PHYSICAL REVIEW C 86, 014907 (2012) Measurement of the azimuthal anisotropy for charged particle production in lead-lead collisions with the ATLAS detector √ sN N = 2.76 TeV G. Aad et al.* (ATLAS Collaboration) (Received 14 March 2012; published 24 July 2012) Differential measurements of charged particle azimuthal anisotropy are presented for lead-lead collisions at sNN = 2.76 TeV with the ATLAS detector at the LHC, based on an integrated luminosity of approximately 8 μb−1 . This anisotropy is characterized via a Fourier expansion of the distribution of charged particles in azimuthal angle relative to the reaction plane, with the coefficients vn denoting the magnitude of the anisotropy. Significant v2 –v6 values are obtained as a function of transverse momentum (0.5 < pT < 20 GeV), pseudorapidity (|η| < 2.5), and centrality using an event plane method. The vn values for n ⩾ 3 are found to vary weakly with both η and centrality, and their pT dependencies are found to follow an approximate scaling relation, 1/2 vn1/n (pT ) ∝ v2 (pT ), except in the top 5% most central collisions. A Fourier analysis of the charged particle pair distribution in relative azimuthal angle (φ = φa − φb ) is performed to extract the coefficients vn,n = cos nφ. For pairs of charged particles with a large pseudorapidity gap (|η = ηa − ηb | > 2) and one particle with pT < 3 GeV, the v2,2 –v6,6 values are found to factorize as vn,n (pTa , pTb ) ≈ vn (pTa )vn (pTb ) in central and midcentral events. Such factorization suggests that these values of v2,2 –v6,6 are primarily attributable to the response of the created matter to the fluctuations in the geometry of the initial state. A detailed study shows that the v1,1 (pTa , pTb ) data are consistent with the combined contributions from a rapidity-even v1 and global momentum conservation. A two-component fit is used to extract the v1 contribution. The extracted v1 is observed to cross zero at pT ≈ 1.0 GeV, reaches a maximum at 4–5 GeV with a value comparable to that for v3 , and decreases at higher pT . √ DOI: 10.1103/PhysRevC.86.014907 PACS number(s): 25.75.Dw, 25.75.Ld I. INTRODUCTION where pT is the transverse momentum, η is the pseudorapidity, and vn and n represent the magnitude and direction of the nth-order harmonic, respectively (see Sec. IV A). The nth-order harmonic has n-fold periodicity in azimuth, and the coefficients at low pT are often given descriptive names, such as “directed flow” (v1 ), “elliptic flow” (v2 ), or “triangular flow” (v3 ). In typical noncentral heavy-ion collisions where the nuclear overlap region has an “elliptic” shape (or quadrupole asymmetry) on average, the azimuthal anisotropy is expected to be dominated by the v2 component [5–7]. However, it was recently pointed out that the positions of the nucleons in the overlap region can fluctuate to create matter distributions with additional shape components, such as dipole (n = 1) and sextupole (n = 3) asymmetries [8–11]. Owing to strong finalstate interactions, manifested as either pressure or jet energy loss, these spatial asymmetries can be converted into final-state momentum anisotropies, leading to nonzero first-order and higher-order harmonic coefficients [11,12]. The observation of large v2 for pT  3–4 GeV at the RHIC [13,14] and LHC [15,16] has led to the conclusion that the hot and dense medium behaves like a “perfect fluid” [14,17,18]. This is because the large v2 values require hydrodynamic models [19–21] with a shear viscosity to entropy density ratio that is close to the conjectured lower bound of 1/4π [22,23]. Precise determination of this ratio using only v2 data is limited by many model uncertainties [24]. Because the shear viscosity tends to dampen the harmonics, with more damping for larger n [11,25,26], measurements of harmonic coefficients beyond v2 can provide stronger constraints for the shear viscosity of the medium. Extending these measurements to higher pT is also valuable for discriminating between The primary goal of high-energy heavy-ion physics is to understand the properties of the hot and dense matter created in nuclear collisions at facilities such as the Relativistic Heavy Ion Collider (RHIC) and the Large Hadron Collider (LHC). An important observable toward this goal is the azimuthal anisotropy of particle emission. At low pT (3–4 GeV), this anisotropy results from a pressure-driven anisotropic expansion of the created matter, with more particles emitted in the direction of the largest pressure gradients [1]. At higher pT , this anisotropy is understood to result from the path-lengthdependent energy loss of jets as they traverse the matter, with more particles emitted in the direction of smallest path-length [2]. These directions of maximum emission are strongly correlated, and the observed azimuthal anisotropy is customarily expressed as a Fourier series in azimuthal angle φ [3,4]: E d 3N dp3 d 2N = 2πpT dpT dη  1+2 ∞   vn (pT , η) cos n (φ − n ) , n=1 (1) * 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. 0556-2813/2012/86(1)/014907(41) 014907-1 ©2012 CERN, for the ATLAS Collaboration G. AAD et al. PHYSICAL REVIEW C 86, 014907 (2012) jet-quenching models, as high-pT vn is sensitive to the pathlength dependence of the jet energy loss [27–29]. These coefficients can also help to distinguish between different models of the initial geometry [30–33] and provide insights into the granularity of the initial state fluctuations [26,34–36]. Another related observable for studying the properties of the medium is the correlation function between two particles in relative azimuthal angle φ = φa − φb and pseudorapidity η = ηa − ηb [37]. The distribution of pairs in φ can be expanded into a Fourier series: ∞    dNpairs ∝1+2 vn,n pTa , pTb cos nφ, dφ n=1 (2) where the coefficients vn,n are symmetric functions with respect to pTa and pTb . The harmonics defined in Eq. (1) also contribute to this distribution: ∞      dNpairs ∝1+2 vn pTa vn pTb cos nφ, dφ n=1 (3) where the global direction n drops out in the convolution, and vn is assumed to be independent of η (which is approximately true within |η| < 2.5 at the LHC; see Sec. V A). Thus, if the anisotropy is driven by collective expansion, vn,n should factorize into the product of two single-particle harmonic coefficients [37]:     vn,n pTa , pTb = vn pTa )vn (pTb . (4) Such factorization may also be valid if the anisotropies of the two particles are independently driven by collective expansion and path-length-dependent jet energy loss (both are associated with the same initial spatial asymmetries). This factorization relation has been used to calculate the single-particle vn [38–40]. However, autocorrelations induced by resonance decays or fragmentation of back-to-back jets, are expected to break the factorization. Therefore, Eq. (4) can be used to identify the regions of pTa and pTb where correlations are dominated by effects controlled by the initial spatial asymmetries. The study of the structures of two-particle correlation in η and φ has been the focus of major experimental and theoretical efforts in the last decade. In typical proton-proton collisions, where a medium is presumably not formed, the pair distributions are dominated by strong correlation peaks at (φ, η) ∼ (0, 0) and φ ∼ π . These peaks reflect mainly autocorrelations among particles from fragmentation of backto-back jets. In heavy-ion collisions, additional structures have been observed for pT < 3–4 GeV and large