The hatched band indicates the combined statistical and systematic uncertainties of the prediction.Error bars denote statistical uncertainties of the data.

Table 1 :
Fidu31]l cross-sections from the combined H and L predictions with S surv = 1 and S surv estimated using Refs.[30,31]asdescribed in the main text.S C 4 [94] predictions include fully kinematically-dependent S surv .Uncertainties of 7% (17%) are assigned for predictions of the exclusive (single-dissociative) processes[95].The bottom row displays the measured cross-sections with statistical and systematic uncertainties combined. 001 Also at Dipartimento di Matematica, Informatica e Fisica, Università di Udine, Udine; Italy.s Also at Faculty of Physics, M.V. Lomonosov Moscow State University, Moscow; Russia.t Also at Giresun University, Faculty of Engineering, Giresun; Turkey.u Also at Graduate School of Science, Osaka University, Osaka; Japan.Also at Institucio Catalana de Recerca i Estudis Avancats, ICREA, Barcelona; Spain.x Also at Institut für Experimentalphysik, Universität Hamburg, Hamburg; Germany.y Also at Institute for Nuclear Research and Nuclear Energy (INRNE) of the Bulgarian Academy of Sciences, Sofia; Bulgaria.z Also at Institute for Particle and Nuclear Physics, Wigner Research Centre for Physics, Budapest; Also at Instituto de Fisica Teorica, IFT-UAM/CSIC, Madrid; Spain.ad Also at Joint Institute for Nuclear Research, Dubna; Russia.ae Also at Moscow Institute of Physics and Technology State University, Dolgoprudny; Russia.a f Also at National Research Nuclear University MEPhI, Moscow; Russia.ag Also at Physics Department, An-Najah National University, Nablus; Palestine.ah Also at Physikalisches Institut, Albert-Ludwigs-Universität Freiburg, Freiburg; Germany.ai Also at The City College of New York, New York NY; United States of America.a j Also at TRIUMF, Vancouver BC; Canada.ak Also at Universita di Napoli Parthenope, Napoli; Italy.al Also at University of Chinese Academy of Sciences (UCAS), Beijing; China.
b Also at Center for High Energy Physics, Peking University; China.c Also at Centro Studi e Ricerche Enrico Fermi; Italy.d Also at CERN, Geneva; Switzerland.e Also at CPPM, Aix-Marseille Université, CNRS/IN2P3, Marseille; France.f Also at Département de Physique Nucléaire et Corpusculaire, Université de Genève, Genève; Switzerland.g Also at Departament de Fisica de la Universitat Autonoma de Barcelona, Barcelona; Spain.h Also at Department of Financial and Management Engineering, University of the Aegean, Chios; Greece.i Also at Department of Physics and Astronomy, Michigan State University, East Lansing MI; United States of America.j Also at Department of Physics and Astronomy, University of Louisville, Louisville, KY; United States of America.k Also at Department of Physics, Ben Gurion University of the Negev, Beer Sheva; Israel.l Also at Department of Physics, California State University, East Bay; United States of America.m Also at Department of Physics, California State University, Fresno; United States of America.n Also at Department of Physics, California State University, Sacramento; United States of America.o Also at Department of Physics, King's College London, London; United Kingdom.p Also at Department of Physics, St. Petersburg State Polytechnical University, St. Petersburg; Russia.q Also at Department of Physics, University of Fribourg, Fribourg; Switzerland.r v Also at Hellenic Open University, Patras; Greece.w aa Also at Institute of Particle Physics (IPP); Canada.ab Also at Institute of Physics, Azerbaijan Academy of Sciences, Baku; Azerbaijan.ac * Deceased