PHYSICAL REVIEW C 91, 064905 (2015) Centrality dependence of particle production in p-Pb collisions at √ sNN = 5.02 TeV J. Adam et al.∗ (ALICE Collaboration) (Received 15 January 2015; published 8 June 2015) We report measurements of the primary charged-particle pseudorapidity density and transverse momentum √ distributions in p-Pb collisions at sNN = 5.02 TeV and investigate their correlation with experimental observables sensitive to the centrality of the collision. Centrality classes are defined by using different event-activity estimators, i.e., charged-particle multiplicities measured in three different pseudorapidity regions as well as the energy measured at beam rapidity (zero degree). The procedures to determine the centrality, quantified by the number of participants (Npart ) or the number of nucleon-nucleon binary collisions (Ncoll ) are described. We show that, in contrast to Pb-Pb collisions, in p-Pb collisions large multiplicity fluctuations together with the small range of participants available generate a dynamical bias in centrality classes based on particle multiplicity. We propose to use the zero-degree energy, which we expect not to introduce a dynamical bias, as an alternative event-centrality estimator. Based on zero-degree energy-centrality classes, the Npart dependence of particle production is studied. Under the assumption that the multiplicity measured in the Pb-going rapidity region scales with the number of Pb participants, an approximate independence of the multiplicity per participating nucleon measured at mid-rapidity of the number of participating nucleons is observed. Furthermore, at high-pT the p-Pb spectra are found to be consistent with the pp spectra scaled by Ncoll for all centrality classes. Our results represent valuable input for the study of the event-activity dependence of hard probes in p-Pb collisions and, hence, help to establish baselines for the interpretation of the Pb-Pb data. DOI: 10.1103/PhysRevC.91.064905 PACS number(s): 25.75.Dw, 24.10.Nz, 25.75.Ag I. INTRODUCTION Proton-lead collisions are an essential component of the heavy ion program at the Large Hadron Collider (LHC) [1]. Measurements of benchmark processes in p-Pb collisions serve as an important baseline for the understanding and the interpretation of the nucleus-nucleus data. These measurements allow one to disentangle hot-nuclear-matter effects which are characteristic of the formation of the quark-gluon plasma (QGP) from cold-nuclear-matter effects. The latter are the effects due to the presence of the nuclei themselves and not the QGP; for example, kT broadening, nuclear modification of parton densities, and partonic energy loss in cold nuclear matter. Of particular interest are studies of nuclear effects on parton scatterings at large momentum transfer (hard processes). To this end, we measure the nuclear modification factor, which is defined as the ratio of particle or jet transverse-momentum (pT ) spectra in minimum-bias (MB) p-Pb to those in pp collisions scaled by the average number of binary p-nucleon (p-N) collisions Ncoll [2]. The latter is given by the ratio of p-N and p-Pb inelastic cross sections times the mass number A. In the absence of nuclear effects, the nuclear modification factor is expected to be unity. In heavy-ion collisions, binary scaling is found to hold in measurements of prompt photons [3] and electroweak probes [4,5], which do not ∗ Full author list given at the end of the article. Published by the American Physical Society under the terms of the Creative Commons Attribution 3.0 License. Further distribution of this work must maintain attribution to the author(s) and the published article’s title, journal citation, and DOI. 0556-2813/2015/91(6)/064905(28) 064905-1 strongly interact with the medium. The observation of binary scaling in p-Pb demonstrates that the strong suppression of hadrons [6], jets [7], and heavy flavor hadrons [8,9] seen in Pb-Pb collisions is due to strong final-state effects. Centralitydependent measurements of the nuclear modification factor RpPb (pT ,cent), defined as pPb dNcent /dpT , cent Ncoll dN pp /dpT RpPb (pT ,cent) = (1) cent require the determination of the average Ncoll for each centrality class. Moreover, it has been recognized that the study of pPb collisions is also interesting in its own right. Several measurements [10–13] of particle production in the lowand intermediate-transverse-momentum region clearly show that p-Pb collisions cannot be explained by an incoherent superposition of pp collisions. Instead, the data are compatible with the presence of coherent [14] and collective [15] effects. Their strength increases with multiplicity, indicating a strong collision-geometry dependence. In order to corroborate this hypothesis a more detailed characterization of the collision geometry is needed. The Glauber model [16] is generally used to calculate geometrical quantities of nuclear collisions (A-A or p-A). In this model, the impact parameter b controls the average number of participating nucleons (hereafter referred as “participants” or also “wounded nucleons” [17,18]), Npart and the corresponding number of collisions, Ncoll . It is expected that variations of the amount of matter overlapping in the collision region will change the number of produced particles, and parameters such as Npart and Ncoll have traditionally been used to describe those changes quantitatively and to relate them to pp collisions. ©2015 CERN, for the ALICE Collaboration J. ADAM et al. PHYSICAL REVIEW C 91, 064905 (2015) By using the Glauber model one can calculate the probability distributions πν (ν), where ν stands for Npart or Ncoll . Since ν cannot be measured directly it has to be related via a model to an observable M, generally called centrality estimator, via the conditional probability P(M|ν) to observe M for a given ν. For each collision system and center-of-mass energy, the model has to be experimentally validated by comparing the measured probability distribution Pmeas (M) to the one calculated from the convolution Pcalc (M) = ν P(M|ν)πν (ν). Once the model has been validated, for each event class defined by an M-interval, the average ν is calculated. In order to unambiguously determine ν, one chooses observables whose mean values depend monotonically on ν. Note that, in p-A collisions, the impact parameter is only loosely correlated to ν. Hence, although one uses traditionally the term centrality to refer to these measurements, the relevant parameters are Npart and Ncoll . The procedure described above can be easily extended to several estimators. Of particular interest are estimators from kinematic regions that are causally disconnected after the collision. The measurement of a finite correlation between them unambiguously establishes their connection to the common collision geometry. Typically these studies are performed with observables from well-separated pseudorapidity (η) intervals, e.g., at zero degree (spectators, slow nucleons, deuteron breakup probability) and multiplicity in the rapidity plateau. The use of centrality estimators in p-A collisions based on multiplicity or summed energy in certain pseudorapidity intervals is motivated by the observation that they show a linear dependence on Npart or Ncoll . This is also in agreement with models for the centrality dependence of particle production (e.g., the wounded nucleon model [17,18]), or also string models like FRITIOF [19]). The total rapidity integrated multiplicity of charged particles measured in hadron-nucleus h-A ) at center-of-mass energies ranging from collisions (Nch 10 to 200 GeV (E178 [20], PHOBOS [21]) is consistent pp h-A = Nch Npart /2. The with a linear dependence on Npart : Nch ratio of particle pseudorapidity (η) densities in d-Au and pp collisions exhibits a dependence on η, which implies that the scaling behavior has a strong rapidity dependence with an approximate Npart scaling at η = 0 and an approximate scaling target with the number of target participants (Npart = Npart − 1) in the Au-going direction [21]. In d-Au collisions at the BNL √ Relativistic Heavy Ion Collider (RHIC; sNN = 200 GeV), the PHENIX and STAR collaborations [22,23] have used the multiplicity measured in an η interval of width 0.9 centered at η ≈ −3.5 (Au-going direction) as a centrality estimator. The multiplicity distribution has been successfully described target by the Glauber model by assuming Npart scaling. Finally, √ in centrality-averaged p-Pb collisions at the LHC ( sNN = 5.02 TeV) the primary charged-particle pseudorapidity density at η = 0 scaled by the mean number of participants is found to be consistent with the corresponding value in pp collisions √ interpolated to the same sNN [24]. At RHIC, the deuteron dissociation probability can be accurately modelled by a Glauber calculation and measured by using the zero-degree calorimeters in the d direction [22,25]. The mean number of participants has been determined for centrality classes obtained with the multiplicity estimator described above and used to calculate the deuteron breakup probability. Inferred and measured probabilities