η at φ ∼ 0 (known as the “ridge”) [41,42] and |φ − π | ∼ 1.1 (known as the “double-hump”) [37,43]. These unexpected structures have been interpreted as the response of the medium to the energy deposited by quenched jets [44,45]. However, similar structures can also be generated by the flow harmonics, as they all contribute constructively at φ ∼ 0 but tend to cancel on the away side according to Eq. (3) [8]. Therefore, a detailed comparison between the measured pair distribution [Eq. (2)] and that expected from anisotropic flow [Eq. (3)] can determine whether the structures in two-particle correlations are a consequence of the so-called “jet-induced medium response,” or whether they are a consequence of a sum of the flow harmonics. The v2 coefficient has been extensively studied at the RHIC [13,28,46–49] and LHC [15,16]. Results for higher-order vn for n ⩾ 3 also became available recently [30,31,40]. In contrast, no experimental measurement of v1 including systematic uncertainties exists at the LHC, although an estimate has recently been performed by a theoretical group [50] based on published ALICE data [39]. A primary complication for v1 measurements is global momentum conservation, which induces a significant dipole component [51,52]. A “sideward” deflection of colliding ions can also lead to a small rapidity-odd (i.e., changes sign crossing η = 0) dipole component [53,54]. Therefore, the extraction of v1 values associated with the initial dipole asymmetry requires careful separation of these contributions, which generally break the factorization relation given by Eq. (4). This paper presents comprehensive results for v1 –v6 over broad ranges of centrality, pseudorapidity, and √ pT for charged particles in lead-lead (Pb-Pb) collisions at sNN = 2.76 TeV with the ATLAS detector at the LHC. The vn values are measured directly using an “event plane” (EP) method for n = 2–6, and are also derived from the vn,n measured using a twoparticle correlation (2PC) method for n = 1–6. These detailed measurements provide new insights into the hydrodynamic picture at low pT , the jet energy loss picture at high pT , and the nature of the fluctuations in the initial geometry. They also allow a detailed study of the factorization relation [Eq. (4)] over broad ranges of centrality, η, pTa , and pTb . Together, these measurements should shed light on the physics underlying the structures observed in two-particle correlation functions. The paper is organized as follows. Sections II and III give a brief overview of the ATLAS detector, trigger, and selection criteria for events and tracks. Section IV discusses the details of the EP method and the 2PC method used to measure the vn . Section V A presents results for v2 –v6 from the EP method as a function of pT , η, and centrality. Section V B presents a detailed Fourier analysis of the twoparticle correlation functions to measure vn,n as a function of pTa , pTb , η, and centrality, which are then used to calculate v2 –v6 via the factorization relation [Eq. (4)]. These vn values are compared with those obtained from the EP method in Sec. V C, with a focus on understanding the structures of the 2PC in terms of single-particle vn . Section V D presents results for v1 based on a two-component fit of the v1,1 data with a modified functional form of Eq. (4) that includes the contribution of global momentum conservation. Section VI gives a summary of the results and main observations. II. ATLAS DETECTOR AND TRIGGER The ATLAS detector [55] provides nearly full solid angle coverage of the collision point with tracking detectors, calorimeters, and muon chambers, well suited for measurements of azimuthal anisotropies over a large pseudorapidity 014907-2 MEASUREMENT OF THE AZIMUTHAL ANISOTROPY FOR . . . PHYSICAL REVIEW C 86, 014907 (2012) range.1 This analysis primarily uses three subsystems for vn measurement: the inner detector (ID), the barrel and endcap electromagnetic calorimeters (ECals), and the forward calorimeter (FCal). The ID is contained within the 2-T field of a superconducting solenoid magnet and measures the trajectories of charged particles in the pseudorapidity range |η| < 2.5 and over the full azimuth. A charged particle passing through the ID typically traverses three modules of the silicon pixel detector (pixel), four double-sided silicon strip modules of the semiconductor tracker (SCT) and, for |η| < 2, a transition radiation tracker composed of straw tubes. The electromagnetic energy measurement of the ECal is based on a liquid-argon sampling technology. The ECal covers the pseudorapidity range |η| < 3.2 and is used as a reference detector in the event plane measurements. The FCal consists of three longitudinal sampling layers and extends the calorimeter coverage to |η| < 4.9. It uses tungsten and copper absorbers with liquid argon as the active medium, and has a total thickness of about ten interaction lengths. The centrality measurement uses towers in all three layers of the FCal, while the event plane measurements use towers in the first two layers of the FCal excluding those at the edge of the FCal η acceptance. These selection criteria are found to minimize the effect of fluctuations in the reaction plane measurement. The minimum-bias level-1 trigger used for this analysis requires signals in two zero-degree calorimeters (ZDCs), each positioned at 140 m from the collision point, detecting neutrons and photons with |η| > 8.3, or either one of the two minimum-bias trigger scintillator (MBTS) counters, covering 2.1 < |η| < 3.9 on each side of the nominal IP. The ZDC level-1 trigger thresholds on each side are set below the peak corresponding to a single neutron, for example, as produced from Coulomb dissociation of the lead ion [56]. A level-2 timing requirement based on signals from each side of the MBTS is imposed to remove beam backgrounds. Events satisfying these conditions are then required to have a reconstructed primary vertex within |zvtx | < 150 mm of the nominal center of the ATLAS detector for the EP analysis. A more stringent vertex cut of 100 mm is required for the 2PC analysis, such that enough events can be found in the same zvtx bin for the event mixing procedure (see discussion in Sec. IV B). About 48 × 106 and 43 × 106 events pass the requirements for the EP and 2PC analysis, respectively. Pileup probability is estimated to be at the 10−4 level and is therefore negligible. The Pb-Pb event  centrality is characterized using the total transverse energy ( ET ) deposited in the FCal over the pseudorapidity range 3.2 < |η| < 4.9 at the electromagnetic energy scale. An analysis of this distribution after all trigger and event selections gives an estimate of the fraction of the sampled non-Coulomb inelastic cross-section to be 98% ± 2% [57]. This estimate is  obtained from a shape analysis of the measured FCal ET distributions compared with a convolution of proton-proton data with a Monte Carlo Glauber calculation [58]. The FCal ET distribution is then divided into a set of 5% or 10% percentile bins, together with a bin defined for the 1% most central events. The uncertainty associated with the centrality definition is evaluated by varying the effect of trigger and event selection inefficiencies as well as background rejection requirements in the most peripheral  FCal ET interval [57]. Tracks are reconstructed within the full acceptance of the ID, requiring pT > 0.5 GeV and |η| < 2.5. To improve the