Transverse momentum and pseudorapidity distributions of charged hadrons in pp collisions at sqrt(s) = 0.9 and 2.36 TeV

Measurements of inclusive charged-hadron transverse-momentum and pseudorapidity distributions are presented for proton-proton collisions at sqrt(s) = 0.9 and 2.36 TeV. The data were collected with the CMS detector during the LHC commissioning in December 2009. For non-single-diffractive interactions, the average charged-hadron transverse momentum is measured to be 0.46 +/- 0.01 (stat.) +/- 0.01 (syst.) GeV/c at 0.9 TeV and 0.50 +/- 0.01 (stat.) +/- 0.01 (syst.) GeV/c at 2.36 TeV, for pseudorapidities between -2.4 and +2.4. At these energies, the measured pseudorapidity densities in the central region, dN(charged)/d(eta) for |eta|<0.5, are 3.48 +/- 0.02 (stat.) +/- 0.13 (syst.) and 4.47 +/- 0.04 (stat.) +/- 0.16 (syst.), respectively. The results at 0.9 TeV are in agreement with previous measurements and confirm the expectation of near equal hadron production in p-pbar and pp collisions. The results at 2.36 TeV represent the highest-energy measurements at a particle collider to date.


Introduction
nal from either of the two BPTX detectors indicating the presence of at least one proton bunch crossing the IP. From these samples, collision events were selected offline by requiring BPTX signals from both beams passing the IP and at least one reconstructed charged particle trajectory in the pixel detector originating from within 0.2 cm of the beam position in the transverse direction (Section 4.1). The total number of collision events and the numbers of collision events passing each requirement are listed in Table 1. Table 1: Numbers of events per data sample used in this analysis. The offline event selection criteria are applied in sequence, i.e., each line includes the selection of the lines above. To select NSD events, a coincidence of at least one HF calorimeter tower with more than 3 GeV total energy on each of the positive and negative sides of the HF was required. Events containing beam-halo muons crossing the detector were identified by requiring the time difference between any two hits from the BSC stations on opposite sides of the IP to be within 73 ± 20 ns. Such events were removed from the data sample. Beam-induced background events producing an anomalously large number of pixel hits were excluded by rejecting events with pixel clusters (Section 4.2) inconsistent with a pp collision vertex. This rejection algorithm was only applied for events with more than 150 pixel clusters, providing a clean separation between collision events and beam background events. Finally, events were required to contain at least one reconstructed primary vertex, as described in Section 4.
To study beam-induced background, the event selection criteria were also applied to a data sample obtained by selecting events with only a single unpaired bunch crossing the IP. The contamination of background events in the colliding-bunch data sample was estimated by taking into account the total unpaired and paired bunch intensities and was found to be negligible (<0.1%). The total number of cosmic-ray muons in the selected data sample was estimated to be less than one event, and was also neglected.
The event selection criteria are expected to have high efficiency for the NSD part of the pp cross section, while rejecting a large fraction of the SD component of pp interactions. The efficiency of the event selection for the different processes and centre-of-mass energies was determined using simulated events obtained from the PYTHIA [11] (version 6.420, tune D6T, [12]) and PHO-JET [13,14] (version 1.12-35) event generators processed with a MC simulation of the CMS detector response (hereafter simply called PYTHIA and PHOJET). In the case of PHOJET, the discussion and numerical values concerning the DD process given in this paper contain both the DD and the Double-Pomeron-Exchange (DPE) processes. The relative event fractions of SD, DD and ND processes and event selection efficiencies at √ s = 0.9 and 2.36 TeV are listed in Table 2 for these two samples.
The measurements were corrected for the selection efficiency of NSD processes and for the fraction of SD events contained in the data sample after the event selection. Based on the PYTHIA (PHOJET) event generator, the fractions of SD events contained in the selected data samples were estimated to be 5.2% (4.9%) at 0.9 TeV and 6.3% (5.0%) at 2.36 TeV. Table 2: Expected fractions of SD, DD, ND and NSD processes ("Frac.") obtained from the PYTHIA and PHOJET event generators before any selection and the corresponding selection efficiencies ("Sel. Eff.") determined from the MC simulation. The generated charged-hadron multiplicity distribution is shown in Fig. 1a in the range |η| < 2.5 for all inelastic events after event selection. The event selection efficiency for NSD events is shown in Fig. 1b as a function of generated charged-hadron multiplicity in the region |η| < 2.5. The correction for the event selection efficiency was applied as a function of number of reconstructed charged particles per event, as illustrated at generator level in Figs. 1a and 1b.
The sum of the corrections to the dN ch /dη measurements due to the NSD event selection efficiency and the SD event contamination typically amounts to 8%. The corrections applied in the analysis are based on PYTHIA using the default SD and DD process fractions as listed in Table 2.
The PYTHIA predictions for the SD and DD fractions differ from those of PHOJET, and are not fully consistent with existing measurements, as explained in Section 5. These differences propagate to a systematic uncertainty of 2% in the dN ch /dη measurement. To estimate the additional systematic uncertainty on the event selection efficiency correction resulting from the possible inaccuracies in the detector simulation, the analysis was repeated after replacing the HF eventselection criterion with a two-sided hit coincidence of signals in the BSC detectors. Based on this comparison, an additional 1% systematic uncertainty was assigned to the dN ch /dη measurements.

