Study of the production of charged pions, kaons, and protons in pPb collisions at sqrt(sNN) = 5.02 TeV

Spectra of identified charged hadrons are measured in pPb collisions with the CMS detector at the LHC at sqrt(sNN) = 5.02 TeV. Charged pions, kaons, and protons in the transverse-momentum range pt approximately 0.1-1.7 GeV and laboratory rapidity abs(y)<1 are identified via their energy loss in the silicon tracker. The average pt increases with particle mass and the charged multiplicity of the event. The increase of the average pt with charged multiplicity is greater for heavier hadrons. Comparisons to Monte Carlo event generators reveal that EPOS LHC, which incorporates additional hydrodynamic evolution of the created system, is able to reproduce most of the data features, unlike HIJING and AMPT. The pt spectra and integrated yields are also compared to those measured in pp and PbPb collisions at various energies. The average transverse momentum and particle ratio measurements indicate that particle production at LHC energies is strongly correlated with event particle multiplicity.


Introduction
The study of hadron production has a long history in high-energy particle and nuclear physics, as well as in cosmic-ray physics. The absolute yields and the transverse momentum (p T ) spectra of identified hadrons in high-energy hadron-hadron collisions are among the most basic physical observables. They can be used to test the predictions for non-perturbative quantum chromodynamics (QCD) processes like hadronization and soft-parton interactions, and the validity of their implementation in Monte Carlo (MC) event generators. Spectra of identified particles in proton-nucleus collisions also constitute an important reference for studies of highenergy heavy-ion collisions, where final-state effects are known to modify the spectral shape and yields of different hadron species [1][2][3][4].
The present analysis focuses on the measurement of the p T spectra of charged hadrons, identified mostly via their energy deposits in silicon detectors, in pPb collisions at √ s NN = 5.02 TeV.
The analysis procedures are similar to those previously used in the measurement of pion, kaon, and proton production in pp collisions at several center-of-mass energies [5].
A detailed description of the CMS (Compact Muon Solenoid) detector can be found in Ref. [6]. The CMS experiment uses a right-handed coordinate system, with the origin at the nominal interaction point (IP) and the z axis along the counterclockwise-beam direction. The pseudorapidity η and rapidity y of a particle (in the laboratory frame) with energy E, momentum p, and momentum along the z axis p z are defined as η = − ln[tan(θ/2)], where θ is the polar angle with respect to the z axis and y = 1 2 ln[(E + p z )/(E − p z )], respectively. The central feature of the CMS apparatus is a superconducting solenoid of 6 m internal diameter. Within the 3.8 T field volume are the silicon pixel and strip tracker, the crystal electromagnetic calorimeter, and the brass/scintillator hadron calorimeter. The tracker measures charged particles within the pseudorapidity range |η| < 2.4. It has 1440 silicon pixel and 15 148 silicon strip detector modules. In addition to the barrel and endcap detectors, CMS has extensive forward calorimetry. Steel/quartz-fiber forward calorimeters (HF) cover 3 < |η| < 5. Beam Pick-up Timing for the eXperiments (BPTX) devices were used to trigger the detector readout. They are located around the beam pipe at a distance of 175 m from the IP on either side, and are designed to provide precise information on the Large Hadron Collider (LHC) bunch structure and timing of the incoming beams.
The reconstruction of charged particles in CMS is bounded by the acceptance of the tracker (|η| < 2.4) and by the decreasing tracking efficiency at low momentum (greater than about 60% for p > 0.05, 0.10, 0.20, and 0.40 GeV/c for e, π, K, and p, respectively). Particle identification capabilities using specific ionization are restricted to p < 0.15 GeV/c for electrons, p < 1.20 GeV/c for pions, p < 1.05 GeV/c for kaons, and p < 1.70 GeV/c for protons. Pions are identified up to a higher momentum than kaons because of their high relative abundance. In view of the (y, p T ) regions where pions, kaons, and protons can all be identified (p = p T cosh y), the band −1 < y < 1 (in the laboratory frame) was chosen for this measurement, since it is a good compromise between the p T range and y coverage.
In this paper, comparisons are made to predictions from three MC event generators. The HI-JING [7] event generator is based on a two-component model for hadron production in highenergy nucleon and nuclear collisions. Hard parton scatterings are assumed to be described by perturbative QCD and soft interactions are approximated by string excitations with an effective cross section. In version 2.1, in addition to modification of initial parton distributions, multiple scatterings inside a nucleus lead to transverse momentum broadening of both initial and final-state partons, responsible for the enhancement of intermediate-p T hadron spectra in proton-nucleus collisions. The AMPT [8] event generator is a multi-phase transport model.