reliability of track reconstruction in the high-multiplicity environment of heavy-ion collisions, more stringent requirements on track quality, compared to those defined for proton-proton collisions [59], are used. At least nine hits in the silicon detectors (out of a typical value of 11) are required for each track, with no missing pixel hits and not more than one missing SCT hit, in both cases where such hits are expected. In addition, the point of closest approach is required to be within 1 mm of the primary vertex in both the transverse and the longitudinal directions [16]. This selection is varied in the analysis to check the influence of both the acceptance and fake tracks. The tracking efficiency for charged particles is studied by comparing data to Monte Carlo calculations based on the HIJING event generator [60] and a full GEANT4 [61] simulation of the detector. This efficiency is estimated to be about 72% near midrapidity in central events. However, this analysis is found to be insensitive to variations in the tracking efficiency, as found previously [16]. Fake tracks from random combinations of hits are generally negligible, reaching only 0.1% for |η| < 1 for the highest multiplicity events. This rate increases slightly at large η. III. EVENT AND TRACK SELECTIONS This paper is based on approximately 8 μb−1 of Pb-Pb data collected in 2010 √ at the LHC with a nucleon-nucleon center-of-mass energy sNN = 2.76 TeV. An offline event selection requires a reconstructed vertex and a time difference |t| < 3 ns between the MBTS trigger counters on either side of the IP to suppress noncollision backgrounds. A coincidence between the ZDCs at forward and backward pseudorapidity is required to reject a variety of background processes, while maintaining high efficiency for non-Coulomb processes. IV. DATA ANALYSIS 1 ATLAS uses a right-handed coordinate system with its origin at the nominal interaction point (IP) in the center of the detector and the z axis along the beam pipe. The x axis points from the IP to the center of the LHC ring, and the y axis points upward. Cylindrical coordinates (r, φ) are used in the transverse plane, φ being the azimuthal angle around the beam pipe. The pseudorapidity is defined in terms of the polar angle θ as η = − ln tan(θ/2). Equation (1) implies that each harmonic component of the final-state momentum distribution is represented by its magnitude vn and azimuthal direction n . In general, any distribution can be expanded into a set of Fourier components. However, the distinguishing feature of correlation owing to initial geometry, as opposed to other sources of correlations, is that it is a “global” correlation. That is, n specifies a common 014907-3 G. AAD et al. PHYSICAL REVIEW C 86, 014907 (2012) direction, independent of the particle species, pT and η, and it drops out in the two-particle correlations [Eq. (3)]. This feature is quite different from the correlations expected from jet fragmentation or resonance decays, which typically involve a subset of particles correlated over a finite range in η with no preferred global direction. Thus, vn can be measured either by correlating tracks with the n estimated in the forward direction, or it can be measured from two-particle correlations with a large η gap. In the following, the details of these two methods are discussed. A. Event plane method The azimuthal angle n and the beam direction define the nth-order reaction plane.2 However, owing to incomplete detector acceptance and finite event multiplicity, the true reaction plane angle n cannot be determined. Instead, it is approximated by the event plane angle n , which is defined − → as the direction of the “flow vector” Q n , calculated in this analysis from the ET deposited in the FCal towers in each event: − → Q n = (Qx,n , Qy,n )   = ET cos nφ , ET cos nφ − (5)   ET sin nφ , ET sin nφ − tan nn = Qy,n , Qx,n where the sum ranges over towers in the first two layers of the FCal (see Sec. II). Subtraction of the event-averaged centroid removes biases owing to detector effects [62]. A standard flattening technique is then used to remove the residual nonuniformities in the event plane angular distribution [63]. These calibration procedures are similar to those used by the RHIC experiments [62,64]. The coefficient vn is measured by correlating tracks with n to obtain the raw values vnobs = cos n (φ − n ). The value of vnobs is then corrected by a resolution factor that accounts for the dispersion of n about n [4]: vn = cos n (φ − n ) vnobs , = Res{nn } cos n (n − n ) (6) where the average is performed over all events for the denominator and all tracks and all events for the numerator. The EP resolution of the FCal, Res{nn }, is [4] Res{nn } = cos n(n − n ) √  χn2 χn π − χn2 e 2 I0 + I1 = 2 2 χn2 2  , (7) where Iα are the modified Bessel functions of the first kind, and χn (known as the “resolution parameter”) is the 2 If the shape of Pb nuclei is approximated by the smooth WoodsSaxon function without fluctuations, n coincides with the azimuthal angle of the reaction plane defined by the beam axis and the impact parameter (the vector separating the barycenters of the two nuclei). fundamental variable that quantifies the precision of a detector for determining the event plane. The value of χn is proportional to the ET -weighted harmonic coefficient vnFCal and the square root of the total multiplicity M in the FCal acceptance [4]: √ χn ∝ vnFCal M. (8) The values of χn and Res{nn } are obtained from a twosubevents (2SE) method and a three-subevents (3SE) method [4]. In the 2SE method, the signal of a detector used to measure the event plane is divided into two “subevents” covering equal pseudorapidity ranges in opposite hemispheres, such that the two subevents nominally have the same resolution. The FCal detectors located at positive and negative η, FCalP and FCalN , provide such a division. The resolution of each FCal subevent is calculated directly from the correlation between the two subevents:         Res nnP(N) = cos n nP(N) − n = cos n nP − nN . (9) The resolution parameter of the FCal subevent χn,sub is determined by inverting √ Eq. (7). The√resolution parameter for the full FCal is χn = 2χn,sub , with 2 accounting for a factor of two increase in the total multiplicity [Eq. (8)]. Finally, χn is incorporated into Eq. (7) to obtain the resolution for the full FCal. In the 3SE method, the Res{nn } value for a given subevent A is determined from its correlations with two subevents B and C covering different regions in η:       A cos n nA − nB cos n nA − nC    Res nn = . (10) cos n nB − nC The large η coverage of the ID and ECal, with their fine segmentation, allows for many choices for subevents B and C. The ID and ECal are divided into a set of 22 reference subevents, each covering 0.5 units in η. The subevents B and C are chosen to ensure a minimum separation in η of 1 unit between all three subevents. This separation in η is required to suppress short-range correlations [62]. Various 3SE combinations are studied to check the sensitivity to the size of the chosen pseudorapidity gaps, as well as potential systematic effects owing to the explicit use of the correlation between FCalP and FCalN in the 2SE method. Figure 1 shows the values of χn and Res{nn } measured as a function of centrality for n = 2–6 using the full FCal. The data points and associated statistical uncertainties are calculated using the 2SE method. However, 5% upward and 15% downward centrality-independent corrections are applied to n = 5 and n = 6, respectively, to adjust to the average of the 2SE and the 3SE estimates for the Res{nn }. The differences between the two estimates are quoted as systematic uncertainties for Res{nn }, and they are propagated via Eq. (7) to obtain a systematic uncertainty for χn . In this analysis, the centrality range for each harmonic n is chosen such that the relative statistical uncertainty for Res{nn } is less than 30% of its mean value, and the 2SE and 3SE estimations show good agreement. They are 0%–80% for v2 , 0%–70% for v3 and v4 , and 0%–50% for v5 and v6 , as indicated in Fig. 1. 