Reconstruction algorithms
The analysis presented in this paper measures the dN ch /dη and dN ch /dp T distributions of primary charged hadrons. The dN ch /dη distributions were obtained using three methods based on counting of (i) reconstructed clusters in the pixel barrel detector; (ii) pixel tracklets composed of pairs of clusters in different pixel barrel layers; and (iii) tracks reconstructed in the full tracker volume, combining the pixel and strip hits. The cluster counting method provides an independent measurement for each pixel barrel layer, and the tracklet method for each pair of layers. The third method also allows a measurement of the dN ch /dp T distribution. All three methods rely on the reconstruction of the primary vertex (PV) described in Section 4.1.
The pixel-cluster-counting method has the advantage of having the largest p T acceptance down to small transverse momentum (30 MeV/c), is insensitive to geometrical misalignment of the detector and does not require detailed knowledge of the primary vertex position resolution. A potential disadvantage is the sensitivity to backgrounds from collisions with residual gas in 36 TeV (open circles and dashed histogram) after the event selection is applied to the reconstructed events, using inelastic events from the PYTHIA (histograms) and PHOJET (symbols) event generators. (b) The event selection efficiency expected for NSD events from the PYTHIA (histograms) and PHOJET (symbols) event generators as a function of generated charged hadron multiplicity in the region |η| < 2.5.
the beam pipe (beam-gas collisions), from secondary particles produced in the detector material and from low-p T particles curling in the axial magnetic field (loopers). The pixel-tracklet method is capable of measuring and correcting for the combinatorial background and has a p T threshold of 50 MeV/c. The third method uses the tracker (pixel and SST) to build tracks. It requires at least two pixel hits in different layers, has the largest p T threshold (≈100 MeV/c) and algorithmic complexity, but is the most robust against background hits produced by particles not originating from the collision. The charged-particle multiplicity was corrected in all three methods for the small contamination (< 1%) of primary charged leptons. The measured dN ch /dη values were evaluated by extrapolating or correcting to p T = 0 for all the three analysis methods.
The three reconstruction methods are described in Sections 4.2-4.4.