Data analysis
It starts from the same initial conditions as HIJING, contains a partonic transport phase, the description of the bulk hadronization, and finally a hadronic rescattering phase. The latest available version (1.26/2.26) is used. The EPOS [9] event generator uses a quantum mechanical multiple scattering approach based on partons and strings, where cross sections and particle production are calculated consistently, taking into account energy conservation in both cases. Nuclear effects related to transverse momentum broadening, parton saturation, and screening have been introduced. The model can be used both for extensive air shower simulations and accelerator physics. The latest version, EPOS LHC [10], contains a three-dimensional viscous event-by-event hydrodynamic treatment. This is a major difference with respect to the HIJING and AMPT models.

Data analysis
The data were taken in September 2012 during a 4-hour-long pPb run with very low multipleinteraction rate (0.15% "pileup"). A total of 2.0 million collisions were collected, corresponding to an integrated luminosity of approximately 1 µb −1 . As the dominant uncertainty is systematic in nature, the much larger 2013 pPb dataset was not needed for the analysis. The beam energies were 4 TeV for protons and 1.58 TeV per nucleon for lead nuclei, resulting in a centerof-mass energy per nucleon pair of √ s NN = 5.02 TeV. Due to the asymmetric beam energies the nucleon-nucleon center-of-mass in the pPb collisions is not at rest with respect to the laboratory frame but moves with a velocity β = −0.434 or rapidity −0.465. Since the higher-energy proton beam traveled in the clockwise direction, i.e. at θ = π, the rapidity of a particle emitted at y cm in the nucleon-nucleon center-of-mass frame is detected in the laboratory frame with a shift, y − y cm = −0.465, i.e. a particle with y = 0 moves with rapidity 0.465 in the Pb-beam direction in the center-of-mass system. The particle yields reported in this paper have been measured for laboratory rapidity |y| < 1.
The event selection consisted of the following requirements: • at the trigger level, the coincidence of signals from both BPTX devices, indicating the presence of both proton and lead bunches crossing the interaction point; in addition, at least one track with p T > 0.4 GeV/c in the pixel tracker; • offline, the presence of at least one tower with energy above 3 GeV in each of the HF calorimeters; at least one reconstructed interaction vertex; beam-halo and beaminduced background events, which usually produce an anomalously large number of pixel hits [11], are suppressed.
The efficiencies for event selection, tracking, and vertexing were evaluated using simulated event samples produced with the HIJING 2.1 MC event generator, where the CMS detector response simulation was based on GEANT4 [12]. Simulated events were reconstructed in the same way as collision data events. The final results were corrected to a particle level selection applied to the direct MC output, which is very similar to the data selection described above: at least one particle (proper lifetime τ > 10 −18 s) with E > 3 GeV in the range −5 < η < −3 and at least one in the range 3 < η < 5; this selection is referred to in the following as the "doublesided" (DS) selection. These requirements are expected to suppress single-diffractive collisions in both the data and MC samples. From the MC event generators studied in this paper, the DS selection efficiency for inelastic, hadronic collisions is found to be 94-97%.
The simulated ratio of the data selection efficiency to the DS selection efficiency is shown as a function of the reconstructed track multiplicity in the left panel of Fig. 1  without a reconstructed track. This fraction, as given by the simulation, is about 0.1%. Since these events do not contain reconstructed tracks, only the number of DS events N ev must be corrected.
The extrapolation of particle spectra into the unmeasured (y, p T ) regions is model dependent, particularly at low p T . A high-precision measurement therefore requires reliable track reconstruction down to the lowest possible p T . The present analysis extends to p T ≈ 0.1 GeV/c by exploiting special tracking algorithms [13], used in previous studies [5,11,14], to provide high reconstruction efficiency and low background rate. The charged-pion mass was assumed when fitting particle momenta.
When at least two hits are required in the pixel detector, the acceptance of the tracker (C a ) is flat in the region −2 < η < 2 and p T > 0.4 GeV/c, and its value is 96-98% ( Fig. 1, right panel). The loss of acceptance at p T < 0.4 GeV/c is caused by energy loss and multiple scattering of particles, both depending on the particle mass. Likewise, the reconstruction efficiency (C e ) is about 75-85%, degrading at low p T , also in a mass-dependent way. The misreconstructed-track rate (C f ) is very small, reaching 1% only for p T < 0.2 GeV/c. The probability of reconstructing multiple tracks (C m ) from a single true track is about 0.1%, mostly due to particles spiralling in the strong magnetic field of the CMS solenoid. The efficiencies and background rates largely factorize in η and p T , but for the final corrections an (η, p T ) matrix is used.
The region where pPb collisions occur (beam spot) is measured by reconstructing vertices from many events. Since the bunches are very narrow in the transverse direction, the xy location of the interaction vertices is well constrained; conversely, their z coordinates are spread over a relatively long distance and must be determined on an event-by-event basis. The vertex position is determined using reconstructed tracks which have p T > 0.1 GeV/c and originate from the vicinity of the beam spot, i.e. their transverse impact parameters d T satisfy the condition d T < 3 σ T .
Here σ T is the quadratic sum of the uncertainty in the value of d T and the root-mean-square of the beam spot distribution in the transverse plane. The agglomerative vertex-reconstruction algorithm [15] was used, with the z coordinates (and their uncertainties) of the tracks at the point of closest approach to the beam axis as input. For single-vertex events, there is no mini-