014907-4 MEASUREMENT OF THE AZIMUTHAL ANISOTROPY FOR . . . n=2 n=3 n=4 n=5 n=6 full FCal Pb−Pb sNN=2.76 TeV Lint= 8 μ b-1 χn 2 ATLAS 1 0 80 n=2 n=3 n=4 n=5 n=6 1 Res{nΨn} 60 40 20 Centrality [%] ATLAS full FCal 0 Pb−Pb sNN=2.76 TeV Lint= 8 μ b-1 0.5 0 80 60 40 20 Centrality [%] 0 FIG. 1. (Color online) The χn (top) and EP resolution factor (bottom) vs centrality (smaller value refers to more central events) for n = 2–6, together with the systematic uncertainty as shaded bands. The EP is measured by both sides of the FCal detector (denote by “full FCal”). Note that the Res{nn } value cannot be greater than 1 [see Eq. (7)]; thus, its systematic uncertainty shrinks as it approaches 1. In the event plane analysis, two complementary methods are employed to measure vn . The first (“full FCal”) method calculates vnobs by correlating tracks in the ID with the EP from the full FCal detector; the resolution correction for the full FCal is then applied to obtain the final vn [Eq. (6)]. In the second (“FCal subevent” or simply FCalP(N) ) method [16], tracks with η ⩾ 0 (η < 0) are correlated with the PHYSICAL REVIEW C 86, 014907 (2012) EP in the opposite hemisphere given by the FCalN (FCalP ). The resolution correction for the FCalP(N) is then applied to obtain the final vn . Note that the relative statistical and systematic uncertainties for Res{nn } are almost identical for the two methods because they both rely on similar subevent correlations [Eqs. (9) and (10)]. However, the values of vnobs from the FCal subevent method are smaller than those from the full FCal method owing to its poorer EP resolution. Therefore, these vnobs values have a larger fractional statistical uncertainty. The primary advantage of the FCal subevent method is that it increases the minimum (maximum) pseudorapidity separation between the track and the EP from about 0.8 (4.8) units for the full FCal method to about 3.3 (7.3) units. Thus, the subevent approach is less affected by short-range autocorrelations, stemming primarily from jet fragmentation and resonance decays. In this analysis, the FCalP(N) method is used for the η dependence of vn to minimize short-range correlations, while the full FCal method is used for the pT and centrality dependence of vn to optimize the EP resolution (see Sec. V A). However, the potential influence of short-range correlations on the full FCal method is cross-checked with the FCalP(N) method. Good agreements are always observed for η-integrated vn , within the systematic uncertainties for the two methods. The systematic uncertainty in vnobs is determined by varying the track quality cuts, comparing data for different running periods, varying the full centrality range by ±2% according to the uncertainty in the trigger and event selections, as well as by determining the value of sin n(φ − n ). The study of track quality cuts accounts for influences of background contaminations and tracking inefficiency. Finite sine terms can arise from detector biases that introduce a correlation between the ID and the FCal. Their magnitudes relative to the cosine terms are included in the uncertainty for vnobs . All these uncertainties are generally small, except for n = 6. They are also quite similar for the full FCal and FCal subevent methods, so the larger of the two is quoted as the main uncertainty. As a cross-check, v2 –v6 values are also extracted using the EP measured either for the three layers of FCal individually or for two η regions of FCal (3.3 < |η| < 4.0 and 4.0 < |η| < 4.8). Although these five FCal subevents have up to a factor of four difference in their resolution corrections, the measured v2 –v6 all agree to within 2%–10%. Tables I–V summarize the systematic uncertainties for v2 –v6 in various centrality intervals. The total uncertainties are calculated as the quadrature sum of all sources in these TABLE I. Summary of relative systematic uncertainties in percentage for v2 for both full FCal and FCalP(N) . See text for explanation of the arrangement of the uncertainties. Centrality 0%–20% 20%–50% 50%–70% 70%–80% Resolution (%) Track selection (%) Residual sine term (%) Running periods (%) Trigger and event selection (%) Total (%) 5.0–2.0 2.0 0.8 0.2 1.0 5.6–3.2 1.0–2.0 0.5 0.6 0.2 1.0–0.5 1.4–2.3 3.0–4.0 0.5 0.5 0.5 1.0 3.4–4.2 4.0–6.0 1.0 0.2 1.0 1.5 4.6–6.4 014907-5 G. AAD et al. PHYSICAL REVIEW C 86, 014907 (2012) TABLE II. Summary of relative systematic uncertainties in percentage for v3 for both full FCal and FCalP(N) . TABLE IV. Summary of relative systematic uncertainties in percentage for v5 for both full FCal and FCalP(N) . Centrality (%) Centrality 0%–20% 20%–50% 50%–70% Resolution (%) Track selection (%) Residual sine term (%) Running periods (%) Trigger and event selection (%) Total (%) 3.0 2.0 1.0 0.5 0.4 3.8 3.0 0.5 1.0 0.5–1.5 0.5–1.0 3.5–3.9 3.0–5.6 0.5–2.0 1.5 2.0 1.5–3.5 4.6–7.4 tables. In most cases, they are specified for multiple 5% wide centrality intervals. For example, 0%–20% in Tables I–V refers to four bins: 0%–5%, 5%–10%, 10%–15%, and 15%–20%; a “5.0–2.0” notation indicates the values of relative systematic uncertainty in percentage at the beginning and at the end of the 0%–20% centrality interval. Tables IV and V also quote the uncertainty for the 0%–1% centrality interval, which generally has the same systematic uncertainty as that for the 0%–5% centrality interval, but with a larger statistical uncertainty. The systematic uncertainties include only those associated with the measurements themselves, and no attempt is made to disentangle potential contributions from various sources of autocorrelations, as their exact origin and quantitative effects on vn are not fully understood [5]. Nevertheless, these autocorrelations should be largely suppressed by the large average η gap between the ID and the detector used for determining the EP. B. Two-particle correlation method The two-particle correlation function is generally defined as the ratio of the same-event pair (foreground) distribution to the combinatorial pair (background) distribution in two-particle phase space (φa , φb , ηa , ηb ): C(φa , φb , ηa , ηb ) = d4N dφa dηa dφb dηb 2N . d2N × dφdb dη dφa dηa b (11) In practice, the correlation function is usually studied as a function of relative azimuthal angle (φ) and relative pseudorapidity (η), by averaging pair distributions over the detector acceptance: C(φ, η) = S(φ, η) , B(φ, η) (12) TABLE III. Summary of relative systematic uncertainties in percentage for v4 for both full FCal and FCalP(N) . 0%–1% 0%–20% 20%–40% 40%–50% Resolution (%) Track selection (%) Residual sine term (%) Running periods (%) Trigger and event selection (%) Total (%) where Resolution (%) Track selection (%) Residual sine term (%) Running periods (%) Trigger and event selection (%) Total (%) 0%–20% 20%–50% 50%–70% 4.0 1.0 2.0 1.0 0.6 4.9 4.0 1.0–2.0 2.0 1.5–2.0 0.7 4.9–5.4 4.4–16.0 4.0 3.0–5.0 4.0 1.0–2.0 7.9–17.5 10.2 1.0 2.0 2.0 12.1 11.6  S(φ, η) = 11.6–12.1 13.0–23.0 d 4N , dφa dηa dφb dηb (13) dφa dηa dφb dηb δab d 2N d 2N . dφa dηa dφb dηb (14)  B(φ, η) = 10.2–10.4 11.2–22.4 1.0 2.0 5.0 2.0 4.0 1.0 dφa dηa dφb dηb δab The δab is a shorthand notation for δ(φa − φb − φ)δ(ηa − ηb − η). For an ideal detector, the combinatorial pair distribution is uniform in φ and has a nearly triangular shape in η owing to the weak dependence of the single-particle distribution on η [57]. In reality, both same-event and combinatorial pair distributions are modulated by detector inefficiencies and nonuniformity. These detector effects influence the two distributions in the same way so they cancel in the ratio. Therefore, B(φ, η) is often referred to as the pair acceptance function [37]. In this analysis, B(φ, η) is estimated from track pairs from two events with similar centrality (matched within 5%) and zvtx (matched within 1 mm). The two particles in the pair are typically selected with different conditions, such as different pT ranges, pseudorapidities, and charge signs. In this analysis, charged particles measured by the ID with a pair acceptance extending up to |η| = 5 are used. Figure 2(a) shows the two-dimensional (2D) correlation function for pairs from a representative pT range of 2–3 GeV and 0%–5% centrality interval. It reveals the characteristic long-range near-side ridge and away-side double-hump structures that were observed previously in heavy-ion collisions at the RHIC for a similar pT range [41–43]. A narrow short-range correlation peak is also seen at (φ, η) ∼ (0, 0), presumably owing to autocorrelations from jet fragmentation and resonance decays. From the 2D correlation function, a oneTABLE V. Summary of relative systematic uncertainties in percentage for v6 for both full FCal and FCalP(N) . Centrality Centrality 10.8 1.0 Resolution (%) Track selection (%) Residual sine term (%) Running periods (%) Trigger and event selection (%) Total (%) 014907-6 0%–1% 0%–20% 20%–40% 40%–50% 58 34–31 31 32–38 36 37–42 10 10 10 1 61 38–35 MEASUREMENT OF THE AZIMUTHAL ANISOTROPY FOR . . . PHYSICAL REVIEW C 86, 014907 (2012) FIG. 2. (Color online) The steps involved in the extraction of vn values for 2–3 GeV fixed-pT correlations in the 0%–5% centrality interval: (a) 2D correlation function; (b) the 1D φ correlation function for 2 < |η| < 5 (rebinned into 100 bins), overlaid with contributions from the individual vn,n components and their sum, as well as the residual difference between the data and the sum; (c) Fourier coefficient vn,n vs |η| for n = 1–6; and (d) vn vs |η| for n = 2–6. The shaded bands in (c) and (d) indicate the systematic uncertainties, as described in the text. According to Eq. (4), if the correlations are dominated by those arising from asymmetry of the initial geometry such as flow, vn,n should factorize into the product of two singleparticle harmonic coefficients. This is found to be the case for n ⩾ 2 at low pT for pairs with a large η gap, but is not true for n = 1 (see Secs. V B and V C), similar to what was also found in other measurements [39,40]. Thus, if the two particles are selected from the same pT interval (“fixed-pT ” correlations) as as in Fig. 2, the single-particle vn for n ⩾ 2 can be calculated  √ vn = vn,n . When vn,n < 0, vn is defined as vn = − |vn,n | (or  vn = vn,n / |vn,n | in general). This calculation is repeated for all 1D correlation functions in each |η| slice. The resulting full |η| dependencies of vn,n and vn are shown in Figs. 2(c) and 2(d), respectively. The vn,n and vn values are found to vary rapidly for |η|  1, presumably reflecting the influence of the shortrange correlation at (φ, η) ∼ (0, 0) [Fig. 2(a)], but they decrease much more slowly for larger |η|. This slow decrease is expected because the single-particle vn also decreases very slowly with η (see Fig. 3), and the factorization relation Eq. (4) is valid for the present pT range (see Sec. V B). These behaviors suggest that the autocorrelations from near-side jet fragmentation and resonance decays can be largely eliminated by requiring a large η gap (e.g., |η| > 2). Each “fixed-pT ” correlation function provides a reference vn for a chosen pT range (denoted by superscript “a”). Tracks from this pT range are then correlated with those from a target pT range (denoted by superscript “b”), and this “mixed-pT ” correlation is used to calculate vn,n and to obtain the vn in the target pT via Eq. (4). Because factorization is expected to be valid for the anisotropies driven by the initial geometry, but is broken by the presence of autocorrelations among the jet fragmentation products, the level of consistency between vn obtained from different reference pT ranges reveals whether the 2PC is dominated by anisotropies driven by the initial geometry. A detailed study of the factorization properties of v1 –v6 is presented in Sec. V B. The correlation function relies on the pair acceptance function to reproduce and cancel the detector acceptance effects in the foreground distribution. Mathematically, the pair acceptance function in φ is simply a convolution of two single-particle azimuthal distributions and should be uniform in φ without detector imperfections. A natural way of quantifying the influence of detector effects on vn,n and vn is to transform the single-particle and pair acceptance functions into the Fourier space. The resulting coefficients det for pair acceptance vn,n are the product of those for the two single-particle acceptances vndet,a and vndet,b . In general, the pair acceptance function is quite flat: The maximum variation from its average is observed to be less than 0.001 for pairs det | integrated over 2 < |η| < 5, and the corresponding |vn,n −4 det values are found to be less than 1.5 × 10 . These vn,n values are expected to mostly cancel in the correlation function, and only a small fraction contributes to the uncertainties of the pair det acceptance function. Three possible residual effects for vn,n are studied: (1) the time dependence of the pair acceptance, (2) the effect of imperfect centrality matching, and (3) the effect of det imperfect zvtx matching. In each case, the residual vn,n values dimensional (1D) φ correlation function can be constructed for a given η interval:  S(φ, η)dη . (15) C(φ) = A ×  B(φ, η)dη The normalization constant A is determined by scaling the number of pairs in 2 < |η| < 5 to be the same between the foreground (S) and background (B). This normalization is then applied to other η intervals. Each 1D correlation function is expanded into a Fourier series according to Eq. (2), with coefficients vn,n calculated directly via a discrete Fourier transformation (DFT): N cos(nφm )C(φm ) , (16) vn,n = cos nφ = m=1N m=1 C(φm ) where n = 1–15, and N = 200 is the number of φ bins. A small upward relative correction is applied (∼0.15% for n = 6 and increasing to 1% for n = 15) to account for the finite φ bin width. Figure 2(b) shows one such 1D correlation function for 2 < |η| < 5, overlaid with the corresponding contributions from individual vn,n components. The shape of the correlation function is well described by the sum of the first six vn,n components. 014907-7 G. AAD et al. PHYSICAL REVIEW C 86, 014907 (2012) comparison in Table VI and the pair acceptance. This fraction (point to point in pT ) is estimated to be about 30% of the final systematic uncertainty, and the remaining uncertainty is treated as a pT -correlated systematic uncertainty. They are used in the discussion of the v1,1 results in Sec. V D. V. RESULTS A. v2 –v6 from the event plane method Figure 3 shows the η dependence of vn for several centrality intervals in the 2- to 3-GeV pT range from the FCalP(N) EP method. Similar behaviors are observed in other pT ranges (see also [16] for v2 ). The v2 values decrease by less than 5% toward large |η| for central and midcentral events, and the decrease is more pronounced both for n ⩾ 3 and for peripheral events. Pb-Pb sNN=2.76 TeV 0.2 Lint= 8 μb-1 0.1 FCalP(N) EP 24 GeV), v2 increases toward more peripheral events, presumably reflecting the dominance of autocorrelations from dijets. In an ideal hydrodynamics scenario, vn at low pT is a power-law function of the radial expansion velocity of the fluid, leading to the qualitative expectation that vn (pT ) is a power-law function of pT [9,66]. Previous RHIC results 1/4 1/2 have shown that v4 /v22 (or equivalently v4 /v2 ) is almost 1/n 1/2 independent of pT [48,49].3 Figure 6 shows vn /v2 vs pT for various centrality intervals. These ratios vary weakly with pT except in the 5% most central events, suggesting that such a scaling relation largely accounts for the pT dependence. 3 This v4 was measured relative to the 2 instead of the 4 reaction plane and is known as mixed harmonics [4]. It can be regarded as a projection of v4 measured in the 4 onto the 2 . 014907-9 G. AAD et al. PHYSICAL REVIEW C 86, 014907 (2012) n=3 3 than those for n = 3. A similar centrality dependence was observed for the v4 /v22 ratio at the RHIC and was argued to reflect the centrality dependence of fluctuations in the initial geometry [49]. n=4 ATLAS Pb-Pb sNN=2.76 TeV L int= 8 μb-1 |η|<2.5 full FCal EP B. v2 –v6 from the two-particle correlation method 1/2 v1/n n /v 2 2 1 n=5 3 0-1% 0-5% n=6 5-10% 10-20% 20-30% 40-50% 2 1 2 4 2 6 4 6 pT [GeV] 1/2 FIG. 6. (Color online) vn1/n /v2 vs pT for several centrality intervals. The shaded bands indicate the total systematic uncertainties. However, the overall magnitudes of the ratios seem to vary with centrality and also vary slightly with n. 1/n 1/2 Figure 7 shows the centrality dependence of vn /v2 for 2 < pT < 3 GeV. Given that the ratios vary weakly with pT , the results for other pT ranges are similar. The ratios are almost independent of centrality in midcentral and peripheral events, but then increase sharply toward more central events, with a total change of almost a factor of two over the 0%–20% centrality range. In addition, the ratios for n = 4–6 are similar to each other, while they are systematically higher ATLAS 2.5 1/2 v1/n n /v 2 Pb-Pb sNN =2.76 Te V L int = 8 μ b-1 2 full FCal EP n=3 n=4 n=5 n=6 2

2 is required to suppress the near-side jet peak and to expose the long-range structures. At low pT , the vn,n components are mainly driven by these long-range structures. However, for pTa , pTb > 6 GeV, they are dominated by the pronounced 014907-10 MEASUREMENT OF THE AZIMUTHAL ANISOTROPY FOR . . . 0-1% PHYSICAL REVIEW C 86, 014907 (2012) 0-5% 5-10% 1.02 1.02 1.01 1 1 0.99 1.04 1.1 1.02 1.05 1 1 0.98 0 Δφ 2 0 2 4 -4 -2 4 0 Δφ Δη 0 2 4 -4 -2 20-30% C(Δφ, Δη) 10-20% 2 0.95 4 0 Δφ Δη 0 2 4 -4 -2 30-40% 2 4 0 Δφ Δη 1.2 1.1 1.1 1.1 1 1 1 1 0.9 0.9 Δφ 2 4 -4 -2 2 0 0 Δφ Δη 2 4 -4 -2 60-70% 0 2 0.9 0 Δφ 2 4 -4 -2 2 0 0 Δφ Δη 2 4 -4 -2 70-80% 0 2 4 0 Δφ Δη 1 0.9 0.8 Δφ Δη 2 4 -4 -2 0 2 4 -4 -2 4 Δη Pb-Pb sNN=2.76 TeV 1 0 2 2 0 ATLAS 1.4 1.2 4 Δη 80-90% 1.1 1 -2 0.9 4 1.2 1.1 -4 4 50-60% 1.2 4 4 40-50% 1.1 0 2 0 2 Lint= 8 μb-1 4 0 Δφ Δη 2 4 -4 -2 0 2 a 2 < p , pb < 3 GeV 4 T T Δη FIG. 8. (Color online) 2D correlations for 2 < pTa , pTb < 3 GeV in several centrality intervals. The near-side jet peak is truncated from above to better reveal the long-range structures in η. 1.04 ATLAS Lint= 8 μb-1 2<|Δη|<5 1.05 a 3 1. Despite a factor of three variation in the reference vn value for the pT ranges from 0.5–1 GeV to 3–4 GeV (see Fig. 4), the extracted vn (pTb ) for n = 2–6 is constant within statistical uncertainties for |η| > 1. This factorization check is repeated for n = 2–6 for various ranges of pTa , pTb , and centrality. For correlations with the default choice of |η| > 2 used in this paper, factorization is found to hold at the 5%–10% level for pT < 3–4 GeV in the 70% most central events. Further studies of this topic are presented in Sec. V C. Figure 13 shows the extracted Fourier coefficients, vn vs n, for several fixed-pT correlations with |η| > 2 and for the 1% most central events. The pTa,b ranges are restricted to below 4 GeV, where the factorization for v2 –v6 works reasonably well. Significant signals are observed for n ⩽ 6, while the T C. Comparison of v2 –v6 from the two methods Figure 14 compares the centrality dependence of vn obtained from the 2PC method and the EP method at low pT . They agree within 5% for v2 –v4 , 10% for v5 , and 15% 1 6 are consistent with zero within statistical and systematic uncertainties. Note that in cases where the uncertainty of vn,n is comparable to its own value, the vn uncertainty becomes asymmetric and highly non-Gaussian. A complete compilation of the Fourier coefficients for other centrality intervals is included in Figs. 22–27 in the Appendix. Recent model comparisons indicate that these results can provide important constraints on the nature of initial geometry fluctuations and the shear viscosity of the created matter [34]. T 2 4–5 GeV, presumably owing to contributions from the away-side jet (see Fig. 10). Figure 16 compares the v2 –v5 obtained from the mixed-pT 2PC method with those from the EP method in the 0%–10% centrality interval. The layout of this figure is similar to that of Fig. 15, except that the vn (pTb ) are calculated using Eq. (17) for four reference pTa ranges. Because only one particle is chosen from high pT , the influence of autocorrelations from jets is expected to be less than when both particles are chosen from high pT . Indeed, when pTa is below 3 GeV, the deviations from the EP method are less than 5%–10% out to much higher pTb , when statistical precision allows. This behavior suggests that as long as the reference pTa is low, the factorization relation is valid and hence the vn can be measured to high pT via the 2PC method. It is also noticed that in central events, the agreement for v2 is generally worse than that for the higherorder vn : The differences for v2 between the two methods reach >20% for pTb > 4 GeV and 3 < pTa < 4 GeV, while the differences for the higher-order vn are small and show no visible dependence on pTb for pTa < 4 GeV. Comparisons for several other centrality intervals are included in Fig. 28 in the Appendix. Good agreement is always observed for v2 –v5 in noncentral events. The agreement between the 2PC and EP methods implies that the structures of the two-particle correlation function 0 1.2 2PC 1 EP 0.8 0 2 4 6 pa [GeV] 8 T FIG. 15. (Color online) Comparision of vn (pT ) and the ratios between the fixed-pT 2PC method and the EP method (v2 –v5 from top row to bottom row) for the 0%–10% centrality interval. The error bars indicate the statistical uncertainties only. The dashed lines in the ratio plots indicate a ±10% band to guide eye. at low pT and large |η| mainly reflect collective flow. This is verified explicitly by reconstructing the correlation 014907-14 MEASUREMENT OF THE AZIMUTHAL ANISOTROPY FOR . . . ATLAS v2 0.15 0-10% 1.002 2PC v EP n,n (n=2-6) +v 1,1 ATLAS n=2 Pb-Pb sNN=2.7 2.76 TeV eV Lint= 8 μ b-1 2<|Δη|<5 2<|Δ η|<5 0-1% v (n=3,5) v 2PC 1,1 1 1 0 1.2 Lint= 8 μb-1 EP n,n 1.001 0.05 Pb-Pb sNN=2.76 TeV 1.005 v EP n,n (n=2,4,6) 0.1 0.999 a C(Δφ) 2PC 1 EP 0.8 a 0.5 < p ,pb < 1 GeV T T a 1 < p ,pb < 2 GeV 0.995 1.02 1.04 1.01 1.02 T T a 2PC 0.5

2. A previous analysis of published RHIC data supports the existence of a significant rapidity-even v1 at lower beam energy [12]. Here the properties of the v1,1 data are examined in detail, with the goal of understanding the relative contributions of v1 and global momentum conservation. Figure 18 shows the |η| dependence of the v1,1 extracted from two-particle fixed-pT correlations for several centrality intervals (see discussions around Fig. 2). The v1,1 values for |η| < 0.5 are always positive, presumably owing to contributions from the near-side jet or other short-range correlations. The v1,1 values at large |η|, however, have a more complex dependence on centrality and pT . In peripheral events, the v1,1 values are always negative at |η| > 1, and their absolute values increase up to |η| ∼ 1.6 and then decrease for larger |η|. This trend is consistent with the gradual falloff of the away-side jet contribution at large |η| seen in 2D correlation function, for example, the last panel of Fig. 8. In central and midcentral events, the v1,1 values at |η| > 1 are negative for pTa , pTb < 2 GeV, but they become positive and relatively independent of |η| at higher pT , consistent with the contribution from a rapidity-even v1 that varies weakly with η. Given that the |η| dependence of Fig. 18 is generally weak at large |η|, v1,1 is integrated over 2 < |η| < 5 to obtain one value for each pTa and pTb combination. The v1,1 (pTa , pTb ) functions are then shown in Fig. 19 for various centrality intervals. In peripheral events, the v1,1 values are always negative and their absolute values increase nearly linearly with pTa and pTb , presumably dominated by autocorrelations from the away-side jet. In more central events, the absolute value of this negative v1,1 component is smaller, reflecting a dilution of the momentum conservation term by the large event multiplicity. Furthermore, Fig. 19 clearly suggests that a positive v1,1 component sets in for 2  pT  6 GeV. Its magnitude increases for more central events and eventually drives v1,1 into the positive region. The v1,1 values reach a       pa pb v1,1 pTa , pTb ≈ v1 pTa v1 pTb − T 2T . M pT 014907-16 MEASUREMENT OF THE AZIMUTHAL ANISOTROPY FOR . . . a PHYSICAL REVIEW C 86, 014907 (2012) a a a 0.5< p <1 GeV 1< p <1.5 GeV 1.5< p <2 GeV 2< p <3 GeV 3< pa <4 GeV T 4< pa <6 GeV T 6< pa <8 GeV T 8< pa <20 GeV T T T 0-5% 0.005 T T 10-20% 20-30% 40-50% 50-60% 0 -0.005 -0.01 30-40% 0.005 ATLAS T v1,1(pa,pb) Pb-Pb sNN=2.76 TeV Lint= 8 μb-1 2<|Δ η|<5 T 0 -0.005 -0.01 0.005 60-70% 70-80% 80-90% × 0.1 × 0.1 × 0.1 0 -0.005 -0.01 0 2 4 6 8 10 0 2 4 6 8 10 0 2 4 6 8 10 pb [GeV] T FIG. 19. (Color online) v1,1 (pTa , pTb ) for 2 < |η| < 5 vs pTb for different pTa ranges. Each panel presents results in one centrality interval. The error bars and shaded bands represent statistical and systematic uncertainties, respectively. The data points for the three highest pTa intervals have coarser binning in pTb , hence are connected by dashed lines to guide the eye. The data in the bottom three panels are scaled down by a factor of ten to fit within the same vertical range. maximum at around 4 GeV, similar to those for the higher-order harmonic coefficients (see Fig. 4). To validate the two-component model of Eq. (20) as a reasonable interpretation of the v1,1 data and to extract the v1 , a least-squares fit of the v1,1 data is performed for each centrality interval with the following form:           v1,1 pTa , pTb − v1Fit pTa v1Fit pTb − cpTa pTb 2 χ = ,  stat 2  sys,p2p 2 σa,b + σa,b a,b (21) 2 sys,p2p stat and σa,b denote the statistical and point-to-point where σa,b systematic uncertainties for v1,1 (pTa , pTb ), respectively (see Sec. IV B). The v1Fit (pT ) function is defined via a smooth interpolation of its values at m discrete pT points, v1Fit (pT,i )|m i=1 , and these together with the parameter c constitute a total of m + 1 fit parameters. An interpolation procedure is used because the functional form of v1Fit (pT ) is, a priori, unknown, yet it is expected to vary smoothly with pT . In the default setup, the v1,1 data used in the fit are restricted to pTa < 6 GeV and pTb < 10 GeV, giving a total of 129 data TABLE VII. Quality of the fit χ 2 /DOF, fit parameter c, and corfor various centrality responding multiplicity scaled values c dN dη intervals. The uncertainty of c dN dη |η=0 is calculated as the quadrature |η=0 sum of uncertainties from c and dN of Ref. [57]. The bottom row dη |η=0 lists the 15 pT interpolation points used in the default fit. Centrality 0%–5% 5%–10% 10%–20% 20%–30% 30%–40% 40%–50% 014907-17 χ 2 /DOF c(×0.001GeV−2 ) c dN dη |η=0 (×GeV−2 ) 159/113 0.24 ± 0.02 0.39 ± 0.04 133/113 0.28 ± 0.02 0.37 ± 0.04 165/113 0.35 ± 0.03 0.36 ± 0.04 134/113 0.50 ± 0.04 0.34 ± 0.03 188/113 0.75 ± 0.05 0.33 ± 0.03 181/113 1.16 ± 0.09 0.32 ± 0.03 15 interpolation points used in the default fit: 0.5, 0.7, 0.9, 1.1, 1.3, 1.5, 2.0, 2.5, 3.0, 3.5, 4.5, 5.5, 6.5, 7.5, 9.0 GeV G. AAD et al. PHYSICAL REVIEW C 86, 014907 (2012) points for each centrality interval. A cubic-spline interpolation procedure is used, and the number of interpolation points is chosen to be m = 15 (listed in Table VII). To account for sys,corr the pT -correlated uncertainty (σa,b , see Sec. IV B), the least-squares minimization is repeated after varying all the sys,corr v1,1 data points up or down by σa,b . Because v1,1 (pTa , pTb ) is a b symmetric with respect to pT and pT , data points for pTa < pTb are correlated with those for pTa > pTb . The effect of this correlation is evaluated by repeating the fit using only v1,1 data for pTa ⩽ pTb or v1,1 data for pTa ⩾ pTb . The variations from the default fit are included in the systematic uncertainties. Figure 20 shows the fit to the v1,1 data from Fig. 19 for the 0%–5% and 40%–50% centrality intervals. The v1,1 data are plotted as a function of pTb for six intervals of pTa . The seemingly complex patterns of the v1,1 data are well a a 0.5 2 (Fig. 18), is consistent with a rapidity-even v1 that is almost independent of η. If the two-component ansatz of Eq. (20) is valid, the fit parameter c should be inversely proportional to multiplicity M and pT2  of the whole event. This ansatz is checked by calculating the product of c and the charged hadron multiplicity at midrapidity dN from [57], with the assumption that dη |η=0 dN is proportional to M. The results are summarized in dη |η=0 Table VII for each centrality interval. Because pT2  for the whole event is expected to vary weakly with centrality (the pT  for charged pions at midrapidity only varies by ∼5% within the 0%–50% centrality interval at the LHC [70]), the product is also expected to vary weakly with centrality. Table VII shows that this is indeed the case, supporting the assumptions underlying Eq. (20). VI. CONCLUSION In summary, differential measurements of harmonic coeffiof charged particles cients v1 –v6 for the azimuthal distributions √ are presented for lead-lead collisions at sNN = 2.76 TeV, based on an integrated luminosity of approximately 8 μb−1 recorded by ATLAS. The vn values are measured for n = 2–6 using an event plane method and are also derived for n = 1–6 from the vn,n measured in a two-particle correlation method. In the event plane method, v2 –v6 are extracted as a function of transverse momentum (0.5 < pT < 20 GeV), pseudorapidity (|η| < 2.5), and centrality (0%–80%) by correlating tracks with the event plane determined at large rapidity. The vn values exhibit a weak η dependence, slightly decreasing toward larger |η|. The vn values show a similar pT dependence, namely, increasing with pT to a maximum around 3–4 GeV and then decreasing for higher pT . The higher-order coefficients exhibit stronger pT variations. They follow an approximate scaling 1/n 1/2 relation, vn (pT ) ∝ v2 (pT ), except in the top 5% most central collisions. Furthermore, the coefficients v3 –v6 show little dependence on centrality, consistent with an anisotropy 014907-19 G. AAD et al. PHYSICAL REVIEW C 86, 014907 (2012) primarily associated with fluctuations in the initial geometry. In contrast, v2 varies strongly with centrality, reflecting an anisotropy mainly associated with a change in the average elliptic geometry with centrality. In the two-particle azimuthal correlation method, harmonic coefficients for the distributions of the charged particle pairs in φ, vn,n = cos nφ are extracted over a broad range of pT , relative pseudorapidity (|η| < 5) and centrality. For pairs of charged particles with a large pseudorapidity gap (|η| > 2) and one particle at pT < 3 GeV, the v2,2 –v6,6 are found to factorize into the product of corresponding vn harmonics, that is, vn,n (pTa , pTb ) ≈ vn (pTa )vn (pTb ), in central and midcentral events. This suggests that these values of v2,2 –v6,6 are consistent with the response of the created matter to fluctuations in the geometry of the initial state. This factorization does not describe the v2,2 –v6,6 data in more peripheral collisions or at higher pT , primarily owing to autocorrelations from dijets. The factorization relation is also found to fail for v1,1 over the entire pT range. A detailed investigation of the v1,1 (|η|, pTa , pTb ) data suggests that they are consistent with the combined contributions from a rapidity-even v1 component, and a global momentum conservation component that increases linearly with the pTa × pTb and is inversely proportional to the event multiplicity. Motivated by this observation, a two-component fit is used to extract v1 . The derived signal is negative at low pT , changes sign at pT ≈ 1.0 GeV, reaches a maximum at around 4–5 GeV and then decreases at higher pT . The magnitude of v1 at the peak is comparable to that of v3 . At low pT where single-particle anisotropy is dominated by collective flow, the two-particle correlation functions are found to be well reproduced by combining the contributions of v2 –v6 from the event plane method and the v1,1 [Eq. (18)]. This implies that the main structures of the low pT two-particle correlation for |η| > 2 largely reflect higher-order collective flow (n ⩾ 2), together with a cos φ term that contains contributions from both a rapidity-even v1 and the global momentum conservation. Therefore, the low pT correlation functions do not allow significant contributions from jet-induced medium responses. The fluctuations in the initial geometry together with the low viscosity of the created matter are potentially responsible for these vn,n coefficients. A detailed comparison of the comprehensive vn and vn,n data presented in this paper with viscous hydrodynamic model calculations at low pT [25,26,32,34,71,72] and jet energy loss model calculations at high pT [27,29,73,74] may help elucidate the nature of these fluctuations and better constrain the transport properties of the medium. ACKNOWLEDGMENTS We thank CERN for the very successful operation of the LHC, as well as the support staff from our institutions without whom ATLAS could not be operated efficiently. We acknowledge the support of ANPCyT, Argentina; YerPhI, Armenia; ARC, Australia; BMWF, Austria; ANAS, Azerbaijan; SSTC, Belarus; CNPq and FAPESP, Brazil; NSERC, NRC, and CFI, Canada; CERN; CONICYT, Chile; CAS, MOST, and NSFC, China; COLCIENCIAS, Colombia; MSMT CR, MPO CR, and VSC CR, Czech Republic; DNRF, DNSRC, and Lundbeck Foundation, Denmark; ARTEMIS and ERC, European Union; IN2P3-CNRS, CEA-DSM/IRFU, France; GNAS, Georgia; BMBF, DFG, HGF, MPG, and AvH Foundation, Germany; GSRT, Greece; ISF, MINERVA, GIF, DIP, and Benoziyo Center, Israel; INFN, Italy; MEXT and JSPS, Japan; CNRST, Morocco; FOM and NWO, Netherlands; RCN, Norway; MNiSW, Poland; GRICES and FCT, Portugal; MERYS (MECTS), Romania; MES of Russia and ROSATOM, Russian Federation; JINR; MSTD, Serbia; MSSR, Slovakia; ARRS and MVZT, Slovenia; DST/NRF, South Africa; MICINN, Spain; SRC and Wallenberg Foundation, Sweden; SER, SNSF, and Cantons of Bern and Geneva, Switzerland; NSC, Taiwan; TAEK, Turkey; STFC, the Royal Society and Leverhulme Trust, United Kingdom; DOE and NSF, United States of America. The crucial computing support from all WLCG partners is acknowledged gratefully, in particular from CERN and the ATLAS Tier-1 facilities at TRIUMF (Canada), NDGF (Denmark, Norway, Sweden), CC-IN2P3 (France), KIT/GridKA (Germany), INFN-CNAF (Italy), NLT1 (Netherlands), PIC (Spain), ASGC (Taiwan), RAL (UK), and BNL (USA) and in the Tier-2 facilities worldwide. APPENDIX: COMPREHENSIVE DATA PLOTS FROM TWO-PARTICLE CORRELATION Section V B presents the results only from selected centrality or pT intervals. One of the primary goals of this paper is the detailed study in a wide range of centrality and pT of the vn,n and vn , obtained from the 2PC in φ with a large pseudorapidity gap. For completeness, the vn coefficients for several fixed-pT correlations and for centrality intervals other than that shown in Fig. 13 are presented in Figs. 22–27. One important study of the factorization properties of v2 –v6 is the comparison between the EP method and mixed-pT 2PC method. This comparison is shown in Fig. 16 for the 0%–10% centrality interval. Figure 28 extends this comparison to several centrality intervals. Good consistency is observed for v2 –v6 in noncentral collisions. Figures 29–31 present the full set of the φ correlation functions with 2 < |η| < 5 for one central, one midcentral, and one peripheral centrality interval. These correlation functions are the inputs for the vn,n and vn spectra. They provide the complete picture of how the structures of the correlation functions are influenced, as a function of pTa and pTb , by the anisotropies related to the initial geometry and autocorrelations from the away-side jets, as well as how these influences map onto the strength and sign of the vn,n coefficients. Note that the correlation functions are symmetric with respect to pTa and pTb . For example, the correlation for 1 < pTa < 2 GeV and 2 < pTb < 3 GeV (denoted as [1–2, 2–3] GeV) is the same as that for 2 < pTa < 3 GeV and 1 < pTb < 2 GeV (denoted as [2–3, 1–2] GeV). Figures 32–36 shows the dependence of v2,2 –v6,6 on pTa and b pT . Together with Fig. 19, they provide a concise summary of vn,n spectra for all the 1D correlation functions such as those in Figs. 29–31 for 2 < |η| < 5. Because the sum of the v1,1 – v6,6 components almost exhaust the shape of the correlation functions, these data provide almost equivalent information as the correlation functions themselves over broad ranges of pTa and pTb . 014907-20 MEASUREMENT OF THE AZIMUTHAL ANISOTROPY FOR . . . 0.5