Primary vertex reconstruction
The x, y and z positions of the luminous region where protons of both beams interact, hereafter referred to as beam spot, are obtained for each data set from three-dimensional vertex fits based on tracks reconstructed with p T > 0.9 GeV/c, using the full event sample. The RMS of the beam spot in the transverse directions was found to be less than 0.05 cm. The beam spot position and dimensions were found to be stable within a given data set.
To reconstruct the z coordinate of the PV for each event, tracks consisting of triplets of pixel hits were formed. The minimum transverse momentum of these tracks is ≈ 75 MeV/c. The tracks were required to originate from the vicinity of the beam spot with a transverse impact parameter (d T ) smaller than 0.2 cm. Of these tracks, only those with d T < 4 σ T , where σ T is the quadratic sum of the uncertainty in d T and the RMS of the beam spot in the transverse The vertex-reconstruction algorithm uses the z coordinate of the tracks at the point of closest approach to the beam axis and the corresponding estimated measurement uncertainty (σ z ). It performs an agglomerative clustering by adding tracks to form groups. These groups (denoted the i th and j th group) are then merged based on their normalized distance, where σ i and σ j are the uncertainties of the z i and z j positions, with a fast nearest-neighbour search algorithm [15]. The z position and its uncertainty σ z for the newly joined group are calculated using a weighted average. The clustering process stops when the smallest normalized distance between the remaining groups gets larger than 8.0, where the stopping condition was optimized using simulated events. Only vertices formed from at least two tracks were considered except when only one track was reconstructed in the event. In the latter case the PV position was defined as the point of the closest approach of the track to the beam axis. The fraction of single-track vertices in the selected data sample is 1.7% (1.3%) at 0.9 TeV (2.36 TeV). The overall PV reconstruction efficiency, evaluated from the data after all other event selection cuts, is in excess of 99% and the fraction of events with more than one primary vertex candidate is 5.0% (7.4%) at 0.9 TeV (2.36 TeV). In the rare case of multiple PV candidates, the vertex composed of the largest set of tracks was chosen.
The reconstructed primary vertex resolution in the z direction is a function of the associated track multiplicity (N) and was found to be parameterized adequately as 0.087 cm /N 0.6 using simulated events.
The distribution of the reconstructed z position of the PV is shown in Fig. 2a. Overlaid is the PV distribution in simulated events, the position and RMS of which were adjusted to reproduce the beam spot measured in data.

Pixel cluster counting method
The pseudorapidity distribution of primary charged hadrons produced in a pp collision can be measured by counting the number of clusters they create when traversing each of the three pixel barrel layers and applying appropriate corrections, as described in this section.
The energy deposited by charged particles traversing a pixel detector layer is spread over multiple pixels. Adjacent pixels with a charge measurement above a readout threshold of typically 2740 electrons are combined into pixel clusters to integrate the total charge deposit [6]. The cluster size and charge depend on the incident angle of the particle with respect to the active detector surface. The cluster length along the z axis ranges from 1-2 pixels at normal incident angle up to 14-16 pixels at shallow crossing angles. Figure 2b shows the measured distribution of cluster charge multiplied by | sin θ| (or 1/cosh η) after the cluster selection discussed below, compared to the simulation. Here, θ is the polar angle of the straight line connecting the PV to the cluster.
The peak position is consistent with the expected charge of 22 ke, while the width of the distribution is slightly larger in the data than in the simulation due to gain calibration nonuniformities.
The cluster counting method correlates the observed pixel-cluster length in the z direction, expressed in number of pixels, with the expected path length traveled by a primary particle at a given η value. For primary particles the cluster length in z is proportional to | cot θ| (or | sinh η|) as displayed in Fig. 3a. Small clusters at large |η| are due to loopers, secondary particles and daughters of long-lived hadrons. Clusters from these background particles were efficiently removed by the cluster-length cut represented by the solid line in Fig. 3a. To allow for an efficient background rejection, only the barrel part of the pixel detector was used, where the detector units are parallel to the beam axis, as opposed to the pixel end-caps. Furthermore, the η range for the cluster counting was restricted to |η| < 2 to avoid acceptance problems due to the slightly off-centred position of the luminous region.
The event selection efficiency and the SD contribution for a given total multiplicity of selected clusters (M) for each pixel barrel layer can be determined from Monte Carlo simulation. The overall change of the dN ch /dη value due to this correction is 9% for both collision energies.
The fraction of clusters created by loopers above the cluster-length cut (1% and 5% for η = 2 and η = 0.5, respectively) can be estimated by measuring the total number of clusters below the cut in data, corrected by the ratio of looper clusters below and above the cut in simulated events. The number of clusters eliminated by the cluster-length cut was found to be higher in data than in simulated events by 10-20% due to a slightly larger abundance of secondary particles and loopers, while the observed number and length distributions of clusters above the cut was found to agree with the simulation.
The corrections for absorption in the beam pipe and detector material, secondary particles, daughters of long-lived hadrons, delta-ray electrons and double hits caused by geometrically overlapping detector units were all evaluated, in bins of η and M, with simulated data. The size of these corrections is 10%, 23% and 41% for the first, second and third detector layer, respectively. Varying the charged-particle multiplicity in the event generator by 50% only causes a ±3% relative change in these corrections. Their dependence on η and pixel-cluster multiplicity is similarly small.