3 Estimation of energy loss rate and yield extraction
mum requirement on the number of tracks associated with the vertex, even one-track vertices are allowed. If multiple vertices are present, it is enough to keep the one with the highest track multiplicity. The resultant bias is negligible since the pileup rate is extremely small.
The distribution of the z coordinates of the reconstructed primary vertices is Gaussian, with a standard deviation of 7.1 cm. The simulated data were reweighted so as to have the same vertex z coordinate distribution as the data.
While the correction (C s ) is around 1% for pions, it rises up to 15% for protons with p T ≈ 0.2 GeV/c. As none of the mentioned weakly decaying particles decay into kaons, the correction for kaons is small. Based on studies comparing reconstructed K 0 S , Λ, and Λ spectra and predictions from the HIJING event generator, the corrections are reweighted by a p T -dependent factor.
For p < 0.15 GeV/c, electrons can be clearly identified. The overall e ± contamination of the hadron yields is below 0.2%. Although muons cannot be separated from pions, their fraction is very small, below 0.05%. Since both contaminations are negligible, no corrections are applied for them.

Estimation of energy loss rate and yield extraction
In this paper an analytical parametrization [16] has been used to approximate the energy loss of charged particles in the silicon detectors. The method provides the probability density P(∆|ε, l) of energy deposit ∆, if the most probable energy loss rate ε at a reference path-length l 0 = 450 µm and the path-length l are known. It was used in conjunction with a maximum likelihood estimation method.
For pixel clusters, the energy deposits were calculated as the sum of individual pixel deposits. In the case of strips, the energy deposits were corrected for capacitive coupling and cross-talk between neighboring strips. The readout threshold, the coupling parameter, and the standard deviation of the Gaussian noise for strips were determined from data, using tracks with closeto-normal incidence.
For an accurate determination of ε, the response of all readout chips was calibrated with multiplicative gain correction factors. The measured energy deposit spectra were compared to the energy loss parametrization and hit-level corrections (affine transformation of energy deposits using scale factors and shifts) were introduced. The corrections were applied to individual hits during the determination of the ln ε fit templates.
The best value of ε for each track was calculated with the corrected energy deposits by minimizing the joint energy deposit negative log-likelihood of all hits on the trajectory (index i), The ln ε values in (η, p T ) bins were then used in the yield unfolding. Hits with incompatible energy deposits (contributing to the joint χ 2 with a large amount) were excluded. At most one hit was removed; this affected about 1.5% of the tracks. Finally, fit templates, giving the expected ln ε distributions for all particle species (electrons, pions, kaons, and protons), were built from tracks. All kinematical parameters and hit-related observables were kept, but the energy deposits were regenerated by sampling from the analytical parametrization.
Distributions of ln ε as a function of total momentum p for positive particles are plotted in the left panel of Fig. 2 and compared to the predictions of the energy loss method for electrons, pions, kaons, and protons. The remaining deviations were taken into account by means of track-level residual corrections (affine transformation of templates using scale factors and shifts, ln ε → α ln ε + δ). Low-momentum particles can be identified unambiguously and can therefore be counted. Conversely, at high momentum, the ln ε bands overlap (above about 0.5 GeV/c for pions and kaons and 1.2 GeV/c for protons); the particle yields therefore need to be determined by means of a series of template fits in ln ε, in bins of η and p T (Fig. 2, right panel). For a less