are consistent, demonstrating the correlation between collision geometry and multiplicity and providing a stringent test for the Npart determination. Since, for example, hard scatterings can significantly contribute to the overall particle multiplicity, correlations between high-pT particle production and bulk multiplicity can also be induced after the collisions and, hence, they are not only related to the collision geometry. Therefore, the use of Ncoll from the Glauber model to scale cross sections of hard processes from pp to p-A has to undergo the same scrutiny as the correlation of the centrality estimator to the collision geometry. This is necessary also due to the enhanced role of multiplicity fluctuations in p-A. While the average of centrality estimators vary monotonically with ν, for a full description of the conditional probability P(M|ν) fluctuations of M for a fixed ν have to be taken into account. In Pb-Pb collisions, these multiplicity fluctuations have little influence on the centrality determination. The range of ν is large and P(M|ν) converges with increasing ν rapidly to a Gaussian with small width relative to the the range of ν. However, in p-Pb collisions, the range of multiplicities used to select a centrality class is of similar magnitude as the fluctuations, with the consequence that a centrality selection based on multiplicity may select a biased sample of nucleon-nucleon collisions (for a discussion of this effect in d + Au; see Ref. [22]). In essence, by selecting high (low) multiplicity one chooses not only large (small) average Npart , but also positive (negative) multiplicity fluctuations leading to deviations from the binary scaling of hard processes. These fluctuations are partly related to qualitatively different types of collisions. High-multiplicity nucleon-nucleon collisions show a significantly higher particle mean transverse momentum. They can be understood as “harder” collisions, i.e., with higher four-momentum transfer squared Q2 or as nucleon-nucleon collisions where multiple parton-parton interactions (MPIs) take place. In contrast, a centrality selection that is not expected to induce a bias on the binary scaling of hard processes is provided by the energy measurement with the zero-degree calorimeters (ZDCs) due to their large η separation from the central barrel detectors. They detect the so-called “slow” nucleons produced in the interaction by nuclear deexcitation processes, or knocked out by wounded nucleons [26,27]. The relationship of the energy deposited in the ZDC to the number of collisions requires a detailed model to describe the slow nucleon production. A heuristic approach, based on a parametrization of data from low-energy experiments, is discussed in the present paper. We show that centrality estimators using forward neutron energy and those using central multiplicity give consistent results for Npart and Ncoll , demonstrating their connection to the collision geometry. Based on the considerations outlined above we study two different procedures for centrality estimation. The first procedure is to determine the centrality with chargedparticle multiplicity. The collision geometry is determined by fitting the measured multiplicity distribution with the 064905-2 CENTRALITY DEPENDENCE OF PARTICLE PRODUCTION . . . PHYSICAL REVIEW C 91, 064905 (2015) Ncoll distribution obtained from the Glauber model [16], convolved with a negative binomial distribution (NBD). Due to the possible dynamical bias introduced by the multiplicity selection, Ncoll should in this case not be used to scale hard cross sections. Additional effort is needed to understand the bias or to extend the Glauber model to include additional dynamical fluctuations. Several possible directions have been discussed, for example, Glauber–Gribov fluctuations of the proton size [28] as well as fluctuations of the number of hard scatterings per collision due to the impact-parameter dependence and purely statistical (Poissonian) fluctuations [29]. The second procedure requires a centrality selection with minimal bias and, therefore, uses the ZDC signal. To relate the ZDC signal to the collision geometry we developed a heuristic model for slow-nucleon emission based on a parametrization of data from low-energy experiments. This heuristic approach, however, can provide only a modeldependent Ncoll determination. However, one can study the correlation of two or more observables out of which at least one is expected to scale linearly with Ncoll . Examples are (i) the target-going multiplicity proportional to the numtarget ber of wounded target nucleons (Npart = Npart − 1 = Ncoll ), (ii) the multiplicity at midrapidity proportional to the number of participants (Npart = Ncoll + 1), (iii) the yield of hard probes, like high-pT particles at midrapidity proportional to Ncoll . These scalings can be used as an ansatz when calculating Ncoll based on an event selection using the ZDC. Both alternatives are discussed in the present paper. The paper is organized as follows: Section II describes the experimental conditions, the event selection, and the event characterization using the multiplicity distributions of charged particles measured in various η ranges, or the energy collected in the ZDC. Section III describes the centrality determination based on charged-particle distributions using an NBD-Glauber fit to extract the average geometrical quantities for typical centrality classes. Section IV presents a phenomenological model describing the relation of the energy deposited in the ZDC calorimeter and Ncoll . Section V discusses the various effects leading to a bias in the centrality measurements based on particle multiplicity. Section VI introduces a hybrid method, where we use the ZDC to characterize the event activity, and base the determination of Ncoll on the assumption that Npart -scaling holds for the central pseudorapidity multiplicity target density or Npart -scaling for particle production in the target region. Section VII discusses the implications of the different choices of a centrality estimator on the physics results, such as the nuclear modification factors, or the pseudorapidity density of charged particles at midrapidity. Section VIII summarizes and concludes the paper. beams in the two rings, implying that the ratio of beam energies is fixed to be exactly equal to the ratio of the charge/mass ratios of each beam. Protons at 4 TeV energy collided onto fully stripped 208 82 Pb ions √ at 1.58 TeV per nucleon energy resulting in collisions at sNN = 5.02 TeV in the nucleon–nucleon center-of-mass system (cms), which moves with a rapidity of yNN = 0.465 in the direction of the proton beam. In the following, we use the convention that y stands for ycms , defined such that the proton moves towards positive ηcms , while η stands for ηlab . The number of colliding bunches varied from 8 to 288. The proton and Pb bunch intensities were ranging from 0.2 × 1012 to 6.5 × 1012 and from 0.1 × 1012 to 4.4 × 1012 , respectively. The luminosity at the ALICE interaction point was up to 5 × 1027 cm−2 s−1 resulting in a 10 kHz hadronic interaction rate. The rms width of the interaction region is 6.3 cm along the beam direction and of about 60 μm in the direction transverse to the beam. The ALICE apparatus and its performance in the LHC Run 1 are described in Refs. [30,31], respectively. The main detector components used for the centrality determination are the Silicon Pixel Detector (SPD), two cylindrical layers of hybrid silicon pixel assemblies covering |η| < 2.0 for the inner layer and |η| < 1.4 for the outer layer for vertices at the nominal interaction point, with 93.5% active channels; the Time Projection Chamber (TPC), a large cylindrical drift detector covering |η| < 0.9; the VZERO scintillator counters, covering the full azimuth within 2.8 < η < 5.1 (VZERO-A) and −3.7 < η < −1.7 (VZERO-C); and the Zero-Degree Calorimeters (ZDC), two sets of neutron (ZNA and ZNC) and proton (ZPA and ZPC) calorimeters positioned at ±112.5 m from the interaction point, with an energy resolution of about 20% for the neutron and 24% for the proton calorimeters. The p-Pb trigger, configured to have high efficiency for hadronic events, requires a signal in both the VZERO-A and VZERO-C (VZERO-AND requirement). Beam-gas and other machine-induced background collisions with deposited energy above the thresholds in the VZERO or ZDC detectors are suppressed by requiring the signal arrival time to be compatible with a nominal p-Pb interaction. The fraction of remaining beam-related background after all requirements is estimated from control triggers on noncolliding or empty bunches and is found to be negligible. The resulting event sample corresponds to a so-called visible cross section of 2.09 ± 0.07 barn measured in a van der Meer scan [32]. From Monte Carlo simulations we expect that the sample consists mainly of non-single-diffractive (NSD) collisions and a negligible contribution of single-diffractive (SD) and electromagnetic interactions. The VZERO-AND trigger is not fully efficient for NSD events. Previous Monte Carlo studies (for details see Ref. [24]) have shown that the inefficiency is observed mostly for events without a reconstructed vertex, i.e., with no particles produced at central rapidities. Given the fraction of such events in the data (1.5%), the corresponding inefficiency was found to be 2.2% with a large systematic uncertainty of 3.1%. Correcting for this inefficiency would mainly concern the most peripheral class (80% to 100%) where the correction amounts up to 11% ± 15.5%. For the results reported in this paper, centrality II. EXPERIMENTAL CONDITIONS The data were recorded during a dedicated LHC run of four weeks in January and February, 2013. Data were taken with two beam configurations by inverting the direction of the two particle species, referred to as p-Pb and Pb-p, respectively, for the situations where the proton beam is moving towards positive rapidities, or vice versa. The two-in-one-magnet design of the LHC imposes the same magnetic rigidity of the 064905-3 J. ADAM et al. PHYSICAL REVIEW C 91, 064905 (2015) 0.56 0.82 0.64 11.0 23.6 8.74 0.44 1.08 0.76 with the Glauber-derived p-nucleus inelastic cross section of 2.1 ± 0.1 b. The Glauber MC determines on an event-by-event basis the properties of the collision geometry, such as Npart , Ncoll , and TpPb , which must be mapped to an experimental observable. Assuming that the average V0A multiplicity is proportional to the number of participants in an individual p-A collision, the probability distribution P (n) of the contributions n to the amplitude from each p-nucleon collisions can be described by the NBD, which is defined as P (n; μ,k) = (μ/k)n (n + k) , (n + 1) (k) (μ/k + 1)n+k (3) where is the gamma function, μ the mean amplitude per participant and the dispersion√parameter k is related to the relative width given by σ/μ = 1/μ + 1/k. From the closure of the NBD under convolution, it follows that the conditional probability P(n|Npart ), i.e., Npart repeated convolutions, is equal to P (n; Npart μ,Npart k). To obtain the NBD parameters μ and k, the calculated V0A distribution, obtained by convolving the Glauber Npart distribution with P(n|Npart ), is fit to the measured V0A distribution. The fit is performed by excluding the low-V0Aamplitude region, VOA < 10. We note, however, that fitting with the full range gives consistent results. The measured V0A distribution together with the NBD-Glauber distribution for the best fit are shown in Fig. 1. Similar fits have been performed ALICE p-Pb sNN = 5.02 TeV Data NBD-Glauber fit 10-2 10-2 Npart x NBD (μ = 11.0, k = 0.44) 10-3 0 10 60-80% with ρ0 being the nucleon density, which provides the overall normalization, a radius of R = 6.62 ± 0.06 fm, and a skin depth of a = 0.546 ± 0.010 fm based on data from lowenergy electron-nucleus scattering experiments [39]. Nuclear collisions are modelled by randomly displacing the projectile proton and the target Pb nucleus in the transverse plane. A hard-sphere exclusion distance of 0.4 fm between nucleons is employed. The proton is assumed to collide with the nucleons of the Pb nucleus if the transverse distance between them is less than the distance corresponding√to the inelastic nucleon-nucleon cross section of 70 ± 5 mb at s = 5.02 TeV, estimated from interpolating data at different center-of-mass energies [40] including measurements at 2.76 and 7 TeV [41]. The VZERO-AND cross section measured in a van der Meer scan [32] was found to be compatible, assuming negligible efficiency and electromagnetic contamination corrections, 9.6 25.2 9.8 80-100% (2) k 20 30 40 10-4 10-5 0 100 200 0-5% 1 , 1 + exp r−R a μ 5-10% ρ (r) = ρ0 k 10-20% To determine the relationship between charged-particle multiplicity and the collision properties, such as the number of participating nucleons Npart , binary pN collisions Ncoll , or inel nuclear overlap TpPb (=Ncoll /σNN ), it is customary to use the Glauber Monte Carlo (Glauber MC) model combined with a simple model for particle production [33–37]. The method was used in Pb-Pb collisions and is described in detail in Ref. [38]. In the Glauber calculation, the nuclear density for 208 82 Pb is modelled by a Woods–Saxon distribution for a spherical nucleus V0A V0M CL1 p-Pb μ 20-40% A. Negative binomial distribution Glauber fit pp 40-60% III. CENTRALITY FROM CHARGED-PARTICLE DISTRIBUTIONS System distribution 60-80% (i) CL1: the number of clusters in the outer layer of the silicon pixel detector, |η| < 1.4; (ii) V0A: the amplitude measured by the VZERO hodoscopes on the A side (the Pb-going side in the p-Pb event sample), 2.8 < η < 5.1; (iii) V0C: the amplitude measured by the VZERO hodoscopes on the C side (the p-going side in the p-Pb event sample), −3.7 < η < −1.7; (iv) V0M: the sum of the amplitudes in the VZERO hodoscopes on the A and C side (V0A + V0C); (v) ZNA: the energy deposited in the neutron calorimeter on the A side (the Pb-going side in the p-Pb event sample). TABLE I. Fit parameters of the Npart × NBD for pp collisions at 7 TeV and p-Pb multiplicity distributions. Events (arb. units) classes have been defined as percentiles of the visible cross section and the measurements are not corrected for trigger inefficiency. The centrality determination is performed by exploiting the rapidity coverage of the various detectors. The raw multiplicity distributions measured in the Central Barrel are modelled by assuming particle production sources are distributed according to a NBD. The zero-degree energy of the slow nucleons emitted in the nucleon fragmentation requires more detailed models. In this context, the main estimators used for centrality in the following are 300 400 500 V0A (Pb-side) amplitude (arb. units) FIG. 1. (Color online) Distribution of the sum of amplitudes in the V0A hodoscopes (Pb-going), as well as the NBD-Glauber fit (explained in the text). Centrality classes are indicated by vertical lines. The inset shows a zoom in on the most peripheral events. 064905-4 CENTRALITY DEPENDENCE OF PARTICLE PRODUCTION . . . PHYSICAL REVIEW C 91, 064905 (2015) TABLE II. Geometric properties (b, TpPb , Npart , Ncoll ) of p-Pb collisions for centrality classes defined by cuts in V0A. The mean values and the σ values are obtained with a Glauber Monte Carlo calculation, coupled to a NBD to fit the V0A distribution. Centrality (%) 0–5 5–10 10–20 20–40 40–60 60–80 80–100 0–100 b (fm) σ (fm) TpPb (mb−1 ) σ (mb−1 ) Npart σ Ncoll σ 3.12 3.50 3.85 4.54 5.57 6.63 7.51 5.56 1.39 1.48 1.57 1.69 1.69 1.45 1.11 2.07 0.211 0.186 0.167 0.134 0.0918 0.0544 0.0277 0.0983 0.0548 0.0539 0.0549 0.0561 0.0516 0.0385 0.0203 0.0728 15.7 14.0 12.7 10.4 7.42 4.81 2.94 7.87 3.84 3.78 3.85 3.93 3.61 2.69 1.42 5.10 14.7 13.0 11.7 9.36 6.42 3.81 1.94 6.87 3.84 3.78 3.85 3.93 3.61 2.69 1.42 5.10 to V0M and CL1 and the corresponding fit parameters are listed in Table I. The values of the parameters μ and k are similar to those obtained by fitting the corresponding multiplicity distributions in pp collisions at 7 TeV. Since the raw distribution is sensitive to experimental parameters such as noise and gain, one cannot expect identical values even in the case of perfect Npart scaling and therefore the comparison is only qualitative. For a given centrality class, defined by selections in the measured distribution, the information from the Glauber MC in the corresponding generated distribution is used to calculate the mean number of participants Npart , the mean number of collisions Ncoll , and the average nuclear overlap function TpPb . These are given in Table II, with the corresponding σ values. Since the event selection dominantly selects NSD events, it is important to note that the number of participants in the Glauber calculation would increase by only 2.5% for NSD events. This was estimated with a modified Glauber calculation to exclude SD collisions [24]. The systematic uncertainties are evaluated by varying the Glauber parameters (radius, skin depth, and hard-sphere exclusion distance) within their known uncertainty. The uncertainties on Ncoll are listed in Table III by adding all the deviations from the central result in quadrature. The uncertainties range from about 4%–5% in peripheral collisions to about 10% in central collisions. Note that, inel inel as TpPb = Ncoll /σNN , the uncertainties on σNN and Ncoll largely cancel in the calculation of TpPb . However, edge effects in the nuclear overlap are large for TpPb in peripheral collisions. The procedure was tested with a MC-closure test using HIJING p-Pb simulations [29] with nuclear modifications of the parton density (shadowing) and elastic scattering switched off. In the MC-closure test, the V0A distribution obtained from a detailed detector simulation coupled to HIJING was taken as the input for the fit with the NBD-Glauber method. The difference between the Ncoll values calculated from the fit and those from the MC truth used in the HIJING simulation range from 3% in central to 23% in peripheral events (see Table III). The large uncertainty in the peripheral events arises from the small absolute values of Ncoll itself. In this case a small absolute uncertainty results in a large relative deviation. The total uncertainty on Ncoll for each centrality class with the CL1, V0M, or V0A estimators is obtained by adding the uncertainty from the variation of the Glauber parameters with those from the respective MC-closure test in quadrature. The NBD-Glauber fit is repeated for the multiplicity distribution of the SPD clusters (CL1) and for the sum of V0A and V0C, V0M, in the same centrality classes as for V0A. The Ncoll values as a function of centrality are given in Table III and shown in Fig. 2 for the various estimators. In addition, the events from the MC-Glauber calculation were ordered according to their impact parameter, and the values of Ncoll were extracted for the same centrality classes. The variation of Ncoll between different centrality estimators is small and of similar magnitude as the systematic uncertainty TABLE III. Comparison of Ncoll values. In the first column results are listed for centrality classes obtained by ordering the events according to the impact parameter distribution (b). In the next three columns Ncoll values are given for the various centrality estimators CL1, V0A, V0M. The systematic uncertainty on Ncoll (in parentheses on TpPb ) is obtained by changing all Glauber parameters by 1σ ; the second column is obtained from the MC-closure test; those two are added in quadrature to obtain the total systematic uncertainty on Ncoll . The last column gives the Ncoll values obtained for the ZNA (see Sec. IV) and the uncertainty on the slow nucleon model (SNM, see Sec. IV). Centrality (%) b Ncoll CL1 Ncoll V0M Ncoll V0A Ncoll Sys. Glauber Sys. MC closure Sys. Total ZNA Ncoll Sys. SNM 0–5 5–10 10–20 20–40 40–60 60–80 80–100 0–100 14.4 13.8 12.7 10.2 6.30 3.10 1.44 6.88 15.6 13.6 12.0 9.49 6.18 3.40 1.76 6.83 15.7 13.7 12.1 9.55 6.26 3.40 1.72 6.87 14.8 13.0 11.7 9.36 6.42 3.81 1.94 6.87 10% (3.7%) 10% (3.5%) 10% (3.2%) 8.8% (3.1%) 6.6% (4.3%) 4.3% (6.7%) 2.0% (9.3%) 8% (3.4%) 3% 1% 2% 2% 3% 20% 23% 10% 10% 10% 9% 7.2% 20% 23% 8% 15.7 13.9 12.4 9.99 6.53 3.04 1.24 6.88 7% 5% 2% 2% 4% 4% 8% 064905-5 16 PHYSICAL REVIEW C 91, 064905 (2015) ALICE p-Pb sNN = 5.02 TeV TABLE IV. Fit parameters of the V0A distributions using standard Glauber and Glauber–Gribov ( = 0.55) distributions of Npart and Ncoll coupled to a NBD. CL1 V0M V0A ZNA b (imp.par.) 14 12 10 8 6 4 Sys: Glauber + MC-closure b 〈 Ncoll 〉 / 〈 Ncoll 〉 2 1.4 1.3 1.2 1.1 1 0.9 0.8 0.7 0 10 20 30 40 50 60 70 80 90 Centrality (%) FIG. 2. (Color online) Values of Ncoll extracted from CL1, V0M, V0A, ZNA and by ordering the events according to the impact parameter distribution (b). The systematic uncertainty, given by the quadrature sum of the uncertainty from the Glauber parameters and the MC-closure test, are drawn around the values obtained with b. obtained by adding in quadrature the uncertainty from the Glauber model and from the MC-closure test. This implies that the Ncoll determination with the NBD-Glauber fit is robust and independent of the centrality estimator used. B. Glauber–Gribov corrections Events (arb. units) Event-by-event fluctuations in the configuration of the incoming proton can change its scattering cross section [28]. In the Glauber MC this phenomenon is implemented by an effective scattering cross section [42–44]. At high energies, the configuration of the proton is taken to be frozen over the timescale of the p-A collision. Analogously to the studies in Refs. [45,46], the effect of these frozen fluctuations of the projectile proton is evaluated with a modified version of the Glauber MC, referred to as “Glauber–Gribov.” This version includes event-by-event variations of the nucleon-nucleon 10-1 Glauber-MC p-Pb sNN = 5.02 TeV Std-Glauber Glauber-Gribov Ω = 0.55 Glauber-Gribov Ω = 1.01 10-2 10-3 Distribution μ k Std-Glauber and Ncoll ⊗ NBD Std-Glauber and Npart ⊗ NBD Glauber-Gribov and Ncoll ⊗ NBD Glauber-Gribov and Npart ⊗ NBD 12.2 11.0 12.6 11.0 0.58 0.44 1.35 0.60 cross section. Here we have used the same values of the parameter , which controls the width of the probability inel distribution of σNN , as used in Ref. [45]; namely, = 0.55 and 1.01, where = 0.0 corresponds to the standard Glauber. The distribution of the number of participants, Npart , obtained from the two Glauber–Gribov parameter variations are shown in the left panel of Fig. 3 together with a standard Npart distribution obtained using a fixed inelastic cross inel = 70 mb. The Glauber–Gribov Npart distributions section, σNN are much broader than the Glauber distribution due to the cross-section fluctuations. We note that by construction the total inelastic p-Pb cross section is unaltered by the proton fluctuations. The Glauber–Gribov distributions of Npart and Ncoll , coupled to a NBD, were fit to the measured distribution of V0A. The right panel of Fig. 3 shows the V0A distribution together with various fits performed with the standard Glauber or the Glauber–Gribov distribution, using = 0.55, and assuming that the signal increases proportionally either to Npart or to Ncoll . As before, no attempt is made to describe the most peripheral region (below ∼90%), where trigger efficiency is not 100%. The extracted parameters are given in Table IV. The standard NBD-Glauber fits yield satisfactory results using either the Npart or the Ncoll scaling, which result in a similar average number of collisions Ncoll , evaluated for each of the centrality intervals as shown in Table V. The Glauber– Gribov fits with = 0.55 provide an equally good description of the measured V0A distribution as the standard Glauber, indicating that the fits cannot discriminate between the models. Events (arb. units) 〈 Ncoll 〉 J. ADAM et al. ALICE p-Pb sNN = 5.02 TeV 10-3 10-4 10-4 10-2 10-2 0 10 20 30 40 Data Std-Glauber (Ω=0) N coll x NBD Std-Glauber (Ω=0) Npart x NBD 10-5 10-5 Glauber-Gribov (Ω=0.55) N coll x NBD Glauber-Gribov (Ω=0.55) Npart x NBD 0 10 20 30 40 50 60 Npart 0 100 200 300 400 500 V0A (Pb-side) amplitude (arb. units) FIG. 3. (Color online) (left) Glauber and Glauber–Gribov Monte Carlo Npart distributions for 5.02 TeV p-Pb collisions. (right) Measured V0A distribution compared to Glauber and Glauber–Gribov fits assuming Npart or Ncoll scaling. The inset shows a zoom in on the most peripheral events. 064905-6 CENTRALITY DEPENDENCE OF PARTICLE PRODUCTION . . . PHYSICAL REVIEW C 91, 064905 (2015) TABLE V. Ncoll values obtained for various fits of the V0A, using Std-Glauber ( = 0.0) and Glauber–Gribov ( = 0.55) distributions for Npart or Ncoll , coupled to a NBD. is expected to be monotonically related to Ncoll [26] and can therefore be used as a centrality estimator. Emitted nucleons are classified as “black” or “gray.” This terminology originates from emulsion experiments where it was related to the track grain density. Black particles, typically defined to have velocity β 0.25 in the nucleus rest frame, are produced by nuclear evaporation processes, while gray particles, 0.25 β 0.7, are mainly nucleons knocked out from the nucleus. Experimental results at lower energies show that the features of the emitted nucleons, such as angular-momentum and multiplicity distributions, are weakly dependent on the projectile energy in a wide range from 1 GeV up to 1 TeV (see Ref. [26] and references therein). These observations suggest that the emission of slow particles is mainly dictated by nuclear geometry. To quantitatively relate the energy deposited in the ZDC to the number of binary collisions requires a model to describe the production of slow nucleons. Since there are no models available that are able to describe the slow nucleon emission at LHC energies, we relied on the weak dependence on collision energy and followed a heuristic approach. For this purpose we developed a model for the slow-nucleon emission (SNM) based on the parametrization of experimental results at lower energies. In the left panel of Fig. 4 it is shown that the energy detected by the neutron calorimeter on the Pb-remnant side (ZNA) is correlated with the energy detected in the proton ZDC (ZPA), up to the onset of a saturation in the emission of neutrons. This saturation effect is commonly attributed to the black component (see Ref. [26] and references therein). The energy detected by ZP is lower. This is due both to the lower number of protons in the Pb nucleus and to the lower acceptance for emitted protons that are affected by LHC magnetic fields. Furthermore, contrary to ZN, ZP response and energy resolution strongly depend on the proton impact point. In the following we focus on the ZN spectrum for these reasons. The energy released in the ZNA is anticorrelated with the signal in the neutron calorimeter placed on the p-remnant side (ZNC) (see Fig. 4, right). The p-remnant-side ZN signal Centrality (%) 0–5 5–10 10–20 20–40 40–60 60–80 80–100 0–100 Std-Glauber Glauber-Gribov Npart × NBD Ncoll × NBD Npart × NBD Ncoll × NBD 14.8 13.0 11.7 9.36 6.42 3.81 1.94 6.87 15.3 13.4 12.0 9.62 6.40 3.42 1.85 6.87 17.8 14.4 12.0 8.82 5.68 3.33 1.72 6.73 19.2 15.2 12.5 9.04 5.56 2.89 1.43 6.75 The broader Npart distributions in the Glauber–Gribov models require smaller intrinsic fluctuations in the NBD at fixed Npart . No satisfactory fit is obtained with = 1.01. As expected, the corresponding values of Ncoll , also shown in Table V, are larger (smaller) for central (peripheral) than those obtained from the standard Glauber, as a consequence of the different shapes of the Npart distributions in these models [see Fig. 3 (left)]. Both assumptions that the multiplicity distribution is proportional to Npart or Ncoll are found to give an equally good description of the experimental data (see Fig. 3, and parameters reported in Table IV). The difference in the extracted geometric quantities is within 10% for 0%–60% and slightly increases for the most peripheral, which is of similar order as the uncertainty derived from the Glauber parameters (see the last two columns of Table V). IV. CENTRALITY FROM ZERO-DEGREE ENERGY 100 0.006 80 0.005 0.004 60 ZNC (arb. units) ZNA energy (TeV) The energy measured in the zero-degree calorimeters (ZDCs) can be used to determine the centrality of the collision. The ZDC detects the so-called “slow” nucleons produced in the interaction: protons in the proton ZDC (ZP) and neutrons in the neutron ZDC (ZN). The multiplicity of slow nucleons 300 ALICE p-Pb sNN = 5.02 TeV 250 200 0.003 40 0.002 150 20 0.001 ALICE p-Pb 0 0 2 4 6 8 sNN = 5.02 TeV 10 12 14 16 18 20 22 24 100 0 0 ZPA energy (TeV) 20 40 60 80 100 ZNA (TeV) FIG. 4. (Color online) (left) Correlation between Pb remnant neutron (ZNA) and proton (ZPA) calorimeter energies. (right) Average signal on the p remnant side (ZNC) versus average energy detected by ZNA in centrality bins selected using ZNA. 064905-7 J. ADAM et al. PHYSICAL REVIEW C 91, 064905 (2015) 30000 ALICE p-Pb sNN = 5.02 TeV25000 20000 Data 15000 105 SNM-Glauber 10000 5000 0 1 2 3 4 5 6 7 8 104 102 20 40 60 80 0-5 % 103 20-40 % 10-20 % 5-10 % where NLCF is the number of light charged fragments; namely, the number of fragments with Z < 8. Since we cannot directly measure the number of light charged fragments in ALICE, we assumed that NLCF is proportional to the number of slow protons as measured by COSY [48]: NLCF = γ Nslow p where the proportionality factor γ = 1.71 is obtained through a minimization procedure. The first term in Eq. (5) describes the gray neutron production that linearly increases with Ncoll and hence with NLCF . The second term reproduces the saturation in the number of black nucleons and is based on a parametrization of results from the COSY experiment where the neutron yield is related to NLCF [48]. The values of the parameters α, a, b, and c are obtained through a minimization procedure and are α = 0.48, a = 50, b = 230, c = 4.2. The relative fraction of black and gray neutrons is evaluated by assuming that 90% of the emitted neutrons are black, as measured in proton-induced spallation reactions in the energy range between 0.1 and 10 GeV [49]. The number of nucleons emitted from 208 82 Pb is finally calculated event by event as a function of Ncoll , assuming binomial distributions with probabilities p = Nslow p /82 for protons and p = Nslow n /126 for neutrons. The kinematical distributions of the black and the gray components are described by independent statistical emission from a moving frame: black nucleons are emitted from a stationary source, while gray nucleons are emitted from a frame slowly moving along the beam direction with βgray = 0.05. 40-60 % This relationship was found to be in good agreement with gray proton data measured by E910 in p-Au collisions with an 18 GeV/c proton beam [47]. The coefficient values taken from the E910 fit are rescaled to Pb nuclei by using the ratio (ZPb /ZAu ): c0 = −0.24, c1 = 0.55, c2 = 0.0007. The linear term is the dominant contribution while the quadratic term is negligible. Neglecting in this context a possible saturation effect for black protons, we approximate the average number of black protons using the ratio between “evaporated” and “direct” proton production measured by the COSY experiment in p-Au interactions at 2.5 GeV [48]: Nblack p = 0.65Ngray p . The average number of slow neutrons is obtained using the following formula: b , (5) Nslow n = αNLCF + a − c + NLCF 60-80 % (4) 80-100 % 2 Ngray p = c0 + c1 Ncoll + c2 Ncoll . The angular distribution for gray tracks is forward peaked in the polar angle θ , while black nucleons are assumed to be uniformly distributed, in agreement with the experimental observations [47,50]. The neutron calorimeter has full geometric acceptance for neutrons emitted from the Pb nucleus, as estimated through Monte Carlo simulations. Experimentally, a fraction of triggered events (4.4%) does not produce a signal in ZN, these are very peripheral events with no neutron emission. The convolution of ZN acceptance and efficiency has been calculated coupling an event generator based on the SNM to HIJING [29] and using a full GEANT 3 [51] description of the ALICE experimental apparatus. Taking into account the experimental conditions (beam-crossing angle and detector configuration), we obtain that 94% of the events have a signal in the neutron calorimeter, in good agreement with the experimental acceptance (95.6%). Since the events without ZNA signal have the same CL1, V0A, and V0M distributions as the those in the 80%–100% centrality bin, they are attributed to this bin. The SNM, coupled to the probability distribution for Ncoll calculated from the Glauber MC as in Sec. III A, is fit to the experimental distribution of the ZDC energy in Fig. 5. The detector acceptance and resolution are fixed to the experimental values. The parameters that are obtained by fitting the data are γ , a, b, c, and α. The main features of the measured energy distribution in the neutron calorimeter on the Pb side are reasonably well described by the SNM. The Ncoll , reported in Table III and in Fig. 2, is then Events (arb. units) cannot be easily calibrated in energy units due to the lack of peaks in the spectrum. Events characterized by low-Ncoll values, corresponding to low energy deposit in ZNA, have the largest contribution in ZNC. This implies that the participant contribution cannot be neglected for very peripheral events, where the sample is also partially contaminated by electromagnetic processes. Therefore, supposing that no nucleons are emitted in the limit that there is no collision, the model is not expected to provide a complete and reliable description for very peripheral data. In the following, we briefly summarize the main ingredients of the developed heuristic model for slow-nucleon emission. The average number of emitted gray protons is calculated as a function of Ncoll by using a second-order polynomial function: 100 120 EZN (TeV) FIG. 5. (Color online) Distribution of the neutron energy spectrum measured in the Pb-remnant-side ZN calorimeter. The distribution is compared with the corresponding distribution from the SNM-Glauber model (explained in the text), which is shown as a line. Centrality classes are indicated in the figure. The inset shows a zoom in on the most peripheral events. 064905-8 103 Data NBD-Glauber fit 102 10 1 Ncoll x NBD (μ = 1.72, k = 0.35) 0 20 40 60 80 100 120 140 0-5% 10-2 -3 10 10-4 0 20 10-20% 10-2 -3 10 10-4 0 10 20 30 40 50 60 70 80 90 100 10-2 -3 10 10-40 5 10 15 20 25 30 35 40 45 50 V0A ring1 (Pb-side) amplitude (arb. units) 80 100 -3 10 10-4 0 10-2 -3 10 10-4 0 10 20 30 40 50 60 70 80 90 80-100% 10-1 10-2 -3 10 20 40 60 80 100 V0A ring1 (Pb-side) amplitude (arb. units) 10 1 40-60% -2 10 -3 10 10-40 V0A ring1 (Pb-side) amplitude (arb. units) Events (arb. units) Events (arb. units) 60-80% 60 20-40% V0A ring1 (Pb-side) amplitude (arb. units) 10-1 40 5-10% 10-2 V0A ring1 (Pb-side) amplitude (arb. units) Events (arb. units) Events (arb. units) V0A amplitude (arb. units) ring1 (Pb-side) Events (arb. units) 104 Events (arb. units) ALICE p-Pb sNN = 5.02 TeV 10 PHYSICAL REVIEW C 91, 064905 (2015) 10 20 30 40 50 60 V0A ring1 (Pb-side) amplitude (arb. units) 〈 V0A ring 1 〉 5 Events (arb. units) Events (arb. units) CENTRALITY DEPENDENCE OF PARTICLE PRODUCTION . . . 30 Data Glauber + SNM Unfolding 25 20 15 10 5 10-40 5 10 15 20 25 30 35 40 V0A ring1 (Pb-side) amplitude (arb. units) 0 20 40 60 80 100 Centrality (%) FIG. 6. (Color online) V0A ring1 signal distributions. The top-left panel shows the distribution for MB events together with a NBD-Glauber fit. The remaining panels show the distributions and mean values for centrality classes selected with ZNA. These are compared to those obtained by the convolution of the P(Ncoll |centZNA ) distributions from the SNM with the NBD from the NBD-Glauber fit to V0A ring 1. Data are also compared to the distributions obtained with an unfolding procedure, where the Ncoll distributions have been fit to the data by using the parameters from the NBD-Glauber fit. The bottom-right panel compares the mean values of these distributions as a function of the centrality. calculated for centrality classes defined by dividing the energy spectrum in percentiles of the hadronic cross section. The systematic uncertainty on the Ncoll values reported in Table III, has been evaluated by varying the model parameters within reasonable ranges: (i) using for the relative fraction of black over gray protons Nblack p = 0.43Ngray p from spallation reaction results [49], (ii) including a saturation effect for black protons, (iii) decreasing the ratio of black over gray neutrons to 0.5 as obtained from DPMJET [52], (iv) neglecting the linear term in Eq. (5) and assuming complete saturation for the neutrons, (v) varying γ by ±10%, and (vi) assuming different parametrization for the fluctuations in the number of slow nucleons for a fixed Ncoll value. We note that this uncertainty corresponds to the variation of the SNM parameters; therefore, it is meant as the uncertainty within our SNM and does not reflect any possible other model that could describe nucleus fragmentation. When using the Ncoll values for the ZNA centrality estimator, the total systematic uncertainty on Ncoll is obtained by adding the uncertainties from the Glauber and SNM parameters in quadrature. Within the Glauber-model, the consistency between measurements of Ncoll in largely separated rapidity regions establishes their relation to centrality. To this end, we correlate the ZNA measurements to the amplitudes measured in the innermost ring of the VZERO-A detector (V0A ring 1), since this ring covers the most-forward rapidity in the Pbgoing direction. The Ncoll distributions [P(Ncoll |centZNA )] for centrality classes selected with ZNA (centZNA ) are obtained from the SNM-Glauber fit. These are convolved with the NBD obtained from the NBD-Glauber fit to the MB V0A ring 1 distribution. Figure 6 compares the distributions of V0A ring 1 obtained from these convolutions to those measured in the same ZNA centrality classes. As expected, the distributions in the most-peripheral events, where the SNM does not provide a reliable description of the data, are not well reproduced by the Glauber-MC convolution. In all other classes, the experimental distributions are well reproduced. The deviations are consistent ZNA Pb-side (see Table III) and Ncoll (see with those between Ncoll Table VII) assuming that the target-going charged-particle multiplicity measured in V0A ring 1 is proportional to the number of wounded target nucleons. In addition, Fig. 6 shows the results of an unfolding procedure. For each V0A ring 1 distribution selected by a ZNA centrality class, we find the Ncoll distribution that, convolved with the NBDMB , fits the data, i.e., the parameters of the fit are the relative contributions of each Ncoll bin. The unfolded distributions (shown in blue) agree well with the data for all centrality bins, apart from a small discrepancy in the 80%–100% distribution at low amplitude, which is affected MB distribution by trigger and event-selection efficiency. The Ncoll which results from the sum of the unfolded distributions of all centrality bins agree well with that from Glauber MC. The existence of Ncoll distributions that, folded with NBD, agree with measured signal distributions is a necessary condition for ZNA to behave as an unbiased centrality selection. In contrast, it is worth noting that a centrality selection based on central multiplicity, as CL1, has no such solution; i.e., no such good agreement can be found when the V0A ring 1 distributions are selected by ordering the events according to CL1. The biases related to centrality selection will be discussed in the next 064905-9 PHYSICAL REVIEW C 91, 064905 (2015) 30 10-2 Glauber-MC p-Pb sNN = 5.02 TeV 10-2 Npart Npart J. ADAM et al. 400 Glauber-MC Pb-Pb sNN = 2.76 TeV 10-3 10-3 300 20 10-4 10 0 0 5 10 15 10-4 200 10-5 100 10-6 0 0 10-5 5 10 15 10-6 b (fm) 10-2 Glauber-MC 600 p-Pb sNN = 5.02 TeV 10-2 Multiplicity Multiplicity b (fm) 20 20000 10-3 Glauber-MC Pb-Pb sNN = 2.76 TeV 10-3 15000 400 10-4 200 0 0 10 20 30 10-4 10000 10-5 5000 10-6 0 0 Npart 10-5 100 200 300 400 10-6 Npart FIG. 7. (Color online) (top) Scatter plot of number of participating nucleons versus impact parameter. (bottom) Scatter plot of multiplicity versus the number of participating nucleons from the Glauber fit for V0A. The quantities are calculated with a Glauber Monte Carlo of p-Pb (left) and Pb-Pb (right) collisions. section. The assumption that the ZNA selection is bias free will be used in Sec. VI as an ansatz for the hybrid method. V. DISCUSSION OF POTENTIAL BIASES ON CENTRALITY A. Multiplicity bias Section III A describes the NBD-Glauber fitting procedure used to determine the collision geometry in terms of Ncoll and Npart for each centrality class. The NBD is used to account for multiplicity fluctuations at fixed Npart . In contrast to Pb-Pb collisions, for p-Pb collisions these multiplicity fluctuations are sizable compared to the width of the Npart distribution, as illustrated in Fig. 7. For large fluctuations, a centrality classification of the events based on multiplicity may select a sample of nucleon-nucleon collisions which is biased compared to a sample defined by cuts on the impact parameter b. This selection bias, which occurs for any system with large relative statistical fluctuations in particle multiplicity per nucleon-nucleon collision can be quantified by using the Glauber fit itself. The left panel of Fig. 8 shows the ratio between the average multiplicity per average participant and the average multiplicity of the NBD as a function of centrality. In Pb-Pb collisions, where the width of the plateau of the Npart distribution is large with respect to multiplicity fluctuations, the ratio deviates from unity only for the most peripheral collisions. As expected, in p-Pb collisions the ratio differs from unity for all centralities with large deviations for the most central and most peripheral collisions; the most central (peripheral) collisions have on average much higher (lower) multiplicity per participant. When selecting event classes using impact parameter b intervals, there is no deviation from unity, as expected. The right panel of Fig. 8 shows, for each centrality estimator, the relative width of the NBD distribution (σ/μ). As expected, the estimators with the largest bias on the multiplicity per participant correspond to those with the largest relative width. It is instructive for the further discussion to consider the clan model [53], which is the standard physical explanation of the NBD distribution in the context of particle production in pp collisions. In this model particle sources, called ancestors, are produced independently according to a Poisson distribution with mean value, N = k ln (1 + μ/k). Each ancestor can produce on average μ/N particles, e.g., by decay and fragmentation, and a clan contains all particles that stem from the same ancestor. Hence, the bias observed above also corresponds to a biased number of clans, which are sources of particle production. Analogously, in all recent Monte Carlo generators a large part of the multiplicity fluctuations is indeed due to the fluctuations of the number of particle sources, i.e., 2QCD ) parton-parton scatterings multiple semihard (Q2 (MPI). 064905-10 2 Glauber-MC p-Pb sNN = 5.02 TeV Pb-Pb sNN = 2.76 TeV 1.8 1.6 1.4 1.2 PHYSICAL REVIEW C 91, 064905 (2015) σ / μ = 1/μ+1/k Multiplicity / N ancestor /μ CENTRALITY DEPENDENCE OF PARTICLE PRODUCTION . . . V0A CL1 V0C V0M b (imp.par) V0M (Pb-Pb) 1 2 Glauber-MC p-Pb sNN = 5.02 TeV Pb-Pb sNN = 2.76 TeV 1.8 1.6 1.4 1.2 1 0.8 0.8 V0A (N 0.6 0.6 V0A (Ncoll X NBD) 0.4 0.4 0.2 0.2 0 0 part x NBD) CL1 V0C V0M V0M (PbPb) 0 10 20 30 40 50 60 70 80 90 100 Centrality [%] FIG. 8. (Color online) (left) Multiplicity fluctuation bias quantified as the mean multiplicity per Npart /μ from the NBD-Glauber MC in p-Pb and Pb-Pb calculations. (right) Relative width of the NBD obtained from the NBD-Glauber fit of various multiplicity distributions in p-Pb and Pb-Pb calculations. As an example, the HIJING generator accounts for fluctuations of the number of MPI per NN interaction via a NN overlap function TNN (bNN ), where bNN is the NN impact parameter, i.e., the impact parameter between the proton and each wounded nucleon of the Pb nucleus. The probability for inelastic NN collisions is given as one minus the probability to have no interaction: (6) where σsoft is the geometrical soft cross section of 57 mb [29] related to the proton size and σhard is the energy-dependent pQCD cross section for 2 → 2 parton scatterings. Furthermore, as in the clan model, there is a Poissonian probability P (nhard ) = nhard nhard −nhard e nhard ! for multiple hard collisions with an average number determined by bNN : nhard = σhard TNN (bNN ) . B. Jet-veto bias (7) Additional kinematic biases exist for events containing high-pT particles. These particles arise from the fragmentation (8) Hence, the biases on the multiplicity discussed above correspond to a bias on the number of hard scatterings (nhard ) and bNN in the event. The latter correlates fluctuations over large rapidity ranges (long-range correlations). As a consequence, for peripheral (central) collisions we expect a lower (higher) than average number of hard scatterings per binary collision, corresponding to a nuclear modification factor less than one (greater than one). In general, the number of binary pN collisions, Ncoll , is used to scale the reference pp yields and obtain the nuclear modification factor, which is used to quantify nuclear matter effects. However, for centrality classes based on multiplicity, owing to the bias induced by such selection, hard processes do not simply scale with Ncoll but rather with an effective number of collisions, obtained by scaling Glauber by the number of hard scatterings per pN the Ncoll Glauber collision: Ncoll nhard pN /nhard pp . As discussed in the HIJING example above, the number of hard scatterings per 〈 nhard 〉 / 〈 Ncoll 〉 2 dσinel = π dbNN [1 − e−(σsoft +σhard )TNN (bNN ) ], pN collision is simulated in Monte Carlo models. In this specific MC, even without bias, the total number of hard scatterings deviates from simple Ncoll scaling due to energy conservation at high Ncoll . Instead, with the objective to study a baseline corresponding to an incoherent and unconstrained superposition of nucleon-nucleon collisions, the PYTHIA [54] event generator has been coupled to the p-Pb MC Glauber calculation. For each MC Glauber event PYTHIA is used to generate Ncoll independent pp collisions. In the following we refer to this model as G-PYTHIA. In this model, the number of hard scatterings per pN collision shows a strong deviation from Ncoll scaling which is illustrated in Fig. 9 and resembles the bias observed in Fig. 8. 3.5 3 2.5 〈 nhard 〉pp 2 1.5 Glauber + Pythia p-Pb sNN = 5.02 TeV 1 0.5 0 20 40 60 80 100 Centrality (%) FIG. 9. Number of hard scatterings [MSTI(31) in PYTHIA6] per Ncoll as a function of the centrality calculated with a toy MC that couples a pp PYTHIA6 calculation to a p-Pb Glauber MC (described in the text). 064905-11 PHYSICAL REVIEW C 91, 064905 (2015) 1 80% 〈bNN〉 (fm) Events (arb. units) J. ADAM et al. 60% 10-1 10-2 1.5 1.45 1.4 HIJING Glauber MC 1.35 p-Pb sNN = 5.02 TeV 1.3 1.25 10-3 1.2 ALICE 10-4 1.1 p-Pb sNN = 5.02 TeV 1.05 pp s = 7 TeV 10-5 1 -1 10 〈bpp〉 1.15 1 10 0 5 10 15 2 10 CL1 FIG. 10. (Color online) Multiplicity distribution used as centrality estimators√in p-Pb collisions, compared to the distribution in pp collisions at s = 7 TeV. The dashed lines mark the 80% and the 60% percentile of the p-Pb cross section, respectively. of partons produced in parton-parton scattering with large momentum transfer. Their contribution to the overall multiplicity rises with parton energy and, thus, can introduce a trivial correlation between the centrality estimator and the presence of a high-pT particle in the event. In particular, for very peripheral collisions, the multiplicity range that governs the centrality for the bulk of soft collisions can represent an effective veto on hard processes, leading to a RpPb < 1. This bias is illustrated in Fig. 10. It shows a multiplicity distribution which is used as centrality estimators in p-Pb collisions, √ compared to the same distribution in pp collisions at s = 7 TeV. The dashed lines mark the 80% and the 60% percentile of the p-Pb cross section. The fraction of the pp cross section selected with the 80%–100% (60%–100%) p-Pb multiplicity cut is 0.8 (0.97). The 80% cut in p-Pb is smaller than the multiplicity range covered in pp, therefore resulting in an effective veto on the large multiplicity events produced by hard processes. C. Geometric bias The bNN dependence of particle production postulated in Sec. V A leads to a purely geometrical centrality estimator independent bias for peripheral p-Pb collisions [55]. As illustrated in Fig. 11, the mean impact parameter between two nucleons (bNN ), calculated from a Monte Carlo Glauber simulation, is almost constant for central collisions, but rises significantly for Npart < 6. This reduces the average number of MPIs for most peripheral events, enhancing the effect of the bias leading to a nuclear modification factor less than (greater than) one for peripheral (central) collisions. In summary, based on simplified models we have identified three different possible biases that are expected to lead to deviations from unity at high pT of the nuclear modification factors in peripheral and central collisions. As will be discussed and studied in the following sections, the effect decreases with increasing rapidity separation between the RpPb measurement and the centrality estimator. 20 Npart FIG. 11. Average nucleon-nucleon impact parameter as a func√ tion of the number of participants for p-Pb at sNN = 5.02 TeV from a Glauber MC calculation as implemented in HIJING (no shadowing, no elastic scattering). The result depends on the modeling of the spatial parton density in the nucleon. In HIJING it is approximated by the Fourier transform of a dipole form factor. For the estimators we used, the main biases are (i) CL1: strong bias due to the full overlap with tracking region. Additional bias from the “jet veto effect,” as jets contribute to the multiplicity and shift events to higher centralities (pT dependent); (ii) V0M: reduced bias since the VZERO hodoscopes are outside the tracking region; (iii) V0A: reduced bias because of the enhanced contribution from the Pb fragmentation region; (iv) ZNA: no bias expected. In addition, independent of the centrality estimator, there is a geometrical bias for peripheral collisions (see Sec. V C). VI. THE HYBRID METHOD A. Basis and assumptions of the method The hybrid method presented in the following section aims to provide an unbiased centrality estimator and relies on two main assumptions. The first is to assume that an event selection based on ZN does not introduce any bias on the bulk at midrapidity and on high-pT particle production. This assumptions is based on the results from the unfolding procedure presented in Sec. IV and the full acceptance of ZN for neutrons emitted from the Pb nucleus, also discussed in Sec. IV. In addition consistent results where obtained with proton calorimeter ZP, in the region of its full acceptance. Therefore, we do not expect a significant bias from the ZN selection and herein this is taken as ansatz. This selection was also used in the method proposed in Sec. IV; however, the Ncoll determination provided by the SNM-Glauber model is model dependent. In contrast, in the hybrid method, the Ncoll determination is based—as an ansatz—on a particular scaling for particle multiplicity (the second assumption), e.g., we assume that the charged-particle multiplicity measured at midrapidity scales with the number of participants. 064905-12 CENTRALITY DEPENDENCE OF PARTICLE PRODUCTION . . . PHYSICAL REVIEW C 91, 064905 (2015) TABLE VI. Average pseudorapidity covered by VZERO detector rings in p-Pb and Pb-p collisions. (10 < pT < 20 GeV/c) particles measured at midrapidity (|η| < 0.3). In order to compare these observables on the same scale and also, at first order, to neglect detector efficiency and acceptance effects, we use so-called normalized signals Si /SMB . These are obtained dividing Si , i.e., the mean value of dNch /dη, number of raw SPD tracklets or raw VZERO signal in a given ZN-centrality class i, by the corresponding mean values in minimum-bias collisions. Figure 12 shows, for bins in ZN centrality, the correlation between a few selected normalized signals and the normalized charged-particle density averaged over −1 < η < 0. The statistical uncertainty is negligible, while the systematic uncertainties largely cancel in the ratio to the MB signals. One can note that the correlation exhibits a clear dependence on the pseudorapidity of the normalized signal. The slope of the normalized signals with dNch /dη diminishes towards the proton direction (C side in p-Pb collisions). For example, in the innermost ring of the VZERO-C detector the signal amplitude range is about a factor three, while for the innermost ring of the VZERO-A detectors it is about twice as large. In the wounded nucleon model [18], the total number of participants Npart is expressed in terms of target and projectile participants. The charged-particle density at midrapidity is thus proportional to Npart , whereas at higher rapidities the model predicts a dependence on a linear combination of the number of target and projectile participants with coefficients which depend on the rapidity. Close to Pb-rapidity a linear target wounded target nucleon scaling (Npart = Npart − 1) is expected. In order to further understand the relative trends of the observables in Fig. 12 and to relate them with geometrical quantities, such as Npart , one can adopt the wounded nucleon model and make the assumption that dNch /dη in −1 < η < 0 is proportional to Npart . In this case, the other observables can be related to Npart , assuming linear or power-law dependence. The linear dependence can be parametrized with Npart − α, where α is a free parameter. Then the normalized signals can Ring VZERO-A ring 1 VZERO-A ring 2 VZERO-A ring 3 VZERO-A ring 4 VZERO-C ring 1 VZERO-C ring 2 VZERO-C ring 3 VZERO-C ring 4 ηcms (p-Pb) ηcms (Pb-p) −5.39 −4.80 −4.28 −3.82 3.34 2.82 2.33 1.86 4.45 3.87 3.35 2.89 −4.26 −3.74 −3.25 −2.78 Normalized signals 2 1.8 1.6 1.4 1.2 1 0.8 0.6 0.4 0.2 V0A ring 1 dN/dη(1.5<η<2.0) dN/dη(-2.0<η<-1.5) V0C ring 1 ALICE p-Pb sNN = 5.02 TeV Data/Fit Data/Fit Normalized signals To obtain more insight into the particle-production mechanisms, we study the correlation of various pairs of observables that, in ZN-centrality classes, are expected to scale linearly with Npart or Ncoll . One of these observables is the chargedparticle density dNch /dη in |η| < 2.0, measured with the SPD. The charged-particle pseudorapidity density is obtained from the measured distribution of tracklets, formed by using the position of the primary vertex and two hits, one on each SPD layer [24]. At larger pseudorapidities, where a direct multiplicity measurement is not available, we study the raw signals of the four rings of the VZERO-A and VZERO-C detectors separately. We exploit both beam configurations, p-Pb and Pb-p, in order to cover the widest possible rapidity range. To take into account the impact of secondary particles, the pseudorapidity coverage of the VZERO detector rings with respect to the primary charged particles was calculated with a full detector simulation based on DPMJET [51,52] and it is given in Table VI in the center-of-mass system (cms), which moves with a rapidity of yNN = 0.465 in the direction of the proton beam (see Sec. II). The information about charged-particle multiplicity, dominated by soft particles, is complemented with observables from hard processes which are expected to scale with the number of binary collisions, such as the yield of high-pT 1.1 1 0.9 0.4 0.6 0.8 1.4 1.6 ( dN/d / dN/d 1 1.2 ) 2 1.8 1.6 1.4 1.2 1 0.8 0.6 0.4 0.2 V0A ring 1 dN/dη(1.5<η<2.0) dN/dη(-2.0<η<-1.5) V0C ring 1 ALICE p-Pb sNN = 5.02 TeV 1.1 1 0.9 0.4 MB -1<η<0 0.6 0.8 1 1.2 1.4 ( dN/d / dN/d 1.6 ) MB -1<η<0 FIG. 12. (Color online) (left) Normalized signal from various observables (the innermost ring of VZERO-A and VZERO-C and two rapidity intervals of dNch /dη) versus the normalized charged-particle density averaged over −1 < η < 0. (left) Fit with the linear function from Eq. (9). (right) Fit with the power-law function from Eq. (10). Only data from p-Pb collisions are shown. 064905-13 PHYSICAL REVIEW C 91, 064905 (2015) α β J. ADAM et al. 2 ALICE p-Pb sNN = 5.02 TeV p-Pb Pb-p ALICE p-Pb sNN = 5.02 TeV p-Pb Pb-p 1.3 target Npart or Ncoll scaling 1 1.4 1.2 Npart scaling 0 1.1 -2 -3 V0 Tracklets dNch/dη dNch/dη(10
5 GeV/c) by scaling the measurements at s = 7 TeV√with the ratio of spectra calculated with NLO pQCD at s = 5.02 and 7 TeV [61]. The systematic uncertainty, given by the largest of the relative systematic uncertainties of the spectrum at 2.76 or 7 TeV at low pT and assigned from the relative difference between the NLO-scaled spectrum for different scales and the difference between the interpolated and the NLO-scaled data at high pT , ranges from 6.8% to 8.2%. For MB collisions the nuclear modification factor RpPb is consistent with unity for pT above 6 GeV/c. The same analysis was repeated by dividing the visible cross section (see Sec. II) in event classes defined by the centrality estimators described above, and the QpPb were calculated by using the values of Ncoll listed in Tables III and VII for each given estimator. Figure 19 shows QpPb for different centrality estimators and different centrality classes. The uncertainties of the p-Pb and pp spectra are added in quadrature, separately, for the statistical and systematic uncertainties. The systematic uncertainty on the spectra is only shown for the V0A 0%–5% centrality bin and is the same for all others, since all the corrections are independent of centrality. The total systematic uncertainty on the normalization, given by the quadratic sum of the uncertainty on the normalization of the pp data and the normalization of the p-Pb data, amounts to 6.0% and is shown as a gray box around unity. The systematic uncertainty on TpPb is shown as a light blue box around unity. For simplicity, we draw only the uncertainty for the minimum-bias value TpPb MB . As expected, for CL1, V0M, and V0A, QpPb strongly deviates from unity at high pT in all centrality classes, with values well above unity for central collisions and below unity for peripheral collisions. However, the spread between 064905-18 QpPb CENTRALITY DEPENDENCE OF PARTICLE PRODUCTION . . . 1.8 ALICE p-Pb 1.6 PHYSICAL REVIEW C 91, 064905 (2015) sNN = 5.02 TeV Charged particles | η |<0.3 1.4 0-5% 40-60% Syst. on 〈 T 5-10% 60-80% Syst. on normalization 80-100% Syst. on dN/dp 10-20% 20-40% pA 〉 T 1.2 1 0.8 1 0.6 0.4 1.15 1.1 1.05 1 0.95 0.9 0.85 0 ZN + N Pb-side coll 5 10 ZN + N mult coll 15 20 25 30 5 p (GeV/c) 10 15 20 25 30 p (GeV/c) T T FIG. 20. (Color online) QpPb spectra with the hybrid method. Spectra are calculated in ZNA classes with Ncoll as given in Table VII, and are obtained with assumptions on particle production described in Sec. VI. centrality classes reduces with increasing rapidity gap between the range used for the centrality estimator and that used for the pT measurement. There is a clear indication of the jet-veto bias in the most peripheral CL1 class, where QpPb has a significant negative slope (pT > 5 GeV/c) since the jet contribution to the total multiplicity increases with pT . This jet-veto bias diminishes for V0M and is absent for V0A, where QpPb < 1 for peripheral collisions, indicating that the multiplicity bias is still present. In order to study the centrality determination biases further, the QpPb spectra are compared to the G-PYTHIA spectra. The event centrality is obtained from the charged-particle multiplicity in the rapidity region covered by each estimator in the same way as in data, and Ncoll is directly obtained from the Monte Carlo. The calculation is shown as lines in Fig. 19. With this approach, the general trend at high pT is reasonably well described for all centrality classes, particularly for CL1. This suggests that particle production at high pT in p-Pb collisions indeed can be approximated by an an incoherent superposition of pp collisions. The agreement, however, is not as good for the V0A and V0M estimators, since the model is not adequate for forward particle production, particularly in the target fragmentation region. G-PYTHIA also reproduces the jet-veto bias, as indicated by the good agreement of the pT dependence in the low- and intermediate-pT region in the most peripheral CL1 collisions. However, for central collisions, the QpPb values show a significant enhancement at intermediate pT ≈ 3 GeV/c (called the Cronin effect; a nuclear modification factor above unity at intermediate pT , observed at lower energies in p-A collisions [25,62–64]), which increases with centrality independently of the estimator used. The enhancement in the intermediate pT region is about 15%, and the differences in the height of the peak among centrality estimators are small with respect to the absolute increase of the p-Pb yields. The enhancement is not reproduced by our model of incoherent superposition of pp collisions. In contrast, in the low-pT region, below the Cronin peak, the yield is overestimated by the model. This overestimate at low pT is expected because this pT region is dominated by soft processes and therefore is not expected to scale with Ncoll . On the other hand, the intermediate-pT region is expected to be dominated by hard scatterings and should scale with Ncoll in the absence of nuclear effects. From this we can conclude that the Cronin enhancement observed is due to nuclear modification effects, as observed in other measurements [10–13], as well as in the minimum-bias RpPb [2]. The bottom right plot of Fig. 19 shows QpPb for the ZNA centrality selection. The classes selected by the ZNA present spectra much more similar to each other than the other estimators, as expected in the absence of a multiplicity bias. The height of the Cronin peak relative to the yield at high pT is larger with the V0A selection, which may be seen as a sign of a remaining small bias in V0A, expected from the G-PYTHIA calculations. However, for peripheral collisions (60%–80% and 80%–100%), the absolute values of the spectra at high pT indicate the presence of a bias on Ncoll in the ZNA measurement. This is not due to the event selection but is due to the inaccurate estimate of Ncoll values for peripheral events, where a small, absolute uncertainty results in a large relative deviation in the QpPb calculation. As discussed in Sec. VI, the hybrid method uses centrality classes selected with ZNA and Ncoll values determined with assumptions on particle production. Figure 20 shows the Pb-side resulting QpPb values, Qmult pPb in the left panel and QpPb in the right panel. Here it is important to note that the ratios in the lower-right panel in Fig. 19 and in both panels in Fig. 20 have the same shape by construction and only differ due to the scaling (Ncoll ) of the reference. The small differences among the Ncoll values (Table VII) are reflected in consistent QpPb , which also remain consistent with unity at high pT for all centrality classes. This confirms the absence 064905-19 J. ADAM et al. PHYSICAL REVIEW C 91, 064905 (2015) 2 1.6 1.4 CL1 V0A V0M V0A Glauber-Gribov 1.8 1.2 1 1.6 1.4 1.2 1 0.8 0.8 0.6 0.6 0.4 0.4 0.2 0.2 0 0 ALICE p-Pb sNN = 5.02 TeV T 1.8 QpPb 10