Pixel-tracklet method
This method was first used to measure charged-hadron multiplicities by the PHOBOS experiment at RHIC [16]. Pixel tracklets are constructed from combinations of two pixel hits in any two pixel barrel layers. The difference in the angular positions of the two clusters with respect to the PV, ∆η and ∆φ, are calculated for each tracklet. If two tracklets share a hit, the tracklet with the larger ∆η is discarded. The ∆η distribution of reconstructed tracklets is shown in Fig. 3b, together with the corresponding distribution from simulated data and a separate distribution for simulated primary particles only. Tracklets from primary particles display a sharp peak at ∆η = 0, while the tracklets from the combinatorial background have an extended tail. The simulated ∆η distributions are in good agreement with data.
To suppress the combinatorial background, only tracklets with |∆η| < 0.1 and |∆φ| < 1.0 were selected. Since the combinatorial background is flat in ∆φ, the remaining fraction of background tracklets in the signal region |∆φ| < 1.0 can be estimated from tracklets with 1.0 < |∆φ| < 2.0. This data-driven estimate of the background accurately describes the raw ∆η distribution of tracklets for |∆η| > 2, where no signal from primary particles is expected from the MC simulation. Typical values of this estimated background fraction in the signal region increase with |η| from 2% to 30%. The η range for the tracklet method was restricted to |η| < 2 to avoid a large acceptance correction.
The contribution from secondary particles, reconstruction efficiency and geometrical acceptance needs to be accounted for to determine the number of primary charged hadrons. These correction factors were calculated using PYTHIA simulations for background-subtracted tracklets in bins of z position of the PV, pseudorapidity, and tracklet multiplicity. The magnitude of the correction varies with |η| from 0 to 20%. The correction factors were also cross-checked by PHOJET simulations and only cause a 2-3% change in the dN ch /dη result.
The correction for the event selection efficiency and the SD contribution was determined for each tracklet multiplicity bin. The overall change in the dN ch /dη value due to this correction is about 8% at η = 0.

Tracking method
Pixel and SST detectors were used to reconstruct tracks, including both barrel and end-cap layers. The acceptance was limited to |η| < 2.4 to avoid edge effects. The iterative reconstruction procedure described below follows Refs. [17,18], but was further optimized for primary-track reconstruction in minimum bias events.
In the first step of track reconstruction, tracks with three pixel hits (triplets) are built using the x and y positions of the beam spot and the z coordinate of the primary vertex as constraints. These clean pixel tracks are used as seeds for the Kalman-filter-based trajectory-building algorithm in the SST. The resulting trajectories are stored. Before the second tracking step, the pixel and strip hits associated with the tracks found in the first step are removed from further consideration. The second step uses pixel triplet seeds as well, but does not require a vertex constraint and has a looser transverse impact parameter requirement than in the first step. After removal of hits associated with tracks found in the second step, the third tracking step finds primary tracks seeded by two hits in the pixel detector. At least three hits were required for a track to be accepted.
Tracks found during the three iterative steps were collected and a second iteration of the PV reconstruction, as described in Section 4.1, was performed to refine primary vertex position determination. Finally, the tracks were refit with the corresponding vertex constraint, thus improving their η and p T resolution.
In this analysis, a reconstructed track was considered as a primary-track candidate if it is compatible with originating from the PV (d T < min(4σ T , 0.2 cm) and d z < 4σ z , where d z is the distance between the point of the closest approach of the track to the beam axis and the PV along the z direction).
Studies with simulated events showed that the combined geometrical acceptance and reconstruction efficiency for the tracking method exceeds 50% around p T ≈ 0.1, 0.2 and 0.3 GeV/c for pions, kaons and protons, respectively. The efficiency is about 96% in the |η| < 1 region for p T > 0.25 GeV/c, and is above 80% for pions at p T = 0.15 GeV/c. By requiring the geometrical shapes of the pixel clusters to be consistent with the crossing angle and direction of the track, the fraction of fake tracks was kept below 1%. The fraction of duplicated tracks (e.g., from loopers) was estimated to be about 0.1% in the central region, rising to 0.5% at large |η|.
The measured yield in data was corrected, based on MC simulation and comparisons with data, for geometrical acceptance (2% correction for p T > 200 MeV/c), efficiency of the reconstruction algorithm (5-10% for p T > 300 MeV/c), fake and duplicate tracks (<1% each). The contamination of less than 2% from decay products of long-lived hadrons, photon conversions and inelastic hadronic interactions with the detector material was also subtracted. To obtain the dN ch /dη result from the p T spectrum, an extrapolation to p T = 0 was necessary, resulting in an increase of 5% in the estimated number of charged hadrons.
Corrections based on the average hit efficiency of pixel layers, size of the beam spot, longitudinal and transverse impact-parameter resolutions of pixel tracks were validated with data. As an example, the average number of pixel and strip hits found on tracks in the range |η| < 1 is shown in Figs. 4a and 4b, together with the expectation from PYTHIA. Somewhat fewer particles are predicted with p T < 500 MeV/c than seen in the data, which results in the small difference in the number of tracks with few SST hits in Fig. 4b. This small difference, which originates from limitations of the PYTHIA generator, does not affect the final measurement. The correction for the event selection efficiency and the SD contribution was determined for each track multiplicity bin, and has an overall magnitude of 8.3%.

Systematic uncertainties
Various corrections and their event-selection and model dependence contribute to the systematic uncertainties of the measured quantities. A summary of these systematic uncertainties averaged over η and p T is given in Table 3 and discussed below.
The uncertainties related to the trigger bias and to the event selection are common to all the analysis methods. The efficiency of the trigger and event-selection was corrected for by the prediction of the PYTHIA D6T event generator combined with the full Geant4 simulation of the CMS detector. The material description relevant for this analysis was verified by studies of photon conversion probabilities in the data, found to be in agreement with those obtained from the simulation.
Because single-and double-diffractive pp collisions have much smaller charged-hadron multiplicities per event than non-diffractive events, they contribute to the uncertainty in the measured dN ch /dη mostly through the uncertainty in the fraction of SD and DD events passing the event selection criteria. The fractions of SD events for √ s = 0.9 TeV in PYTHIA and PHO-JET (Table 2) are 23% and 19%, respectively. The UA5 experiment measured 15.5% for this fraction [19]. Based on the simulated trigger efficiencies for the different event types, only 5.5% of events passing the analysis event selection are expected to be single-diffractive events. From the aforementioned variations of SD fractions, an uncertainty of ±1% is attributed to this correction. The contribution of the uncertainty of the fraction of DD events was estimated sim-ilarly to be ±1%. Since underestimated DD and SD fractions both lead to an underestimated dN ch /dη result, a conservative linear sum of 2% was assigned to the above systematic uncertainty. The trigger efficiency of the BSC is more than 98% for events with a valid vertex, and even a 5% uncertainty in single-particle detection efficiency of its individual segments results in a negligible uncertainty in the final result. The trigger efficiency of the BSC and the event selection efficiency of the HF detector were both measured from data and found to be consistent within 1% with the MC simulation. The total systematic uncertainty from propagating all event selection and trigger related uncertainties is 3%. The measurement of the average transverse momentum is less sensitive to the trigger selection efficiency. A smaller, 1% uncertainty was therefore assigned to that result.
The geometrical acceptance was studied by comparing the hit occupancy of the pixel barrel with the predictions from the simulation. The efficiency of the pixel hit reconstruction was estimated using tracks propagated from the SST to the pixel detector and by extrapolating pixel tracklets to the unused pixel barrel layer. The measured pixel hit efficiency was found to exceed 99% with a 0.5% uncertainty from both methods, which propagates into 0.5% uncertainty in the pixel-counting-based, 1% in the tracklet-based, and 0.3% in the track-based results. If the collected charge in one or more pixels in a cluster remains below the threshold, the cluster may be split. The splitting rate was estimated from the geometrical distance distributions of closeby pixel clusters found in the data and in the Monte Carlo simulation and found to be 0.5-0.9% in the simulation and 1.0-1.5% in data.
The uncertainty related to the cluster and tracklet selections was estimated by varying the selection cuts. An additional 3% and 2% uncertainty was assigned to the tracklet and track reconstruction algorithm efficiencies, respectively. Corrections for loopers and secondary particles are simulation dependent; the tracklet-and pixel-counting-based methods have low rejection power compared to the tracking method, thus carry a larger systematic uncertainty (as shown in Table 3).
The effects of the geometrical misalignment of the pixel barrel detector were simulated and a 1% uncertainty was assigned to the results from the tracklet-based method. Hits from beaminduced backgrounds coinciding with the collision were estimated to be very rare, and a conservative 1% random hit contribution was propagated to obtain the uncertainty of the results. The corrections for multiple track counting and fake track rate were estimated from the Monte Carlo simulation and found to be less than 1%. The uncertainty of the extrapolation to the full p T -range depends on the low-p T reach of the three methods and varies between 0.2 and 0.5%. While the sources of uncertainties are largely independent from each other, they are correlated among all the data points.

Charged hadron transverse-momentum distributions
Tracks with |η| < 2.4 and p T > 0.1 GeV/c were used for the measurement of dN ch /dp T . The measured average charged-hadron yields per NSD event are shown in Fig. 5a, as a function of p T in bins of |η|. The yields were fit by the Tsallis function (Eq. 1), which empirically describes both the low-p T exponential and the high-p T power-law behaviours [20,21]: where y = 0.5 ln[(E + p z )/(E − p z )] is the rapidity; C(n, T, m) is a normalization constant that depends on n, T and m; E T = m 2 + p 2 T − m and m is the charged pion mass. This function Table 3: Summary of systematic uncertainties. While the various sources of uncertainties are largely independent, most of the uncertainties are correlated between data points and between the analysis methods. The event selection and acceptance uncertainty is common to the three methods and affects them in the same way. The values in parentheses apply to the p T measurement. provides both the inverse slope parameter T, characteristic for low p T , and the exponent n, which parameterizes the high-p T power-law tail. These fit parameters change by less than 5% with η, thus a fit to the whole region |η| < 2.4 was performed. The p T spectrum of charged hadrons, 1/(2π p T )d 2 N ch /dηdp T , in the region |η| < 2.4, was also fit with the empirical function (Eq. 1) and is shown in Fig. 5b. The p T resolution of the CMS tracker was found to have a negligible effect on the measured spectral shape and was therefore ignored in the fit function. For the 0.9 TeV data, the inverse slope parameter and the exponent were found to be T = 0.13 ± 0.01 GeV and n = 7.7 ± 0.2. For the 2.36 TeV data, the values were T = 0.14 ± 0.01 GeV and n = 6.7 ± 0.2. The average transverse momentum, calculated from the measured data points adding the low-and high-p T extrapolations from the fit is p T = 0.46 ± 0.01 (stat.) ± 0.01 (syst.) GeV/c for the 0.9 TeV and 0.50 ± 0.01 (stat.) ± 0.01 (syst.) GeV/c for the 2.36 TeV data.

Source
The dN ch /dη spectrum was obtained by summing the measured differential yields for 0.1 < p T < 3.5 GeV/c and adding the result to the integral of the fit function for p T < 0.1 GeV/c and p T > 3.5 GeV/c. The latter term amounts to 5% of the total.

Charged hadron pseudorapidity density
The summary of results on the pseudorapidity density distribution of charged hadrons is shown in Fig. 6. The dN ch /dη results for the three layers in the cluster-counting method and the three layer-pairs in the pixel-tracklet method are consistent within 3%. These results from the various layers and from the different layer pairs were combined to provide one set of data from each analysis method, as shown in Fig. 6a. shown. The results from the three analysis methods are in agreement. The larger fraction of background hits in the data compared to simulation affects the cluster-counting method differently from the other two, which results in a small difference at high η, well accounted for by the systematic uncertainty of the measurement.

Discussion
The average transverse-momentum and pseudorapidity densities of charged hadrons derived from the measured data can be compared to results from earlier experiments as a function of the collision energy. The average transverse momentum of charged hadrons was obtained from the fits (Eq. 1) to the transverse-momentum spectrum (Fig. 5b). At low energies the energy dependence of p T can be described by a quadratic function of ln s. The p T from this measurement, shown in Fig. 7a, follows the general trend. At 0.9 TeV it is similar to the results from pp collisions at the same energy [22].
The dN ch /dη distribution was calculated as the weighted average of the data from the three reconstruction methods, taking into account their systematic uncertainties, excluding the common ones, as listed in Table 3. The averaged result is shown in Fig. 6b and compared to measurements at the same accelerator (ALICE, pp [23]) and to previous measurements at the same energy but with different colliding particles (UA5, pp [24]). The shaded error band on the CMS data indicates systematic uncertainties, while the error bars on the data from UA5 and ALICE display statistical uncertainties only. No significant difference is observed between the dN ch /dη distributions measured in pp and pp collisions at √ s = 0.9 TeV.

Summary
Inclusive measurements of charged-hadron densities, dN ch /dp T and dN ch /dη, have been presented based on the first pp collisions recorded at √ s = 0.9 and 2.36 TeV by the CMS experiment during LHC commissioning in December 2009. The numerical values of the data presented in this paper can be found in Ref. [30]. For NSD interactions, the average chargedhadron transverse momentum has been measured to be 0.46 ± 0.01 (stat.) ± 0.01 (syst.) GeV/c at 0.9 TeV and 0.50 ± 0.01 (stat.) ± 0.01 (syst.) GeV/c at 2.36 TeV. The three reconstruction methods employed for the dN ch /dη measurement have yielded consistent results, demonstrating the excellent performance and detailed understanding of the CMS tracker. The pseudorapidity density in the central region, dN ch /dη| |η|<0.5 , has been measured to be 3.48 ± 0.02 (stat.) ± 0.13 (syst.) at 0.9 TeV and 4.47 ± 0.04 (stat.) ± 0.16 (syst.) at 2.36 TeV. The results at 0.9 TeV have been found to be in agreement with previous measurements in pp and pp collisions. With the new measurements at 2.36 TeV, which show a steeper-than-expected increase of charged-hadron multiplicity density with collision energy, the study of particle production in pp collisions has been extended into a new energy regime.