biased determination of track-level residual corrections, enhanced samples of each particle type were employed. For electrons and positrons, photon conversions in the beam-pipe and innermost first pixel layer were used. For high-purity π and enhanced p samples, weakly decaying hadrons were selected (K 0 S , Λ/Λ). Some relations and constraints were also exploited [5], this way better constraining the parameters of the fits: fitting the ln ε distributions in number of hits (n hits ) and track-fit χ 2 /ndf slices simultaneously; fixing the distribution n hits of particle species, relative to each other; using the expected continuity for refinement of track-level residual corrections, in neighboring (η, p T ) bins; using the expected convergence for track-level residual corrections, as the ln ε values of two particle species approach each other.
The results of the (iterative) ln ε fits are the yields for each particle species and charge in bins of (η, p T ) or (y, p T ), both inclusive and divided into classes of reconstructed primary chargedtrack multiplicity. In the end, the histogram fit χ 2 /ndf values were usually close to unity. Although pion and kaon yields could not be determined for p > 1.30 GeV/c, their sum was measured. This information is an important constraint when fitting the p T spectra.
The statistical uncertainties for the extracted yields are given by the fits. The observed local variations of parameters in the (η, p T ) plane for track-level corrections cannot be attributed to statistical fluctuations and indicate that the average systematic uncertainties in the scale factors and shifts are about 10 −2 and 2 · 10 −3 , respectively. The systematic uncertainties in the yields in each bin were obtained by refitting the histograms with the parameters changed by these

Corrections and systematic uncertainties
The measured yields in each (η, p T ) bin, ∆N measured , were first corrected for the misreconstructedtrack rate (C f ) and the fraction of secondary particles (C s ): The distributions were then unfolded to take into account the finite η and p T resolutions. The η distribution of the tracks is flat and the η resolution is very good. Conversely, the p T distribution is steep in the low-momentum region and separate corrections in each η bin were necessary. An unfolding procedure with linear regularization [18] was used, based on response matrices obtained from MC samples for each particle species.
The corrected yields were obtained by applying corrections for acceptance (C a ), efficiency (C e ), and multiple track reconstruction rate (C m ): where N ev is the corrected number of DS events (Fig. 1). Bins with acceptance smaller than 50%, efficiency smaller than 50%, multiple-track rate greater than 10%, or containing less than 80 tracks were not used.
Finally, the differential yields d 2 N/dη dp T were transformed to invariant yields d 2 N/dy dp T by multiplying with the Jacobian E/p and the (η, p T ) bins were mapped into a (y, p T ) grid. As expected, there is no strong y dependence in the narrow region considered (|y| < 1) and therefore the yields as a function of p T were obtained by averaging over rapidity.
The systematic uncertainties are very similar to those in Ref. [5] and are summarized in Table 1.
The uncertainties of the corrections related to the event selection and pileup are fully or mostly correlated and were treated as normalization uncertainties: 3.0% uncertainty on the yields and 1.0% on the average p T . The tracker acceptance and the track reconstruction efficiency generally have small uncertainties (1% and 3%, respectively), but change rapidly at very low p T (right panel of Fig. 1), leading to a 6% uncertainty on the yields in that range. For the multipletrack and misreconstructed-track rate corrections, the uncertainty is assumed to be 50% of the correction, while for the case of the correction for secondary particles it is 20%. The uncertainty of the fitted yields is in the range 1-10% depending mostly on total momentum.

Results
In previously published measurements of unidentified and identified particle spectra [11,19], the following form of the Tsallis-Pareto-type distribution [20,21] was fitted to the data: where and m T = m 2 + p 2 T (factors of c are omitted from the preceding formulae). The free parameters are the integrated yield dN/dy, the exponent n, and parameter T. The above formula is useful for extrapolating the spectra to p T = 0 and for extracting p T and dN/dy. Its validity was cross-checked by fitting MC spectra and verifying that the fitted values of p T and dN/dy were consistent with the generated values. According to some models of particle production based on non-extensive thermodynamics [21], the parameter T is connected with the average particle energy, while n characterizes the "non-extensivity" of the process, i.e. the departure of the spectra from a Boltzmann distribution (n = ∞).
As discussed earlier, pions and kaons cannot be unambiguously distinguished at higher momenta. Because of this, the pion-only (kaon-only) d 2 N/dy dp T distribution was fitted for |y| < 1 and p < 1.20 GeV/c (p < 1.05 GeV/c); the joint pion and kaon distribution was fitted in the region |η| < 1 and 1.05 < p < 1.5 GeV/c. Since the ratio p/E for the pions (which are more abundant than kaons) at these momenta can be approximated by p T /m T at η ≈ 0, Eq. (3) becomes: The approximate fractions of particles outside the measured p T range depend on track multiplicity; they are 15-30% for pions, 40-50% for kaons, and 20-35% for protons. The average transverse momentum p T and its uncertainty were obtained by numerical integration of Eq. (3) with the fitted parameters, under the assumption that the particle yield distributions follow the Tsallis-Pareto function in the unmeasured p T region. The results discussed in the following are for laboratory rapidity |y| < 1. In all cases, error bars indicate the uncorrelated statistical uncertainties, while boxes show the uncorrelated systematic uncertainties. The fully correlated normalization uncertainty is not shown. For the p T spectra, the average transverse momentum, and the ratio of particle yields, the data are compared to AMPT 1.26/2.26 [8], EPOS LHC [9,10], and HIJING 2.1 [7] MC event generators.

Inclusive measurements
The transverse momentum distributions of positively and negatively charged hadrons (pions, kaons, protons) are shown in Fig. 3, along with the results of the fits to the Tsallis-Pareto parametrization (Eqs. (3) and (5)). The fits are of good quality with χ 2 /ndf values in the range 0.4-2.8 ( Table 2). Figure 4 presents the data compared to the AMPT, EPOS LHC, and HIJING predictions. EPOS LHC gives a good description, while other generators predict steeper p T distributions than found in data. Ratios of particle yields as a function of the transverse momentum are plotted in Fig. 5. While the K/π ratios are well described by the AMPT simulation, only EPOS LHC is able to predict both K/π and p/π ratios. The ratios of the yields for oppositely charged particles are close to one, as expected for pair-produced particles at midrapidity.

Multiplicity dependent measurements
A study of the dependence on track multiplicity is motivated partly by the intriguing hadron correlations measured in pp and pPb collisions at high track multiplicities [22][23][24][25], suggesting possible collective effects in "central" pp and pPb collisions at the LHC. At the same time, it was seen that in pp collisions the characteristics of particle production ( p T , ratios) at LHC energies are strongly correlated with event particle multiplicity rather than with the centerof-mass energy of the collision [5]. In addition, the multiplicity dependence of particle yield ratios is sensitive to various final-state effects (hadronization, color reconnection, collective flow) implemented in MC models used in collider and cosmic-ray physics [26].
The event multiplicity N rec is obtained from the number of reconstructed tracks with |η| < 2.4, where the tracks are reconstructed using the same algorithm as for the identified charged hadrons [13]. (The multiplicity variable N offline trk , used in Ref. [23], is obtained from a different track reconstruction configuration and a value of N offline trk = 110 corresponds roughly to N rec = 170.) The event multiplicity was divided into 19 classes, defined in Table 3. To facilitate comparisons with models, the corresponding corrected charged particle multiplicity in the   same acceptance of |η| < 2.4 (N tracks ) is also determined. For each multiplicity class, the correction from N rec to N tracks uses the efficiency from the simulation with HIJING in (η, p T ) bins, followed by integration over the whole p T range of the corrected data in slices of η (with a linear extrapolation for p T < 0.1 GeV/c, down to zero yield at p T = 0), and finished by summing the integrals for each η slice. The average corrected charged-particle multiplicity N tracks , and also its values with the condition p T > 0.4 GeV/c, are shown in Table 3 for each event multiplicity class. The value of N tracks is used to identify the multiplicity class in Figs. 6-10.
The normalized transverse-momentum distributions of identified charged hadrons in selected multiplicity classes for |y| < 1 are shown in Fig. 6 for pions, kaons, and protons. The distribu-

13
tions of negatively and positively charged particles have been summed. The distributions are fitted with the Tsallis-Pareto parametrization. For kaons and protons, the parameter T increases with multiplicity, while for pions both T and the exponent n are independent of multiplicity (not shown).
The ratios of particle yields, obtained by integration of the fitted Tsallis-Pareto function over the whole p T range, are displayed as a function of track multiplicity in Fig. 7. The K/π and p/π ratios are flat, or slightly rising, as a function of N tracks . While none of the models is able to precisely reproduce the track multiplicity dependence, the best and worst matches to the overall scale are given by EPOS LHC and HIJING, respectively. The ratios of yields of oppositely charged particles are independent of N tracks as shown in the right panel of Fig. 7. The average transverse momentum p T is shown as a function of multiplicity in Fig. 8. As expected from the discrepancies between theory and data shown in Fig. 4, EPOS LHC again gives a reasonable description, while the other event generators presented here underpredict the measured values. For the dependence of T on multiplicity (not shown), the predictions match the pion data well; the kaon and proton values are much higher than in AMPT or HIJING.

Comparisons to pp and PbPb data
The comparison with pp data taken at various center-of-mass energies (0.9, 2.76, and 7 TeV) [5] is shown in Fig. 9, where the dependence of p T and the particle yield ratios (K/π and p/π) on the track multiplicity is shown. The plots also display the ranges of these values measured

5 Results
by ALICE in PbPb collisions at √ s NN = 2.76 TeV for centralities from peripheral (80-90% of the inelastic cross-section) to central (0-5%) [27]. These ALICE PbP data cover a much wider range of N tracks than is shown in the plot. Although PbPb data are not available at √ s NN = 5.02 TeV for comparison, the evolution of event characteristics from RHIC ( √ s NN = 0.2 TeV, [2,4,28]) to LHC energies [27] suggests that yield ratios should remain similar, while p T values will increase by about 5% when going from √ s NN = 2.76 TeV to 5.02 TeV. For low track multiplicity (N tracks 40), pPb collisions behave very similarly to pp collisions, while at higher multiplicities (N tracks 50) the p T is lower for pPb than in pp. The first observation can be explained since low-multiplicity events are peripheral pPb collisions in which only a few proton-nucleon collisions are present. Events with more particles are indicative of collisions in which the projectile proton strikes the thick disk of the lead nucleus. Interestingly, the pPb curves (Fig. 9, left panel) can be reasonably approximated by taking the pp values and multiplying their N tracks coordinate by a factor of 1.8, for all particle types. In other words, a pPb collision with a given N tracks is similar to a pp collision with 0.55 × N tracks for produced charged particles in the |η| < 2.4 range. Both the highest-multiplicity pp and pPb interactions yield higher p T than seen in central PbPb collisions. While in the PbPb case even the most central collisions possibly contain a mix of soft (lower-p T ) and hard (higher-p T ) nucleon-nucleon interactions, for pp or pPb collisions the most violent interaction or sequence of interactions are selected.
The transverse momentum spectra could also be successfully fitted with a functional form proportional to p T exp(−m T /T ), where T is called the inverse slope parameter, motivated by the success of Boltzmann-type distributions in nucleus-nucleus collisions [29]. In the case of pions, the fitted range was restricted to m T > 0.4 GeV/c in order to exclude the region where resonance decays would significantly contribute to the measured spectra. The inverse slope parameter as a function of hadron mass is shown in Fig. 10, for a selection of event classes, both for pPb data and for MC event generators (AMPT, EPOS LHC, and HIJING). While the data pPb DS π ± K ± p, − p CMS Figure 11: Rapidity densities dN/dy (left) and average transverse momenta p T (right) as a function of center-of-mass energy for pp [5] and pPb collisions, for charge-averaged pions, kaons, and protons. Error bars indicate the uncorrelated combined uncertainties, while boxes show the uncorrelated systematic uncertainties. The curves show parabolic (for dN/dy) or linear (for p T ) interpolation on a log-log scale. The pp and pPb data are for laboratory rapidity |y| < 1, which is the same as the center-of-mass rapidity only for the pp data.
display a linear dependence on mass with a slope that increases with particle multiplicity, the models predict a flat or slowly rising behavior versus mass and only limited changes with track multiplicity. This is to be compared with pp results [5], where both data and the PYTHIA 6.420 event generator [30] show features very similar to those in pPb data. A similar trend is also observed in nucleus-nucleus collisions [2,4], which is attributed to the effect of radial flow velocity boost [1].
Rapidity densities dN/dy and average transverse momenta p T of charge-averaged pions, kaons, and protons as a function of center-of-mass energy are shown in Fig. 11 for pp and pPb collisions, both corrected to the DS selection. To allow comparison at the pPb energy, a parabolic (linear) interpolation of the pp collision values at √ s = 0.9, 2.76, and 7 TeV is shown for dN/dy ( p T ). The rapidity densities are generally about three times greater than in pp interactions at the same energy, while the average transverse momentum increases by about 20%, 10%, and 30% for pions, kaons, and protons, respectively. The factor of three difference in the yields for pPb as compared to pp can be compared with the estimated number of projectile collisions N coll /2 = 3.5 ± 0.3 or with the number of nucleons participating in the collision N part /2 = 4.0 ± 0.3, based on preliminary cross-section measurements, that have proven to be good scaling variables in proton-nucleus collisions at lower energies [31].

Conclusions
Measurements of identified charged hadron spectra produced in pPb collisions at √ s NN = 5.02 TeV have been presented, based on data collected in events with simultaneous hadronic activity at pseudorapidities −5 < η < −3 and 3 < η < 5. Charged pions, kaons, and protons were identified from the energy deposited in the silicon tracker and other track information. In

6 Conclusions
the present analysis, the yield and spectra of identified hadrons for laboratory rapidity |y| < 1 have been studied as a function of the event charged particle multiplicity in the range |η| < 2.4. The p T spectra are well described by fits with the Tsallis-Pareto parametrization. The ratios of the yields of oppositely charged particles are close to one, as expected at mid-rapidity for collisions of this energy. The average p T is found to increase with particle mass and the event multiplicity.
The results can be used to further constrain models of hadron production and contribute to the understanding of basic non-perturbative dynamics in hadron collisions. The EPOS LHC event generator reproduces several features of the measured distributions, a significant improvement from the previous version, attributed to a new viscous hydrodynamic treatment of the produced particles. Other studied generators (AMPT, HIJING) predict steeper p T distributions and much smaller p T than found in data, as well as substantial deviations in the p/π ratios.
Combined with similar results from pp collisions, the track multiplicity dependence of the average transverse momentum and particle ratios indicate that particle production at LHC energies is strongly correlated with event particle multiplicity in both pp and pPb interactions. For low track multiplicity, pPb collisions appear similar to pp collisions. At high multiplicities, the average p T of particles from pPb collisions with a charged particle multiplicity of N tracks (in |η| < 2.4) is similar to that for pp collisions with 0.55 × N tracks . Both the highest-multiplicity pp and pPb interactions yield higher p T than seen in central PbPb collisions.

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: