betaFeedback

WorksPhysics and Astronomy

Neural network enhanced Bayesian global analysis of relativistic heavy ion collisions

Abstract

We introduce a novel deep convolutional neural network (NN) -enhanced Bayesian global analysis of bulk observables in highest-energy heavy-ion collisions, using relativistic 2+1 D second-order viscous hydrodynamics with a dynamical freeze-out, and with perturbative QCD and saturation -based initial conditions from the event-by-event EKRT-model. Our analysis has 13+2 free parameters for the QCD-matter properties + initial state, which are constrained by the experimental data from sNN=200\sqrt{s_{NN}}=200 GeV Au+Au collisions at RHIC and 2.762.76 TeV Pb+Pb, 5.025.02 TeV Pb+Pb, and 5.445.44 TeV Xe+Xe collisions at the LHC. We replace the computationally demanding hydrodynamical simulations by NNs, which predict bulk observables directly from the initial energy density profiles, event-by-event, and account for the QCD-matter properties. With the NN output, we train the Gaussian process emulators for obtaining centrality-class averaged observables and their uncertainties. The NNs reduce the computing time significantly, enabling us to include also statistics-hungry flow observables like v4v_4 and the normalized symmetric cumulant NSC(4,2)NSC(4,2) in the analysis. In this paper, we demonstrate the feasibility of the NN based Bayesian global analysis. We find the data favoring a specific shear viscosity η/sη/s with a minimum-value plateau at temperatures 150≲T≲230150\lesssim T \lesssim 230 MeV, with 0.12≲(η/s)min≲0.180.12 \lesssim (η/s)_{\mathrm{min}} \lesssim 0.18. The bulk viscous coefficient ζ/sζ/s is non-zero at 200≲T≲300200\lesssim T \lesssim 300 MeV. The Knudsen number at the freeze-out is 0.8−2.30.8-2.3, while the ratio of the mean free path to the system size at freeze-out is in the range 0.3−1.20.3-1.2, implying that the freeze-out indeed happens at the expected limit of the applicability of hydrodynamics.

Introduction

According to Quantum Chromodynamics (QCD), the strongly interacting matter at high temperatures (T≳150T \gtrsim150 MeV) and vanishing baryochemical potential takes the form of the quark-gluon plasma (QGP), a fluid of deconfined quarks and gluons [1, 2, 3, 4]. The properties of hot QCD matter can be currently studied in ultra-relativistic heavy-ion collisions at the CERN Large Hadron Collider (LHC) and at the BNL Relativistic Heavy Ion Collider (RHIC), where such extreme temperatures are reachable.

A cornerstone in modeling QCD matter evolution is dissipative relativistic fluid dynamics, “hydrodynamics”, where the matter properties are embedded in the transport coefficients and QCD equation of state, which includes the cross-over transition from the QGP to a hadron gas. Besides the matter properties, an essential part of the hydrodynamical models are the initial conditions that can be either parametrized, as done e.g. in TrENTo [5], or computed from QCD-based initial state models such as color glass condensate based IP-Sat (Impact parameter dependent saturation) [6, 7, 8], the EKRT (Eskola-Kajantie-Ruuskanen-Tuominen) model [9, 10, 11, 12, 13, 14, 15], and EPOS (Energy conservation + Parallel scattering + factOrization + Saturation) [16, 17, 18, 19, 20, 21]. The hydrodynamical evolution can be further supplemented by additional modeling for pre-hydrodynamic evolution such as KøMPøST [22] and A{L}PACA [23], and for late hadronic evolution that can be modeled using hadron cascade simulations, such as UrQMD [24, 25] or SMASH [26]. Alternatively, one may describe the whole evolution using hydrodynamics, the approach taken in this work, but supplement it with dynamical freeze-out conditions, and chemical freeze-out [27]. Over the last several years hydrodynamics-based approaches have provided a consistent description of the measured bulk (low transverse momentum) observables, and consequently a very strong support for the formation of a hot collectively behaving QGP in these collisions [28].

The determination of the QCD matter properties, such as the temperature dependencies of the specific shear and bulk viscosities, η/s(T)\eta/s(T) and ζ/s(T)\zeta/s(T), from the LHC and RHIC data is a very active field of research currently. The effective model frameworks mentioned above, which are necessary for describing the full collision event, typically come altogether with a relatively large number of free parameters, O(10−20)\mathcal{O}(10-20), which must be fitted from the experimental data; the matter properties, together with the unknowns in the initial state dynamics and in the decoupling dynamics, are then expressed in terms of these parameters. Furthermore, many of the fit parameters are correlated in such a way that various different parameter combinations can produce similar effects on the final state observables. For this reason – to understand these correlations and to have controlled error estimates – a Bayesian global analysis has become a standard tool in the determination of QCD matter properties and their uncertainties [29, 30, 31, 32, 33, 34, 35, 36, 37, 38, 39, 40, 41, 42, 43, 44, 45, 47] [46]. For a recent review, see Ref. [48].

One of the major challenges in these analyses is that typically the final experimental results are averages of millions of collision events, and a meaningful data comparison with the models can require a similar amount of simulated collision events. This is combined with the fact that the parameters for the QCD matter properties, together with those for the decoupling conditions and the initial state, form a multidimensional parameter space. The Bayesian analysis then requires that we can compute experimentally measured observables fast and easily at an arbitrary point in the parameter space. By a direct model computation this is clearly an impossible task as it would require a new set of at least tens of thousands of model runs, or even significantly more for rare observables, for each new parameter combination. Even with tools like Gaussian process (GP) emulators [49, 50, 51] the task is still computationally very demanding, as it may still require millions of model runs to produce sufficiently comprehensive training data for these emulators.

In this proof-of-principle study, we introduce a novel framework that can speed-up this process potentially by several orders of magnitude. The main new feature in our approach is that instead of computing millions of hydrodynamical simulations, we use the hydrodynamical simulations to train deep convolutional neural networks (NN) that can predict the final experimental observables event-by-event (EbyE) directly from the initial energy density profile. We can then replace the time consuming hydrodynamical simulations by NNs, and use them to generate the training data for the Gaussian process emulators. While the training of the NNs still requires a considerable amount of real hydrodynamical simulations, it is orders of magnitude less than what would be needed for generating data directly to the training of GP emulators using hydrodynamics. The more simulated events are required to obtain meaningful results for a given observable, the more significant is the difference in the computational time. Besides our previous works [52, 53], the use of generative neural networks as a fast way to emulate EbyE hydrodynamics simulations has recently been studied in Refs. [54, 55, 56]. However, to our knowledge, this is the first time that any of these methods have been applied in the context of Bayesian global analysis.

Our main physics goal here is an improved determination of the QCD matter viscosities η/s(T)\eta/s(T) and ζ/s(T)\zeta/s(T), and an understanding how these properties are correlated with the EKRT initial state calculation and its uncertainties. Our NNs, introduced in [52], are now further developed to accept also the QCD matter parameters as input [53]. These new NNs enable a very fast EbyE computation of flow coefficients, average transverse momenta and multiplicities. With the centrality-class averaged results, we then train the GP emulators, which are employed in the Bayesian analysis.

For the EbyE training of the NNs, we employ 2+1 D relativistic shear and bulk viscous hydrodynamics from Ref. [27], with initial conditions computed from the perturbative QCD + saturation -based EbyE-EKRT (Eskola-Kajantie-Ruuskanen-Tuominen) model [13], and with the decoupling conditions determined dynamically as introduced in Ref. [27]. Instead of coupling our hydrodynamics to a hadron cascade afterburner, we run hydrodynamics in partial chemical equilibrium (PCE) until the final kinetic freeze-out. In doing so, we avoid further assumptions for the various unknown hadronic cross sections and unphysical discontinuities e.g. in the shear viscosity. Strong and electromagnetic hadron decays, however, we still account for after the freeze-out.

Our Bayesian inference set-up has 15 input parameters in total: 10 for the η/s(T)\eta/s(T) and the ζ/s(T)\zeta/s(T), 1 for the chemical decoupling temperature, 2 for the dynamical freeze-out conditions, and 2 for the initial state calculation. Exploring such a high-dimensional parameter space with full EbyE 2+1 D hydrodynamic simulations is computationally expensive. Even when utilizing statistical model emulators such as Gaussian processes, full EbyE simulations still need to be run for O(1000)\mathcal{O}(1000) different parameter combinations, and for each of these the EbyE statistics required by the observables needs to be generated. With the NNs trained in these O(1000)\mathcal{O}(1000) parameter points, we can, however, easily generate such a large number of events that inclusion of more statistics-hungry flow observables in the analysis, such as here the Fourier coefficient v4v_4, and the normalized symmetric cumulant NSC(4,2)NSC(4,2), becomes possible.

This paper is structured as follows: We describe the EbyE-EKRT + viscous hydrodynamics model in Sec. II and the neural network framework for predicting model output in Sec. III. We go through the statistical analysis procedure in Sec. IV and present the analysis results in Sec. V. A summary of our findings can be found in Sec. VI.

Model description

The EbyE-EKRT + viscous hydrodynamics model with dynamical freeze-out utilized in this study has been described in detail in Ref. [27]. Here we summarize the main features and parameters relevant to the current study.

EbyE-EKRT initial state

The initial energy density profiles of QCD matter are obtained from the EbyE-EKRT minijet saturation model [13], which has been developed from Refs. [9, 10, 11, 12]. The model is based on a collinearly factorized next-to-leading order (NLO) perturbative QCD calculation of transverse energy (ETE_T) of minijets that are produced with transverse momenta (pTp_T) above a cut-off scale p0p_0 and rapidities in a mid-rapidity interval Δy=1\Delta y = 1 [11, 13, 57, 58, 59]. Saturation of the minijet production, emerging from QCD non-linearities that become dominant at high parton densities [9, 60], is conjectured to suppress the primary parton production towards lower values of pTp_T. For the transverse density of ETE_T in an AAAA collision of an impact parameter b\mathbf{b}, the saturation limit is obtained from a local saturation condition [11],

dETd2r(TA(r1,σn),TA(r2,σn),p0,sNN,Δy,β)=Ksatπp03Δy,(1)\begin{aligned} \frac{dE_T}{d^2\mathbf{r}}\left(T_A(\mathbf{r}_1,\sigma_n),T_A(\mathbf{r}_2,\sigma_n),p_0,\sqrt{s_{NN}},\Delta y,\beta\right) \\ &= \frac{K_{\mathrm{sat}}}{\pi}p_0^3\Delta y, \tag*{(1)} \end{aligned}

where r\mathbf{r} is the transverse coordinate, r1,2=r±b/2\mathbf{r}_{1,2}=\mathbf{r}\pm\mathbf{b}/2, TAT_A’s are the nuclear thickness functions, σn\sigma_n is the width parameter of the nucleon density distribution, sNN\sqrt{s_{NN}} is the nucleon-nucleon center-of-mass system (cms) collision energy, β\beta is a parameter encoding the freedom in the infrared-collinear safe NLO definition of the ETE_T (see [11, 13] for the details), and KsatK_{\mathrm{sat}} is a parameter controlling the onset of saturation.

The nuclear collision geometry enters the calculation via the nuclear overlap density,

ρAA(b,r,σn)=TA(r1,σn)TA(r2,σn),(2)\rho_{AA}(\mathbf{b},\mathbf{r},\sigma_n)=T_A(\mathbf{r}_1,\sigma_n)T_A(\mathbf{r}_2,\sigma_n), \tag*{(2)}

and also via the TAT_A-dependent nuclear effects in the employed EPS09s nuclear parton distribution functions [61]. The TAT_A’s here fluctuate due to the EbyE fluctuating configurations of nucleons in the colliding nuclei. The nucleon positions ri1,2\mathbf{r}_{i1,2} around the centers of the nuclei are sampled from the Woods-Saxon densities, with thicknesses d=0.55d=0.55, 0.550.55, 0.540.54 fm and radii R=6.38R=6.38, 6.76.7, 5.495.49 fm for the Au, Pb, Xe nuclei, correspondingly, and for Xe with fixed deformation parameters β2=0.162\beta_2=0.162 and β4=−0.003\beta_4=-0.003 [62]. The TAT_A’s are obtained as sums of individual nucleon thickness functions, which are assumed to be Gaussians of a width σn\sigma_n,

TA(r1,σn)=∑i=1A12πσn2exp⁡(−∣r1−ri1∣22σn2),(3)T_A(\mathbf{r}_1,\sigma_n)=\sum_{i=1}^{A}\frac{1}{2\pi\sigma_n^2}\exp\left(-\frac{\lvert\mathbf{r}_1-\mathbf{r}_{i1}\rvert^2}{2\sigma_n^2}\right), \tag*{(3)}

and similarly for TA(r2,σn).T_A(\mathbf{r}_2,\sigma_n). We leave σn\sigma_n here as a free fit parameter to be determined from the experimental data.

The saturation scale that regulates the parton production at low pTp_T is obtained as a solution of Eq. (1),

psat=psat(ρAA(b,r;σn),sNN,Δy,β,Ksat).(4)p_{\mathrm{sat}}=p_{\mathrm{sat}}\left(\rho_{AA}(\mathbf{b},\mathbf{r};\sigma_n),\sqrt{s_{NN}},\Delta y,\beta,K_{\mathrm{sat}}\right). \tag*{(4)}

As noticed in [13], to a very good precision psatp_{\mathrm{sat}} does not depend individually on TA(r1)T_A(\mathbf{r}_1) and TA(r2)T_A(\mathbf{r}_2), but effectively only on the value of the overlap density ρAA\rho_{AA} (as the above list of arguments to psatp_{\mathrm{sat}} is to indicate). Thus, for fixed AA and sNN\sqrt{s_{NN}}, it has been possible to parametrize the ρAA\rho_{AA} dependence of psatp_{\mathrm{sat}} with KsatK_{\mathrm{sat}} and β\beta dependent coefficients [13, 63, 64]. To minimize the number of free parameters in our Bayesian inference study, we fix β=0.8\beta=0.8 as in our earlier works [12, 13, 27, 63, 64], but leave KsatK_{\mathrm{sat}} as another free initial-state parameter to be constrained by the data.

Assuming the formation time of the minijet system to be τsat=1/psat\tau_{\mathrm{sat}}=1/p_{\mathrm{sat}}, the EbyE-EKRT model gives the local energy density at the mid-rapidity unit, for each nucleon configuration, impact parameter and cms-energy, and fixed β\beta, as [13]

e(r,τsat(r))=Ksatπ[psat(ρAA(b,r;σn),Ksat)]4.(5)e(\mathbf{r},\tau_{\mathrm{sat}}(\mathbf{r}))=\frac{K_{\mathrm{sat}}}{\pi}\left[p_{\mathrm{sat}}\left(\rho_{AA}(\mathbf{b},\mathbf{r};\sigma_n),K_{\mathrm{sat}}\right)\right]^4. \tag*{(5)}

where now the dependence on the fit parameters σn\sigma_n and KsatK_{\mathrm{sat}} is indicated. Concretely, for a fast computation of the EbyE fluctuating energy densities for the present Bayesian inference study, we employ the parametrizations of psatp_{\mathrm{sat}} introduced in Refs. [13, 63, 64].

Following our earlier EbyE-EKRT studies, in the transverse coordinate region where psat(r)≥psatmin⁡=1p_{\mathrm{sat}}(\mathbf{r})\geq p_{\mathrm{sat}}^{\min}=1 GeV, we evolve the energy densities from τsat(r)\tau_{\mathrm{sat}}(\mathbf{r}) to τ0=1/psatmin⁡≈0.2\tau_0=1/p_{\mathrm{sat}}^{\min}\approx0.2 fm using 0+1 D Bjorken hydrodynamics. In the region where psat(r)<psatmin⁡p_{\mathrm{sat}}(\mathbf{r})<p_{\mathrm{sat}}^{\min}, we connect the energy density profile at the time τ0\tau_0 smoothly to a binary collision profile (∝ρAA\propto\rho_{AA}), see again [13] for the details. We then use the obtained full energy density profile e(r,τ0)e(\mathbf{r},\tau_0), assuming no initial transverse flow, as the initial condition for our 2+1 D viscous hydrodynamics described in the next section.

For triggering the nuclear collisions here, we require that between the colliding two randomly sampled nucleon configurations the transverse distance of at least one pair of nucleons is less than σNNin/π\sqrt{\sigma_{NN}^{\mathrm{in}}/\pi}, where σNNin\sigma_{NN}^{\mathrm{in}} is the inelastic nucleon-nucleon cross section. As in [27], we use here the values σNNin=42\sigma_{NN}^{\mathrm{in}}=42, 6464, 7070, 7272 mb for sNN=200\sqrt{s_{NN}}=200 GeV Au+Au, 2.76 TeV Pb+Pb, 5.02 TeV Pb+Pb, 5.44 TeV Xe+Xe collisions, correspondingly.

2+1 D viscous hydrodynamics

The fluid dynamical equations describe local conservation of energy and momentum, ∂μTμν=0\partial_\mu T^{\mu\nu}=0. The energy-momentum tensor can be decomposed in the Landau frame as

Tμν=euμuν−PΔμν+πμν,(6)T^{\mu\nu}=e u^\mu u^\nu-P\Delta^{\mu\nu}+\pi^{\mu\nu}, \tag*{(6)}

where uμu^\mu is the fluid 4-velocity, ee is the local energy density, PP is the isotropic pressure, and πμν\pi^{\mu\nu} is the shear-stress tensor. Here P=−13ΔμνTμνP=-\frac{1}{3}\Delta_{\mu\nu}T^{\mu\nu}, with Δμν=gμν−uμuν\Delta^{\mu\nu}=g^{\mu\nu}-u^\mu u^\nu. The bulk viscous pressure is defined as Π=P−P0\Pi=P-P_0, where P0P_0 is the equilibrium pressure provided by the equation of state (EoS), P0=P0(e)P_0=P_0(e), with conserved charges neglected. For the EoS we use the s95p parametrization [65] with a variable chemical freeze-out temperature TchemT_{\mathrm{chem}} [66].

The evolution of the dissipative quantities is given by second-order transient fluid dynamics, [67, 69] [68]

τΠddτΠ+Π=−ζθ−δΠΠΠθ+λΠππμνσμν,(7)\tau_\Pi\frac{d}{d\tau}\Pi+\Pi=-\zeta\theta-\delta_{\Pi\Pi}\Pi\theta+\lambda_{\Pi\pi}\pi^{\mu\nu}\sigma_{\mu\nu}, \tag*{(7)}
τπddτπ⟨μν⟩+πμν=2ησμν+2τπππ  α⟨μων⟩α−δπππμνθ−τπππ  α⟨μσν⟩α+φ7π  α⟨μπν⟩α+λπΠΠσμν,(8)\tau_{\pi}\frac{d}{d\tau}\pi^{\langle\mu\nu\rangle}+\pi^{\mu\nu}=2\eta\sigma^{\mu\nu}+2\tau_{\pi\pi}\pi^{\langle\mu}_{\ \ \alpha}\omega^{\nu\rangle\alpha} -\delta_{\pi\pi}\pi^{\mu\nu}\theta-\tau_{\pi\pi}\pi^{\langle\mu}_{\ \ \alpha}\sigma^{\nu\rangle\alpha} +\varphi_{7}\pi^{\langle\mu}_{\ \ \alpha}\pi^{\nu\rangle\alpha}+\lambda_{\pi\Pi}\Pi\sigma^{\mu\nu}, \tag*{(8)}

where η\eta is the shear viscosity, ζ\zeta is the bulk viscosity, τΠ=[15(13−cs2)2(e+P0)]−1\tau_{\Pi}=\left[15\left(\frac{1}{3}-c_{s}^{2}\right)^{2}(e+P_{0})\right]^{-1} is the bulk relaxation time with csc_{s} denoting the speed of sound, τπ=5ηe+P0\tau_{\pi}=\frac{5\eta}{e+P_{0}} is the shear relaxation time, θ=∇μuμ\theta=\nabla_{\mu}u^{\mu} is the expansion rate, σμν=∇⟨μuν⟩\sigma^{\mu\nu}=\nabla^{\langle\mu}u^{\nu\rangle} is the strain-rate tensor and ωμν=12(∇μuν−∇νuμ)\omega^{\mu\nu}=\frac{1}{2}(\nabla^{\mu}u^{\nu}-\nabla^{\nu}u^{\mu}) is the vorticity tensor. The angular brackets denote the symmetric and traceless projection that is orthogonal to the 4-velocity, and the spatial gradient is defined as ∇μ=Δμν∂ν\nabla^{\mu}=\Delta^{\mu\nu}\partial_{\nu}. The second order transport coefficients are obtained from the 14-moment approximation for massless gas [69, 70] [68] and bulk related coefficients from Ref. [71], δΠΠ=23τΠ\delta_{\Pi\Pi}=\frac{2}{3}\tau_{\Pi}, λΠπ=85(13−cs2)τΠ\lambda_{\Pi\pi}=\frac{8}{5}\left(\frac{1}{3}-c_{s}^{2}\right)\tau_{\Pi}, δππ=43τπ\delta_{\pi\pi}=\frac{4}{3}\tau_{\pi}, τππ=107τπ\tau_{\pi\pi}=\frac{10}{7}\tau_{\pi}, φ7=970P0\varphi_{7}=\frac{9}{70P_{0}}, λπΠ=65τπ\lambda_{\pi\Pi}=\frac{6}{5}\tau_{\pi}.

The temperature dependence of the specific shear viscosity is parametrized as [27]

(ηs)(T)={(ηs)min⁡+SHT((TTH)−PH−1),T<TH(ηs)min⁡,TH≤T≤TQ(ηs)min⁡+SQ(T−TQ),T>TQ(9)\left(\frac{\eta}{s}\right)(T)= \begin{cases} \left(\frac{\eta}{s}\right)_{\min}+S_{H}T\left(\left(\frac{T}{T_{H}}\right)^{-P_{H}}-1\right), & T<T_{H}\\ \left(\frac{\eta}{s}\right)_{\min}, & T_{H}\leq T\leq T_{Q}\\ \left(\frac{\eta}{s}\right)_{\min}+S_{Q}(T-T_{Q}), & T>T_{Q} \end{cases} \tag*{(9)}

where ss is the equilibrium entropy density, and TQ=TH+Wmin⁡T_{Q}=T_{H}+W_{\min}. The parametrization allows both wide and narrow constant-η/s\eta/s region near the QCD transition temperature, and also a rapid increase in η/s\eta/s towards the lower temperatures in the hadronic matter. The latter is particularly important in connection with the dynamical freeze-out, discussed below, that relates decoupling and the temperature dependence of the hadronic shear viscosity.

The temperature dependence of the specific bulk viscosity is parametrized as [27]

(ζs)(T)=(ζ/s)max⁡1+(T−Tmax⁡ζ/sw(T))2,(10)\left(\frac{\zeta}{s}\right)(T)=\frac{(\zeta/s)_{\max}}{1+\left(\frac{T-T_{\max}^{\zeta/s}}{w(T)}\right)^{2}}, \tag*{(10)}

where

w(T)=2(ζ/s)width1+exp⁡(aζ/s(T−Tmax⁡ζ/s)(ζ/s)width).(11)w(T)=\frac{2(\zeta/s)_{\mathrm{width}}}{1+\exp\left(\frac{a_{\zeta/s}(T-T_{\max}^{\zeta/s})}{(\zeta/s)_{\mathrm{width}}}\right)}. \tag*{(11)}

The parametrization gives a maximum of ζ/s\zeta/s at temperature Tmax⁡ζ/sT_{\max}^{\zeta/s}, and the width of the peak is controlled by w(T)w(T). The peak is allowed to be asymmetric (regulated by aζ/sa_{\zeta/s}) so that the most important part of the bulk viscous effects at temperatures below the QCD transition is given by the chemical freeze-out [72, 73, 74].

Thus, the η/s(T)\eta/s(T) and ζ/s(T)\zeta/s(T)-related 10 free parameters to be determined from the experimental data are (η/s)min⁡(\eta/s)_{\min}, THT_{H}, Wmin⁡W_{\min}, PHP_{H}, SHS_{H}, SQS_{Q}, and (ζ/s)max⁡(\zeta/s)_{\max}, Tmax⁡ζ/sT_{\max}^{\zeta/s}, (ζ/s)width(\zeta/s)_{\mathrm{width}}, and aζ/sa_{\zeta/s}.

Dynamical freeze-out conditions

In order to model the system size dependence of freeze-out, while still treating hadronic QCD-matter evolution using fluid dynamics, we employ a dynamical decoupling condition, where the decoupling surface is determined by imposing simultaneously two different conditions2. We take the microscopic scale to be given by the shear relaxation time τπ\tau_{\pi} and assume that the fluid decouples when the local Knudsen number exceeds a limit CKnC_{\mathrm{Kn}}. Thus, at decoupling

Kn=τπmax⁡(∣∇μe∇μe∣/e,θ)=CKn.(12)\mathrm{Kn}=\tau_{\pi}\max\left(\sqrt{\left|\nabla_{\mu}e\nabla^{\mu}e\right|/e},\theta\right)=C_{\mathrm{Kn}}. \tag*{(12)}

In addition, we require that the system decouples when the mean free path, assumed to be proportional to τπ\tau_{\pi}, becomes larger than the system size. This gives a non-local system size decoupling condition

γτπR=CR,(13)\frac{\gamma\tau_{\pi}}{R}=C_{R}, \tag*{(13)}

where the Lorentz-γ\gamma factor accounts for τπ\tau_{\pi} being in the fluid rest frame while the system size is determined in the center-of-momentum frame of the nuclear collision, and R=AKnπR=\sqrt{\frac{A_{\mathrm{Kn}}}{\pi}}, with AKnA_{\mathrm{Kn}} denoting the area in the transverse-coordinate plane where Kn<CKn\mathrm{Kn}<C_{\mathrm{Kn}}. In the cases where there are multiple separate areas of the fluid, the system size is calculated separately for each region. Both CKnC_{\mathrm{Kn}} and CRC_{R} are free parameters to be determined from the experimental data.

A clear advantage of treating also the hadronic QCD-matter evolution with fluid dynamics is that it is then possible to avoid artificial discontinuities in the temperature dependence of transport coefficients, and at the same time obtain constraints from the experimental data also for the transport coefficients below the QCD transition temperature.

Predicting hydrodynamic simulation output with neural networks

Some observables, such as higher flow harmonics v3v_{3}, v4v_{4}, ... and their correlations, require a lot of events to constrain statistical errors, which makes them computationally expensive. We use neural networks to obtain quick estimates for single-event model output for any given energy density profile and parameter combination. The performance gain is significant; we can run approximately two full simulations in one CPU-hour, while with neural networks we can produce ∼100\sim100 events per second using a GPU (NVIDIA V100).

BlockOutput sizeLayers
Convolution128×128×64128 \times128 \times647×77 \times7 conv, stride 2
Pooling64×64×6464 \times64 \times643×33 \times3 max pool, stride 2
Dense Block64×64×25664 \times64 \times2561×11 \times1 conv; 3×33 \times3 conv; ×6\times6
Transition Layer67×67×12867 \times67 \times1281×11 \times1 conv
Transition Layer32×32×12832 \times32 \times1282×22 \times2 average pooling, stride 2
Dense Block32×32×51232 \times32 \times5121×11 \times1 conv; 3×33 \times3 conv; ×12\times12
Transition Layer33×33×25633 \times33 \times2561×11 \times1 conv
Transition Layer16×16×25616 \times16 \times2562×22 \times2 average pooling, stride 2
Dense Block16×16×89616 \times16 \times8961×11 \times1 conv; 3×33 \times3 conv; ×20\times20
Transition Layer16×16×44816 \times16 \times4481×11 \times1 conv
Transition Layer8×8×4488 \times8 \times4482×22 \times2 average pooling, stride 2
Dense Block8×8×12168 \times8 \times12161×11 \times1 conv; 3×33 \times3 conv; ×24\times24
Pooling12168×88 \times8 global average pooling

Table 1. The structure of a neural network with initial energy density and model parameter inputs. The numbers in parentheses next to the input variable describe the dimensions of the input. The output of the initial energy density part of the network is connected to the model parameter input branch via concatenation.

LayerOutputs size
Input parametersη/s\eta/s (6)ζ/s\zeta/s (4)Decoupling (3)
Fully Connected128128128
Concatenation384
Fully Connected1024
Fully Connected512
Fully Connected256
Fully ConnectedNoutN_{\mathrm{out}}

Table 1 (PDF p. 5). The structure of a neural network with initial energy density and model parameter inputs. The numbers in parentheses next to the input variable describe the dimensions of the input. The output of the initial energy density part of the network is connected to the model parameter input branch via concatenation.

Following Refs. [53, 75], the neural network architecture consists of two main parts. The first part extracts the most relevant information from the initial energy density profile at midrapidity, which is discretized into a 256×256256 \times256 matrix in the transverse xyxy-plane with a resolution of 0.07 fm. This segment of the network is built from a series of convolutional layers connected via dense connections. This design, known as the Densely Connected Convolutional Network (DenseNet), was originally developed for computer vision tasks [76]. Dense connections preserve information from all preceding convolution operations within a block by concatenating the output channels of each layer with those of all previous layers in that block. Formally, for nn successive convolution layers, the output of one densely connected block is:

y=[x,C1(x),C2([x,C1(x)]),…,Cn([x,C1(x),C2([x,C1(x)]),…)](14)\mathbf{y} = [\mathbf{x}, C_1(\mathbf{x}), C_2([\mathbf{x}, C_1(\mathbf{x})]), \ldots, C_n([\mathbf{x}, C_1(\mathbf{x}), C_2([\mathbf{x}, C_1(\mathbf{x})]), \ldots)] \tag*{(14)}

where CiC_i denotes a convolution operation and x\mathbf{x} is the input to the block. Using a convolutional design allows the network to learn local features hierarchically, with smaller-scale patterns building up to recognize larger structures. Furthermore, dense connectivity enhances training stability and mitigates the vanishing gradient problem [76].

As in Refs. [52, 53], we use a DenseNet-BC variant with a growth rate k=32k = 32 and compression parameter θ=0.5\theta= 0.5, which reduces the number of feature maps by a factor of 2 after each densely connected block. The DenseNet part contains a total of four densely connected blocks. Before each block, the spatial size is reduced by a factor of two by performing average pooling, except for the first block, where 3×33 \times3 average pooling is used. Each convolution is followed by batch normalization and a Rectified Linear Unit (ReLU) activation function, F(x)=max⁡(0,x)F(x) = \max(0, x). The DenseNet section concludes with a global average pooling layer, after which the condensed information is passed forward.

The second part of the network processes the model parameters (excluding those related to the initial state, as their information is already encoded in the energy density profile). This segment consists of two fully connected layers. The output of the second segment is then concatenated with the output from the DenseNet, followed by two additional dense layers and a final output layer. A ReLU activation is applied after each fully connected layer. The detailed structure of the full neural network is shown in Table I.

Separate neural networks are trained to produce different types of pTp_T-integrated event-by-event observables. For this purpose, the observables are divided into the following classes for which the networks are trained:

  • Particle multiplicities: protons, kaons, pions, charged particles

  • Mean transverse momenta ⟨pT⟩\langle p_T \rangle: protons, kaons, pions, charged particles

  • Charged particle flow harmonics v2v_2, v3v_3, and v4v_4, each in a separate class.

Each class contains observables with multiple pTp_{\mathrm{T}} integration ranges matched to the kinematic cuts of various experiments. Furthermore, each observable is normalized such that the typical values are O(1)\mathcal{O}(1).

Network inputs are normalized based on the training data. The initial energy density profiles are scaled to have zero mean and unit standard deviation. Other inputs are scaled to the range [0,1][0, 1].

For training, we run full simulations for 40 events per training point for each investigated collision system. 36 of these events are used for the training, and the remaining 4 are used in validation.

The training points have been sampled using the following iterative procedure: First, we sample 950 points uniformly from the full input parameter space using Latin hypercube sampling and perform initial training of the neural networks. The training was done using the Adam optimizer [77] for a period of 120 epochs. After each epoch, the network performance was tested against the validation set, and the network was saved if it outperformed the previously best-performing network. A logarithmic root-mean-squared (RMS) error loss is used for multiplicities, while the standard RMS error is used for other observables. During the training, the training data is augmented by applying random rotations, flips, and shifts.

Next, we perform the global analysis as described in the following section. After this initial analysis, we sample an additional 100 training points from the obtained posterior distribution and add these to the training set. We then repeat the procedure by retraining the neural network with this enhanced training set by an additional 20 epochs to obtain more accurate estimates in the more relevant region of the parameter space.

The neural networks are used solely to increase event statistics for individual parameter points, while interpolation across the parameter space is handled by a Gaussian process emulator, which will be described in the next section. For each of the (950+100)(950 + 100) parameter sets, we generate 10510^5 events per collision system utilizing the trained NNs.

To validate the neural network’s performance for the event-averaged observables, we selected 10 parameter points from the second training set and performed 5,000 hydrodynamic simulations for 5.02 TeV Pb+Pb collisions at each point. A comparison with neural network predictions is shown in Figure 1. The accuracy is very good, with an average deviation of ∼1%\sim1\% of the observable value. This error is subleading relative to the error from parameter-space interpolation when the training data for the Gaussian process are sparsely distributed in the parameter space (see appendix A).

Figure 1. Validation of neural network predictions against full-simulation results for various observables in sNN=5.02\sqrt{s_{NN}} = 5.02 TeV Pb+Pb collisions. The predictions are compared with hydrodynamical simulation results for 10 parameter points across the centrality classes listed in Table III. The straight line is added as a reference to indicate where the two results match exactly. E\mathcal{E} is the root-mean-square (RMS) error of the observable, defined in Eq. (27).

Unfortunately, 5,000 collision events are not enough to do the same test for the normalized symmetric cumulants. Therefore, to validate the neural network performance in this case, we chose the training parameter point with the highest value of the likelihood function and ran 50,000 hydrodynamic simulations. The results for this test for NSC(4,2)NSC(4,2), NSC(3,2)NSC(3,2), and NSC(4,3)NSC(4,3) are shown in Figure 2. The accuracy of the neural network is of the order 1–10% for the NSC(4,2)NSC(4,2) in the 10–70% centrality range. In the most central collision systems and for other normalized symmetric cumulants, the relative errors can be slightly larger, but these data were not included in the analysis.

Figure 2. Validation of the neural network prediction for normalized symmetric cumulants against hydrodynamic simulations with 50,000 simulated events. Note that only NSC(4,2)NSC(4,2) is included in our Bayesian analysis.

The code package used for the neural network training and the event generation is available at [78].

Statistical Analysis

The goal of a Bayesian global analysis is to obtain the posterior probability distribution of the model parameters. According to Bayes’ theorem, the posterior probability PP for x0x_0 to be the “true” value of parameter xx, on the condition of the observed data yexpy_{\mathrm{exp}}, is proportional to the prior probability pp of x0x_0 and the likelihood of observation yexpy_{\mathrm{exp}} if x=x0x=x_0:

P(x=x0∣yexp)∝p(x=x0)⋅L(yexp∣x0).(15)P(x=x_0|y_{\mathrm{exp}}) \propto p(x=x_0)\cdot\mathcal{L}(y_{\mathrm{exp}}|x_0). \tag*{(15)}

In the case of a uniform prior probability distribution, determining the posterior amounts to determining the likelihood function.

Input parameters

Our EbyE-EKRT + (2+1) D viscous hydrodynamics with dynamical freeze-out model has altogether 15 tunable parameters, which were introduced in § 2. To focus the sampling in the physically most meaningful areas of the input parameter space, we do the following replacements when constructing our uniform prior:

  • SH→Δ(η/s)H≡(η/s)(T=80 MeV)−(η/s)min⁡S_H \to\Delta(\eta/s)_H \equiv(\eta/s)(T=80\ \mathrm{MeV})-(\eta/s)_{\min},

  • CKn→CKn/(η/s)min⁡C_{\mathrm{Kn}} \to C_{\mathrm{Kn}}/(\eta/s)_{\min},

  • CR→CR/(η/s)min⁡C_R \to C_R/(\eta/s)_{\min}.

The final 15 fit parameters in our Bayesian inference study are summarized in Table II, together with their assumed prior ranges. Furthermore, for the model emulation and Markov chain Monte Carlo, the prior ranges have been scaled into a unit hypercube.

ParameterRange
Initial state
KsatK_{\mathrm{sat}}0.2−1.00.2-1.0
σn\sigma_n0.35−0.60 fm0.35-0.60\ \mathrm{fm}
QCD matter
THT_H120−250 MeV120-250\ \mathrm{MeV}
(η/s)min⁡(\eta/s)_{\min}0.08−0.220.08-0.22
PHP_H2.0−10.02.0-10.0
Δ(η/s)H\Delta(\eta/s)_H0.1−1.50.1-1.5
SQS_Q0.0−3.0 GeV−10.0-3.0\ \mathrm{GeV}^{-1}
Wmin⁡W_{\min}0−300 MeV0-300\ \mathrm{MeV}
(ζ/s)max⁡(\zeta/s)_{\max}0.0−0.150.0-0.15
Tmax⁡ζ/sT_{\max}^{\zeta/s}180−350 MeV180-350\ \mathrm{MeV}
aC/sa_{C/s}−2.0−0.0-2.0-0.0
(ζ/s)width(\zeta/s)_{\mathrm{width}}10−100 MeV10-100\ \mathrm{MeV}
Decoupling
TchemT_{\mathrm{chem}}135−165 MeV135-165\ \mathrm{MeV}
CKn/(η/s)min⁡C_{\mathrm{Kn}}/(\eta/s)_{\min}2.0−15.02.0-15.0
CR/(η/s)min⁡C_R/(\eta/s)_{\min}0.5−7.50.5-7.5

Table II. TABLE II: Free parameters of the model and their prior ranges.

Observables

Simulations have been performed for four collision systems: Au+Au at sNN=200\sqrt{s_{NN}}=200 GeV, Pb+Pb at sNN=2.76\sqrt{s_{NN}}=2.76 TeV and 5.02 TeV, and Xe+Xe at sNN=5.44\sqrt{s_{NN}}=5.44 TeV. For each collision system, the following centrality classes have been included: 0–5 %, 5–10 %, 10–20 %, 20–30 %, 30–40 %, 40–50 %, 50–60 %, 60–70 %, and 70–80 %. For each centrality class, the following 15 observables have been determined:

  • Charged particle multiplicity dNch/dη∣∣η∣<0.5dN_{\mathrm{ch}}/d\eta|_{|\eta|<0.5}

  • Pion multiplicity dN(π+)/dy∣∣y∣<0.5dN(\pi^+)/dy|_{|y|<0.5}

  • Kaon multiplicity dN(K+)/dy∣∣y∣<0.5dN(K^+)/dy|_{|y|<0.5}

  • Proton multiplicity dN(p)/dy∣∣y∣<0.5dN(p)/dy|_{|y|<0.5}

  • Charged particle mean transverse momentum ⟨pT⟩(ch)\langle p_T\rangle(\mathrm{ch})

  • Pion mean transverse momentum ⟨pT⟩(π+)\langle p_T\rangle(\pi^+)

  • Kaon mean transverse momentum ⟨pT⟩(K+)\langle p_T\rangle(K^+)

  • Proton mean transverse momentum ⟨pT⟩(p)\langle p_T\rangle(p)

  • Charged particle v2{2}v_2\{2\} ... v4{2}v_4\{2\}, pTp_T integrated from 0.2 GeV to 3.0 GeV

  • Charged particle v2{2}v_2\{2\} ... v4{2}v_4\{2\}, pTp_T integrated from 0.2 GeV to 5.0 GeV

  • Normalized symmetric cumulant NSC(4,2)=⟨v42v22⟩⟨v42⟩⟨v22⟩−1NSC(4,2) = \frac{\langle v_4^2v_2^2\rangle}{\langle v_4^2\rangle\langle v_2^2\rangle} - 1

We provide a detailed list of the included experimental data in Table III.

ObservableAu+Au 200 GeV Centrality (%)Ref.Pb+Pb 2760 GeV Centrality (%)Ref.Pb+Pb 5020 GeV Centrality (%)Ref.Xe+Xe 5440 GeV Centrality (%)Ref.
dNch/dη∣∣η∣<0.5dN_{\mathrm{ch}}/d\eta\big|_{|\eta|<0.5}0 − 5, ..., 50 − 60[80]0 − 5, ..., 60 − 70[81]0 − 5, ..., 60 − 70[82]0 − 5, ..., 70 − 80[83]
dN(π+)/dy∣∣y∣<0.5dN(\pi^{+})/dy\big|_{|y|<0.5}0 − 5, ..., 70 − 80[84]0 − 5, ..., 70 − 80[85]0 − 5, ..., 70 − 80[86]--
dN(K+)/dy∣∣y∣<0.5dN(K^{+})/dy\big|_{|y|<0.5}0 − 5, ..., 70 − 80[84]0 − 5, ..., 70 − 80[85]0 − 5, ..., 70 − 80[86]--
dN(p)/dy∣∣y∣<0.5dN(p)/dy\big|_{|y|<0.5}0 − 5, ..., 70 − 80[84]0 − 5, ..., 70 − 80[85]0 − 5, ..., 70 − 80[86]--
⟨pT⟩(ch)\langle p_T\rangle(\mathrm{ch})------0 − 5, ..., 70 − 80[83]
⟨pT⟩(π+)\langle p_T\rangle(\pi^{+})0 − 5, ..., 70 − 80[84]0 − 5, ..., 70 − 80[85]0 − 5, ..., 70 − 80[86]--
⟨pT⟩(K+)\langle p_T\rangle(K^{+})0 − 5, ..., 70 − 80[84]0 − 5, ..., 70 − 80[85]0 − 5, ..., 70 − 80[86]--
⟨pT⟩(p)\langle p_T\rangle(p)0 − 5, ..., 70 − 80[84]0 − 5, ..., 70 − 80[85]0 − 5, ..., 70 − 80[86]--
v2{2}v_{2}\{2\}0 − 5, ..., 70 − 80[87]0 − 5, ..., 70 − 80[88]0 − 5, ..., 70 − 80[88]0 − 5, ..., 60 − 70[89]
v3{2}v_{3}\{2\}0 − 5, ..., 70 − 80[87]0 − 5, ..., 60 − 70[88]0 − 5, ..., 70 − 80[88]0 − 5, ..., 50 − 60[89]
v4{2}v_{4}\{2\}0 − 5, ..., 70 − 80[87]0 − 5, ..., 60 − 70[88]0 − 5, ..., 70 − 80[88]0 − 5, ..., 50 − 60[89]
NSC(4,2)\mathrm{NSC}(4,2)10 − 20, ..., 40 − 50[90]10 − 20, ..., 40 − 50[91]10 − 20, ..., 40 − 50[92]--

Table III. Summary of experimental data. For Au+Au, the proton target multiplicity is an average of protons and antiprotons and identified particle multiplicities at 10-20 % are averages of 10-15 % and 15-20 % centrality classes; for ⟨pT⟩\langle p_T\rangle, 10-20 % centrality class is not included. For Pb+Pb, in cases where vn{2}v_n\{2\} data have narrower centrality bins than the model output, an average over bins has been taken.

Principal component analysis

The dimension of the observable space can be reduced considerably via principal component analysis, where the goal is to find the linear combination of observables with the largest variance (so the largest sensitivity to the change in input). For the variance comparison to make sense, the observables are centered so that their mean is 0, and scaled with their sample standard deviation to make them dimensionless and of similar scale with regards to the variance.

Given a matrix Y^\hat{Y} of (centered and scaled) model data, where each row corresponds to one point in input parameter space, with the number of rows corresponding to the number of parameter sets that we will use for the training of the GPs (900 in the first analysis and 980 in the second, see Sec. IV D below) and each column corresponds to one of mm data points, we find an eigenvalue decomposition of the covariance matrix Y^TY^\hat{Y}^{T}\hat{Y}:

Y^TY^=VΛVT,(16)\hat{Y}^{T}\hat{Y} = V\Lambda V^{T}, \tag*{(16)}

where Λ\Lambda is the diagonal matrix containing the eigenvalues λ1,…,λm\lambda_1,\ldots,\lambda_m and VV is an orthogonal matrix containing the eigenvectors of the covariance matrix. Since λ1≥λ2≥…≥λm\lambda_1 \ge\lambda_2 \ge\ldots\ge\lambda_m, the fraction of the total variance explained by the principal components above some index kk will be negligible. This allows us to define a lower-rank approximation of the original transformed data matrix Z=Y^VZ = \hat{Y}V as Zk=Y^VkZ_k = \hat{Y}V_k, where VkV_k contains the first kk columns of VV. The transformation of a vector y⃗\vec{y} from the space of observables to a vector z⃗\vec{z} in the reduced-dimension principal component space with kk components is then defined as

z⃗=y⃗Vk.(17)\vec{z} = \vec{y}V_k. \tag*{(17)}
  • In Au+Au collisions, 89 data points are reduced to 15 principal components

  • In Pb+Pb collisions at sNN=2.76\sqrt{s_{NN}} = 2.76 TeV, 91 data points are reduced to 12 principal components

  • In Pb+Pb collisions at sNN=5.02\sqrt{s_{NN}} = 5.02 TeV, 93 data points are reduced to 12 principal components

  • In Xe+Xe collisions, 40 data points are reduced to 8 principal components

Gaussian process emulators

Given a list of model data (X,Zk)(X,Z_k) for a principal component z⃗(x⃗)∈Zk\vec{z}(\vec{x}) \in Z_k evaluated at parameter points x⃗∈X\vec{x} \in X, we obtain the Gaussian process predictive mean and uncertainty for zz at a point x⃗0\vec{x}_0 as follows:

zGP(x⃗0)=C0,ZCZ,Z−1Z,(18)z_{\mathrm{GP}}(\vec{x}_0) = C_{0,Z}C_{Z,Z}^{-1}Z, \tag*{(18)}
σGP2(x⃗0)=C0,0−C0,ZCZ,Z−1CZ,0(19)\sigma_{\mathrm{GP}}^2(\vec{x}_0) = C_{0,0} - C_{0,Z}C_{Z,Z}^{-1}C_{Z,0} \tag*{(19)}

where the covariance matrix has the form

C=(C0,0C0,ZCZ,0CZ,Z)(20)C = \begin{pmatrix} C_{0,0} & C_{0,Z} \\ C_{Z,0} & C_{Z,Z} \end{pmatrix} \tag*{(20)}

further defined by the covariance function also known as GP kernel kk: C0,0=k(x⃗0,x⃗0)C_{0,0}=k(\vec{x}_0,\vec{x}_0), CZ,Z=k(X,X)C_{Z,Z}=k(X,X). In this work we use the radial basis function with an additional white noise term for the kernel,

k(x⃗i,x⃗j)=Aexp⁡(−12∑k=115(xi,k−xj,k)2ℓk2)+δnoise.(21)k(\vec{x}_i,\vec{x}_j)=A\exp\left(-\frac{1}{2}\sum_{k=1}^{15}\frac{(x_{i,k}-x_{j,k})^2}{\ell_k^2}\right)+\delta_{\mathrm{noise}}. \tag*{(21)}

The white noise term δnoise\delta_{\mathrm{noise}} applies only on the diagonal elements of the covariance matrix, k(x⃗i,x⃗i)k(\vec{x}_i,\vec{x}_i), and accounts for the statistical and neural network uncertainties in the event averages. The kernel hyperparameters AA, ℓ⃗\vec{\ell}, δnoise\delta_{\mathrm{noise}} are initially tuned using 900 points from the neural network training dataset, while the remaining 50 were used for validation. On the second round, 80 of the additional 100 training points are used for hyperparameter tuning and remaining 20 are added to the validation set.

To compare the emulator predictions against physical observables, the prediction z⃗GP\vec{z}_{\mathrm{GP}} in the principal component space is transformed back to the space of observables:

y⃗GP=z⃗GPVkT.(22)\vec{y}_{\mathrm{GP}}=\vec{z}_{\mathrm{GP}}V_k^{T}. \tag*{(22)}

One must also remember to invert the scaling and centering of the model data which was performed in preparation for the principal component analysis.

We utilise the SCIKIT-LEARN [79] Python library for both the principal component analysis and the Gaussian process regression. A summary of the emulator quality checks can be found in Appendix A.

Likelihood function

For each collision energy, the likelihood function is defined as

L(z⃗exp∣x⃗)=1∣2πΣ∣e−dM2/2,(23)\mathcal{L}(\vec{z}_{\mathrm{exp}}|\vec{x})=\frac{1}{\sqrt{|2\pi\Sigma|}}e^{-d_M^2/2}, \tag*{(23)}

where dMd_M is the Mahalanobis distance

dM2≡(z⃗GP(x⃗)−z⃗exp)TΣ−1(z⃗GP(x⃗)−z⃗exp).(24)d_M^2\equiv(\vec{z}_{\mathrm{GP}}(\vec{x})-\vec{z}_{\mathrm{exp}})^T\Sigma^{-1}(\vec{z}_{\mathrm{GP}}(\vec{x})-\vec{z}_{\mathrm{exp}}). \tag*{(24)}

The covariance matrix Σ\Sigma contains three different sources of uncertainty: the GP prediction errors (19), the experimental uncertainties, and an additional theoretical uncertainty σth\sigma_{\mathrm{th}} reflecting the remaining modeling imperfections. Such a theoretical uncertainty was included also in Refs. [30] [46]; we, however, do not treat σth\sigma_{\mathrm{th}} as a free parameter, but assume it to be ∼O(10%)\sim\mathcal{O}(10\%) of the measured value for each observable.3 We investigate three possible error values, 10%, 20%, and 30%, to get a clearer view on its effect on the posterior distribution.4 Such a study is relevant since a theoretical model, like EbyE-EKRT here, can be too rigid in some of its predictions. If the model is not matching perfectly with data for all observables, the rigidity easily leads to an artificial bias in an automated global analysis. There is also some degree of uncertainty in the numerical simulations due to the discretization, which is now effectively accounted for by σth\sigma_{\mathrm{th}}.

The total likelihood at input parameter point x⃗\vec{x} is then the product of the likelihoods from all collision systems:

Ltotal=L200×L2760×L5020×L5440.(25)\mathcal{L}_{\mathrm{total}}=\mathcal{L}_{200}\times\mathcal{L}_{2760}\times\mathcal{L}_{5020}\times\mathcal{L}_{5440}. \tag*{(25)}

We use the Markov chain Monte Carlo code emcee [93] to draw samples from the likelihood function. We run an ensemble of 300 walkers for 40000 steps and discard the first 30000 steps to ensure that all samples used in results are drawn after MCMC has converged to the target distribution.

Results

Posterior distribution

We present the summary statistics of the posterior probability distributions in Table IV, i.e., the median values and 90% credible intervals for the marginal distributions of each parameter.

Parameter10%20%30%
KsatK_{\mathrm{sat}}0.57−0.09+0.100.57^{+0.10}_{-0.09}0.53−0.11+0.120.53^{+0.12}_{-0.11}0.51−0.13+0.160.51^{+0.16}_{-0.13}
σn\sigma_n [fm]0.38−0.02+0.030.38^{+0.03}_{-0.02}0.38−0.03+0.050.38^{+0.05}_{-0.03}0.40−0.04+0.070.40^{+0.07}_{-0.04}
THT_H [MeV]181−59+61181^{+61}_{-59}178−55+61178^{+61}_{-55}178−54+63178^{+63}_{-54}
(η/s)min⁡(\eta/s)_{\min}0.14−0.02+0.010.14^{+0.01}_{-0.02}0.15−0.02+0.020.15^{+0.02}_{-0.02}0.15−0.03+0.030.15^{+0.03}_{-0.03}
PHP_H8.4−3.9+1.58.4^{+1.5}_{-3.9}8.4−3.7+1.58.4^{+1.5}_{-3.7}8.3−3.8+1.58.3^{+1.5}_{-3.8}
Δ(η/s)H\Delta(\eta/s)_H0.3−0.2+0.30.3^{+0.3}_{-0.2}0.3−0.2+0.60.3^{+0.6}_{-0.2}0.4−0.3+0.80.4^{+0.8}_{-0.3}
SQS_Q [GeV−1^{-1}]1.2−1.1+1.61.2^{+1.6}_{-1.1}1.3−1.2+1.51.3^{+1.5}_{-1.2}1.4−1.2+1.51.4^{+1.5}_{-1.2}
Wmin⁡W_{\min} [MeV]225−130+68225^{+68}_{-130}183−151+104183^{+104}_{-151}165−144+120165^{+120}_{-144}
(ζ/s)max⁡(\zeta/s)_{\max}0.09−0.05+0.050.09^{+0.05}_{-0.05}0.09−0.06+0.050.09^{+0.05}_{-0.06}0.08−0.06+0.060.08^{+0.06}_{-0.06}
Tmax⁡ζ/sT_{\max}^{\zeta/s} [MeV]215−31+66215^{+66}_{-31}223−37+59223^{+59}_{-37}230−43+76230^{+76}_{-43}
aζ/sa_{\zeta/s}−0.5−1.1+0.4-0.5^{+0.4}_{-1.1}−0.7−1.1+0.7-0.7^{+0.7}_{-1.1}−0.8−1.0+0.7-0.8^{+0.7}_{-1.0}
(ζ/s)width(\zeta/s)_{\mathrm{width}} [MeV]57−43+3857^{+38}_{-43}60−43+3660^{+36}_{-43}61−43+3561^{+35}_{-43}
TchemT_{\mathrm{chem}} [MeV]147−4+3147^{+3}_{-4}148−5+5148^{+5}_{-5}149−6+7149^{+7}_{-6}
CKnC_{\mathrm{Kn}}1.2−0.4+0.61.2^{+0.6}_{-0.4}1.4−0.5+0.71.4^{+0.7}_{-0.5}1.5−0.6+0.81.5^{+0.8}_{-0.6}
CRC_R0.7−0.4+0.30.7^{+0.3}_{-0.4}0.7−0.4+0.30.7^{+0.3}_{-0.4}0.8−0.4+0.40.8^{+0.4}_{-0.4}

Table IV. Summary of the median values and 90% credible intervals from the marginal posterior distributions of the model parameters, for three different values of theoretical uncertainty.

For the initial state parameters, we find the saturation parameter has a median value Ksat≈0.5K_{\mathrm{sat}}\approx0.5 and the full 90% credible range considering all three theoretical uncertainties is 0.38<Ksat<0.670.38<K_{\mathrm{sat}}<0.67, while the nucleon width favors the lowest allowed values, 0.35 fm<σn<0.47 fm0.35\ \mathrm{fm}<\sigma_n<0.47\ \mathrm{fm}. These values for KsatK_{\mathrm{sat}} and σn\sigma_n are consistent with the earlier studies [13, 27]. Interestingly, the σn\sigma_n obtained here is quite close to the value 0.43±0.01 fm0.43\pm0.01\ \mathrm{fm}, extracted from the HERA data for exclusive electroproduction of J/ΨJ/\Psi [94].

The chemical freeze-out temperature TchemT_{\mathrm{chem}} is constrained within the range 143−156 MeV143-156\ \mathrm{MeV}, with the increase in theoretical uncertainty effecting mainly the upper bound. The current posterior distribution is extending to a slightly lower range than the previously obtained 150−158 MeV150-158\ \mathrm{MeV} for s95ps95p EoS from the global analysis presented in Ref. [31].

The Knudsen number for freeze-out is 0.8−2.30.8-2.3, while the ratio of mean free path to system size at freeze-out is in the range 0.3−1.20.3-1.2 with the median around CR≈0.7−0.8C_R\approx0.7-0.8. Thus, according to the experimental data, the freeze-out takes place, as expected, at the very limit of the applicability of hydrodynamics.

For the transport coefficients, rather than investigating each parameter individually, we illustrate their combined effect by plotting the actual temperature dependence of η/s\eta/s and ζ/s\zeta/s based on samples from the posterior distribution. Even if the individual parameters themselves are not very tightly constrained, they are not independent from each other, and the net effect is that in particular shear viscosity itself is much more tightly constrained than the constraints on the individual parameters would suggest. We also note that the final constraints for the transport properties of QCD matter are very robust against the variations in the assumed theoretical error, which is an encouraging result. The most significant difference is seen in bulk viscosity, with the smallest 10 % error there is a clear lower limit, but this diminishes with larger theoretical errors. The upper limit for bulk is much less sensitive. On the other hand, the limits for the shear viscosity are practically independent of the assumed error.

We see from Figure 3 that η/s\eta/s has a strong temperature dependence below T≈150 MeVT\approx150\ \mathrm{MeV} (parameter PHP_H leaning strongly towards the maximum of the prior range), while the minimum-value plateau is between 150−220 MeV150-220\ \mathrm{MeV} with 0.12<(η/s)min⁡<0.180.12<(\eta/s)_{\min}<0.18. For temperatures higher than 200 MeV, a rising trend is allowed but not favored; for all three cases the median remains close to constant all the way to 350 MeV. This is a clear difference from the previous global analysis with EKRT initial state [31] where the model with s95ps95p equation of state had η/s\eta/s increasing with temperature beyond T>250 MeVT>250\ \mathrm{MeV}. However, the two analyses are not directly comparable, as the previous analysis was done using averaged initial conditions and without bulk viscosity.

Figure 3. The median and 90% credible interval for the specific shear viscosity η/s\eta/s with respect to temperature TT.

Comparing our result for η/s\eta/s against other recent Bayesian analyses, we obtain quite a good agreement with the JETSCAPE collaboration (Ref. [33], figure 6) for the temperature range investigated in that study, T=150−350 MeVT=150-350\ \mathrm{MeV}. While the JETSCAPE parametrization for η/s(T)\eta/s(T) did not explicitly include a constant-value plateau, the slope at high temperatures was allowed to be either positive or negative and no strong preference for the sign of the slope was found; a later Bayesian study with the same parametrisation but wider prior and IP-Glasma initial state also found temperature-independent η/s\eta/s to be consistent with data [42]. Our result also fits within the credible intervals from the various Trajectum analyses presented in Ref. [37]; indeed, our 90% credible interval for the 10% theoretical uncertainty is fully consistent with the Trajectum results for η/s\eta/s. Meanwhile, the rising trend observed by Virta et al. in Ref. [46] cannot be fully accommodated within the uncertainties of the present analysis; this discrepancy likely stems from the more restricted parametrization of (η/s)(T)(\eta/s)(T) in Ref. [46] which does not allow a constant-value plateau or a decrease of η/s\eta/s with increasing temperature.

For the bulk viscosity in this study, we obtain much weaker constraints compared to η/s\eta/s; see Figure 4. We can say with 90% credibility that ζ/s\zeta/s is non-zero in the temperature range T=200−300T = 200 - 300 MeV, but the variation in the exact value remains considerable. This temperature range for nonzero ζ/s\zeta/s is slightly higher than the range 150−200150 - 200 MeV obtained in the analyses of JETSCAPE [33] and Virta et al. [46]. On the other hand, both the IP-Glasma study [42] and the Trajectum analysis with the total hadronic cross section σAA\sigma_{AA} included in the fit data and no additional weighting of observables [37] find a nonzero minimum bound for ζ/s\zeta/s in a temperature range similar to our result.

Figure 4. The median and 90% credible interval for the bulk viscous coefficient ζ/s\zeta/s with respect to temperature TT. Note that the lower limit of the prior is zero.

We note that bulk viscosity remains small at and below the chemical freeze-out temperature even if the prior range would allow larger values. This is on par with the discussion in Ref. [27], where the parametrization was constructed so that the main part of the bulk viscous effect in hadron gas is described by chemical freeze-out.

Comparison to data

In the following, for each of the three investigated scenarios, we sample 5000 parameter combinations from the posterior distribution and compare the distribution of emulator predictions for these points against the data.

The centrality dependence of charged particle multiplicities for the investigated collision systems are presented in Figure 5, along with the average transverse momentum in 5.44 TeV Xe+Xe collisions. The agreement with the multiplicity data is very satisfactory across the whole collision energy range, from 200 GeV Au+Au up to 5.44 TeV Xe+Xe collisions, with larger theoretical uncertainty merely increasing the variation in the predictions while the median distance from the data remains very similar. However, in the case of Xe+Xe in particular, we see that the centrality dependence comes off a bit too steep; the multiplicity falls slightly too strongly towards peripheral collisions. This overly strong centrality dependence applies also for the mean transverse momentum of charged particles; in addition, here the uncertainties tend to increase more towards higher values of ⟨pT⟩\langle p_T\rangle, and consequently the increase in theoretical uncertainty also increases the median value of ⟨pT⟩\langle p_T\rangle predictions.

Figure 5. Posterior distribution based emulator predictions for the centrality dependence of charged particle multiplicity for all four collision systems (top panel) and average transverse momentum in Xe+Xe collisions at sNN=5.44\sqrt{s_{NN}} = 5.44 TeV (bottom panel). The violin plots reflect the distribution of emulator predictions, with more points concentrated on the broader parts of the violin. The horizontal line within the violin indicates the mean value of predictions. PHENIX data is from Ref. [80] and ALICE data from Refs. [81] (2.76 TeV), [82] (5.02 TeV), and [83] (5.44 TeV).

Here, we also want to emphasize that none of the model parameters depend on the collision energy sNN\sqrt{s_{NN}} or nuclear mass number AA, and that the sNN\sqrt{s_{NN}}, AA, and centrality dependencies of multiplicity are mainly predictions of the EKRT model.

We show the data comparison of emulator predictions for pion, kaon and proton multiplicities in Au+Au and Pb+Pb collisions in Figure 6. We obtain good descriptions of pion and proton multiplicities in all cases; however, the kaon production is systematically overestimated. As was already seen in the case of charged particle multiplicity, the centrality dependence tends to be too steep at the LHC energies, suggesting perhaps a collision energy dependent nucleon width σn\sigma_n and/or possibly also that nuclear substructure (“hotspots”) might be required for a better fit.

Figure 6. Posterior distribution emulator predictions for identified particle multiplicities in 200 GeV Au+Au (top panel), 2.76 TeV Pb+Pb (middle panel), and 5.02 TeV Pb+Pb (bottom panel) collisions. PHENIX data from [84]. ALICE data from [85] (2.76 TeV) and [86] (5.02 TeV).

The data comparison for the transverse momenta of pions, kaons and protons is presented in Figure 7. Here the overall agreement with the data is very good for all three particle species, the main problem being the proton ⟨pT⟩\langle p_T \rangle at 200 GeV Au+Au collisions which is systematically overestimated. Similarly to the charged particle ⟨pT⟩\langle p_T \rangle in Xe+Xe collisions, it seems easier to increase ⟨pT⟩\langle p_T \rangle rather than decrease it within the posterior distribution, so the increase in theoretical uncertainty allows more variation towards higher ⟨pT⟩\langle p_T \rangle values.

Figure 7. Posterior distribution emulator predictions for identified particle transverse momenta in 200 GeV Au+Au (top panel), 2.76 TeV Pb+Pb (middle panel), and 5.02 TeV Pb+Pb (bottom panel) collisions. PHENIX data from [84]. ALICE data from [85] (2.76 TeV) and [86] (5.02 TeV).

The flow harmonics vnv_n for n=2,3,4n = 2, 3, 4 are shown in Figure 8. The overall agreement with the data over multiple collision systems and three different harmonics is quite good. However, the centrality dependence is again slightly off; there’s too little flow in the most central collisions, and too much v2v_2 produced in more peripheral collisions. The underestimation of flow in the most central collisions may necessitate implementing nuclear de-

Figure 8. Posterior distribution emulator predictions for charged particle flow at RHIC Au+Au (top panel), LHC Pb+Pb (middle panels) and Xe+Xe (bottom panel) collisions. STAR data from [87]. ALICE data from [88] (Pb+Pb) and [89] (Xe+Xe).

formation effects for Au+Au and Pb+Pb collisions, and further deformations in the case of Xe+Xe. Including nucleon substructure would likely improve the v3/v2v_3/v_2 ratio, as we have demonstrated in a recent study [15]. We note that in particular v3v_3 and v4v_4 in Au+Au collisions are in a significantly better agreement with the measurements than in our previous non-Bayesian analysis using the same model setup [27].

To complete the data comparison, we show the emulator predictions for the normalized symmetric cumulant NSC(4,2)NSC(4,2) in Figure 9. We get a good description of the data in the 10-20 % centrality class, but the rise towards more peripheral collisions is too weak compared to measurements.

Figure 9. Posterior distribution emulator predictions for normalized symmetric cumulant NSC(4, 2) in 200 GeV Au+Au (top panel), 2.76 TeV Pb+Pb (middle panel), and 5.02 TeV Pb+Pb (bottom panel) collisions. STAR data from [90]. ALICE data from [91] (2.76 TeV) and [92] (5.02 TeV).

Correlations

Correlations between parameters and observables

Correlations between the parameter-parameter pairs and parameter-observable pairs can serve as a useful guide when investigating the posterior distribution. Strong correlations between parameters can indicate either redundancies in the parametrization (multiple combinations are producing the same effect) or particularly strong constraints from the data, limiting the posterior parameter space and “forcing” a correlation. Correlations between parameters and observables provide further guidance where to look for these data constraints.

In Table V we list the notable Pearson correlations between the parameters within the posterior distribution,

ParameterTh. uncertainty
AB10%20%30%
KsatK_{\mathrm{sat}}σn\sigma_n-0.41-0.44-0.47
KsatK_{\mathrm{sat}}THT_H0.420.300.25
KsatK_{\mathrm{sat}}Wmin⁡W_{\min}0.310.400.35
σn\sigma_nTchemT_{\mathrm{chem}}0.460.350.33
THT_HPHP_H0.600.450.33
(η/s)min⁡(\eta/s)_{\min}CKnC_{\mathrm{Kn}}0.570.650.71
(ζ/s)max⁡(\zeta/s)_{\max}Tmax⁡ζ/sT_{\max}^{\zeta/s}0.350.380.32
Tmax⁡ζ/sT_{\max}^{\zeta/s}(ζ/s)width(\zeta/s)_{\mathrm{width}}0.650.580.47

Table V. Notable Pearson correlations (∣r∣>0.3|r| > 0.3 for at least two of the three scenarios of different theoretical uncertainty) between parameters in the posterior distribution.

rij=∑k(xi,k−⟨xi⟩)(xj,k−⟨xj⟩)∑k(xi,k−⟨xi⟩)2∑k(xj,k−⟨xj⟩)2(26)r_{ij} = \frac{\sum_k (x_{i,k}-\langle x_i\rangle)(x_{j,k}-\langle x_j\rangle)}{\sqrt{\sum_k (x_{i,k}-\langle x_i\rangle)^2\sum_k (x_{j,k}-\langle x_j\rangle)^2}} \tag*{(26)}

where “notable” means ∣r∣>0.3|r| > 0.3 for at least two of the three investigated scenarios of varying theoretical uncertainty. In addition, summary tables for all notable correlations between model parameters and final state observables for 20% theoretical uncertainty can be found in Appendix C.

There are three pairs of parameters with particularly strong correlations which appear regardless of the magnitude of the theoretical uncertainty. The initial state parameters KsatK_{\mathrm{sat}} and σn\sigma_n are anticorrelated. This emerges from the tension between multiplicity data in central and peripheral collisions, as illustrated in Figure 10 for the Pb+Pb collisions at 2.76 TeV. Central collisions favor the combination of large KsatK_{\mathrm{sat}} and small σn\sigma_n, while for the multiplicities in peripheral collisions, smaller values of KsatK_{\mathrm{sat}} combined with larger values of σn\sigma_n are preferred.

Figure 10. Relative deviation from the measured value of charged particle multiplicity N(ch)N(\mathrm{ch}) in Pb+Pb collisions at sNN=2.76\sqrt{s_{NN}}=2.76 TeV for different combinations of initial state parameters KsatK_{\mathrm{sat}} and σn\sigma_n, sampled from the posterior distribution with 20% theoretical uncertainty. Black contours indicate the 10%, 50% and 90% levels of point density, such that the highest 10% of point density is within the innermost curve. Top panel: 0-5 % centrality. Bottom panel: 60-70 % centrality.

Minimum value of the specific shear viscosity (η/s)min⁡(\eta/s)_{\min} correlates with the freeze-out Knudsen number CKnC_{\mathrm{Kn}}. Here the allowed combinations are limited by mean transverse momentum of protons, which imposes constrains to the upper limit of (η/s)min⁡(\eta/s)_{\min} in particular, and the centrality dependence of v3v_3; we demonstrate this in Figure 11 for 5.02 TeV Pb+Pb collisions.

Figure 11. Relative deviation from the measured value of charged particle v3{2}v_3\{2\} (top and middle panel) and proton ⟨pT⟩\langle p_T\rangle (bottom panel) in Pb+Pb collisions at sNN=5.02\sqrt{s_{NN}}=5.02 TeV for different combinations of (η/s)min⁡(\eta/s)_{\min} and CKnC_{\mathrm{Kn}}.

The temperature where the bulk viscosity reaches its peak Tmax⁡ζ/sT_{\max}^{\zeta/s} is correlated with the width of the peak (ζ/s)width(\zeta/s)_{\mathrm{width}}. This correlation does not stem from data constraints, as these parameters have hardly any notable correlations with any of the observables. Rather, the analysis is driven to maximise ζ/s\zeta/s at T≈200T \approx200 MeV which makes the combination of narrow peak located at high temperature unpreferable, see Figure 12.

Figure 12. Value of bulk viscosity coefficient ζ/s\zeta/s at T=200T = 200 MeV for different combinations of the peak temperature Tmax⁡T_{\max} and peak width (ζ/s)width(\zeta/s)_{\mathrm{width}} from the posterior distribution.

Further investigating the correlations between model parameters and observables, there is a particularly strong correlation between chemical freeze-out temperature TchemT_{\mathrm{chem}} and proton multiplicity N(p)N(p). We find notable tension in the N(p)N(p) data regarding the optimal value of TchemT_{\mathrm{chem}}, both in terms of collision energy and centrality, as shown in Figure 13. This tension sets strong constraints on TchemT_{\mathrm{chem}}, and also suggests that the chemical freeze-out temperature might depend on the collision energy.

Figure 13. The posterior distribution of TchemT_{\mathrm{chem}} (same in all panels). The color indicates the average relative deviation of proton multiplicity N(p)N(p) from the measured value in 200 GeV Au+Au (top panels) and 5.02 TeV Pb+Pb collisions (bottom panels) in 0-5 % and 50-60 % centrality classes for different values of TchemT_{\mathrm{chem}}. Average is taken over the posterior samples within each bin.

We already noted above that (η/s)min⁡(\eta/s)_{\min} is constrained by ⟨pT⟩(p)\langle p_T\rangle(p) and v3v_3, and it is not surprising that ⟨pT⟩(K)\langle p_T\rangle(K) and v4v_4 are also correlated with (η/s)min⁡(\eta/s)_{\min}. However, it is interesting to note that ⟨pT⟩(π)\langle p_T\rangle(\pi) and v2v_2 do not have notable correlations with (η/s)min⁡(\eta/s)_{\min}. Instead, both show sensitivity to the initial state parameters on one hand, and the conditions at freeze-out on the other; as a correlation to Δ(η/s)H\Delta(\eta/s)_{H} in the case of ⟨pT⟩(π)\langle p_{T} \rangle(\pi), and as a correlation to CKnC_{\mathrm{Kn}} in the case of v2v_{2}.

As v2v_{2} is correlated with σn\sigma_{n} and CKnC_{\mathrm{Kn}}, while v4v_{4} is correlated with (η/s)min⁡(\eta/s)_{\min}, the question arises if these correlations are reflected in NSC(4,2)NSC(4,2). This turns out to be true only partially, and the dependence on centrality and collision energy is quite different, as shown in Figure 14. For v2v_{2}, the σn\sigma_{n} correlation is strongest in more central collisions, while for NSC(4,2)NSC(4,2) it is notable in more peripheral collisions. For v4v_{4}, the correlation with (η/s)min⁡(\eta/s)_{\min} is strongest at higher collision energies, for NSC(4,2)NSC(4,2) 200 GeV has the strongest correlation. The consistently strong correlation between v2v_{2} and CKnC_{\mathrm{Kn}} is greatly diminished for NSC(4,2)NSC(4,2) and survives only for the 2.76 TeV Pb+Pb collisions at 10-20 % centrality.

Figure 14. Centrality dependence of Pearson correlations for v2{2}v_2\{2\} (top row), v4{2}v_4\{2\} (middle row), and NSC(4,2)NSC(4,2) (bottom row) with parameters σn\sigma_n, (η/s)min⁡(\eta/s)_{\min}, and CKnC_{Kn} (left, middle, and right column, respectively). Gray band indicates negligible correlation strength.

Correlations between observables and viscous coefficients at different temperatures

While there are some notable correlations between observables and parameters related to the extrema of the shear viscosity coefficient, we notice that many of the parameters controlling the temperature dependence of η/s\eta/s and ζ/s\zeta/s have no systematic correlations with observables. In the following, we investigate whether the observables are correlated with η/s\eta/s and ζ/s\zeta/s in any particular temperature regions. In addition, tabulated correlation data for a selection of temperatures can be found in Appendix § 4.4.

For the shear viscosity, the observables with notable correlations are the mean transverse momenta ⟨pT⟩\langle p_{T} \rangle of pions, kaons and protons and the flow observables vn{2}v_{n}\{2\} and NSC(4,2)NSC(4,2). We show the correlations between ⟨pT⟩\langle p_{T} \rangle and η/s\eta/s in the temperature range 80-400 MeV in Figure 15. Here the findings are consistent with what we already

Figure 15. Temperature dependence of Pearson correlations between mean transverse momenta and η/s\eta/s. Top three panels: Collision energy dependence of correlations at 0-5 % centrality. Bottom three panels: Centrality dependence of correlations at sNN=5.02\sqrt{s_{NN}} = 5.02 TeV collision energy.

saw in the correlations between model parameters and observables; pion ⟨pT⟩\langle p_T\rangle correlates with η/s\eta/s at low temperatures, as was suggested by the observed correlation between ⟨pT⟩(π)\langle p_T\rangle(\pi) and Δ(η/s)H\Delta(\eta/s)_H, while ⟨pT⟩(K)\langle p_T\rangle(K) and ⟨pT⟩(p)\langle p_T\rangle(p) correlate with η/s\eta/s in the plateau range ≈130\approx130–230230 MeV, as suggested by their correlations with (η/s)min⁡(\eta/s)_{\min}. These correlations have a very weak dependence on collision energy, and also the centrality dependence is notable only at the lowest temperatures, effecting mostly the correlation between ⟨pT⟩(π)\langle p_T\rangle(\pi) and η/s\eta/s, which is strongest at the most central collisions and becomes insignificant around 40-50 % centrality.

The flow observables, however, show more variable behavior; see Figure 16. While at RHIC the correlations tend to vanish beyond temperatures T>200T > 200 MeV, at LHC energies they persist up to 300 MeV. At these high temperatures, the correlation strengths for v3v_3 and v4v_4 seem to first increase with collision energy, but then abruptly decrease again when moving from 5.02 TeV Pb+Pb collisions to 5.44 TeV Xe+Xe collisions, which may hint to a system size dependence. We also observe similar weakening of the correlations at high temperatures when moving to more peripheral collisions, supporting the idea that the flow coefficients are not probing these temperatures in smaller systems. It also turns out that the correlator NSC(4,2)NSC(4,2) is not sensitive to η/s\eta/s at high temperatures either, having the strongest correlations with η/s\eta/s around T≈100T \approx100 MeV.

Figure 16. Temperature dependence of Pearson correlations between flow observables and η/s\eta/s. Top four panels: Collision energy dependence of correlations at 30-40% centrality. Bottom four panels: Centrality dependence of correlations at sNN=5.02\sqrt{s_{NN}} = 5.02 TeV collision energy.

Finally we move on to the bulk viscosity ζ/s\zeta/s whose temperature dependence remains poorly constrained. Indeed, it only shows notable correlations with mean transverse momenta, which we present in Figure 17. Here all three investigated particle species tell a remarkably similar story: the correlation is the strongest in the most peripheral collisions and peaks between 150 and 200 MeV, which is the temperature range where the lower limit of our 90% credible interval for ζ/s\zeta/s becomes nonzero. The correlation peak does move towards higher temperatures in more central collisions, but the overall strength of the correlation also weakens, suggesting that no observable in the dataset used in this analysis is sensitive to ζ/s\zeta/s at temperatures beyond T≳250T \gtrsim250 MeV.

Figure 17. Temperature dependence of Pearson correlations between mean transverse momenta and ζ/s\zeta/s. Top three panels: Collision energy dependence of correlations at 70–80 % centrality. Bottom three panels: Centrality dependence of correlations at sNN=5.02\sqrt{s_{NN}} = 5.02 TeV collision energy.

Summary and Conclusions

We have introduced a novel neural network–enhanced framework for performing a Bayesian global analysis to determine the QCD matter properties from the high-energy heavy-ion collision data. The main feature of the framework is the replacement of the computationally demanding hydrodynamical simulations with neural networks that can predict the experimental observables directly from the initial energy density profile event by event. The new feature of the networks compared to our earlier work [52] is that they now take also the matter properties as an input, so that it is no longer necessary to train separate networks for each viscosity parametrization and dynamical freeze-out conditions. This makes them very suitable for global analysis. The NN-generated simulation data are then applied to the training of Gaussian process emulators, which in turn are used to evaluate likelihoods in the Bayesian analysis. Even if the training of the networks still requires a considerable amount of full hydrodynamical simulations, the number of simulated events (here 40 events ×\times four collision systems ×\times 1050 parameter sets = O(105)\mathcal{O}(10^{5}) events in total) is orders of magnitude less than what would be needed in computing simulations directly for the training of GP emulators (105×4×1050=O(108)10^{5} \times4 \times1050 = \mathcal{O}(10^{8}) events in total).

Here we have demonstrated the feasibility of the framework by performing a Bayesian analysis on the parameters of the EbyE-EKRT + (2+1) D viscous hydrodynamics model with dynamical freeze-out conditions. We incorporate data from four collision systems, and thanks to the neural networks, are able to include statistically demanding observables such as quadrangular flow v4v_{4} and the normalized symmetric cumulant NSC(4,2)NSC(4,2). We obtain tightest constraints for the initial state saturation strength KsatK_{\mathrm{sat}} and the nucleon width σn\sigma_{n}, chemical freeze-out temperature TchemT_{\mathrm{chem}} and most importantly, as the QCD matter property, for the temperature dependence of η/s\eta/s. In particular, we find the data favoring a relatively narrow nucleon width 0.35<σn<0.470.35 < \sigma_{n} < 0.47 fm, chemical freeze-out in the range 143–156 MeV, and η/s\eta/s with minimum-value plateau extending from 150 MeV to at least 230 MeV, with 0.12<(η/s)min⁡<0.180.12 < (\eta/s)_{\min} < 0.18. Bulk viscous coefficient ζ/s\zeta/s is non-zero at T=200−300T = 200 - 300 MeV, but otherwise remains unconstrained. Overall, our findings for (η/s)(T)(\eta/s)(T) and (ζ/s)(T)(\zeta/s)(T) are consistent with the other recent Bayesian analyses [33, 37, 42] [46]. Our results are also robust against the variations in theoretical uncertainty between 10% and 30% of the experimental data values.

The uncertainties are also reflected in the correlations between parameters and observables, as the most constrained parameters also have most notable correlations with the observables included in the analysis. Investigating the correlations of observables with transport coefficients η/s\eta/s and ζ/s\zeta/s at different temperatures, we find most notable correlations at temperatures ≈150−250\approx150 - 250 MeV. For η/s\eta/s, the strongest constraints in this temperature range appear to come from the mean transverse momenta of kaons and protons, as well as the flow observables vn{2}v_n\{2\}, while the v2v_2-v4v_4 correlator NSC(4,2)NSC(4, 2) and mean transverse momentum of pions are more sensitive to η/s\eta/s at low temperatures. For ζ/s\zeta/s, the constraining observable is ⟨pT⟩\langle p_T \rangle, which shows a very similar correlation peak between 150 - 200 MeV for all three investigated particle species.

There remains some room for improvement both in the model and the analysis. For a better agreement with the data, implementing further nuclear deformation effects is likely required, as well as updating the initial state model to the MC-EKRT model [14] with hotspots [15]. To obtain further constraints in the future analyses, we aim to incorporate even more statistically demanding observables such as NSC(4,3)NSC(4, 3) or mixed harmonic cumulants, and consequently we may need to further improve the architecture of the neural networks and the quality of the training data, for example, utilizing active learning approach [34].

Acknowledgments

J.A. acknowledges support by the program Excellence Initiative–Research University of the University of Wrocław of the Ministry of Science and Education. The project is co-financed by the Polish National Agency for Academic Exchange, number of proposal: BNI/ULM/2024/1/00179/DEC/I. H.H. is supported by Vanderbilt University and by the U.S. Department of Energy, Office of Science, under Award Numbers DE-SC-0024347 and DE-SC-0024711, and in part by the National Science Foundation under Grant No. DMS-2406870. We acknowledge the financial support from the Research Council of Finland Project No. 330448 (K.J.E.). This research was funded as a part of the Center of Excellence in Quark Matter of the Research Council of Finland (Projects No. 346325 and 364192). This research is part of the European Research Council Project No. ERC-2018-ADG-835105 YoctoLHC. We acknowledge the computation resources from the Finnish IT Center for Science (CSC), project jyy2580, and from the Finnish Computing Competence Infrastructure (FCCCI), persistent identifier urn:nbn:fi:research-infras-2016072533.

Emulator validation

Following Ref. [95], we define the emulator prediction error as

E=⟨(prediction−truthtruth)2⟩,(27)\mathcal{E} = \sqrt{\left\langle\left(\frac{\mathrm{prediction}-\mathrm{truth}}{\mathrm{truth}}\right)^2 \right\rangle}, \tag*{(27)}

and the emulator error estimate accuracy as

H≡ln⁡(⟨(prediction−truthGP uncertainty)2⟩),(28)\mathcal{H} \equiv\ln\left(\sqrt{\left\langle\left(\frac{\mathrm{prediction}-\mathrm{truth}}{\mathrm{GP\ uncertainty}}\right)^2 \right\rangle}\right), \tag*{(28)}

where the angle brackets indicate the average over test points and the GP uncertainty σGP\sigma_{\mathrm{GP}} is defined in Eq. (19). We summarize the results of emulator validation for the 20 points sampled from the initial posterior distribution in Table VI. The quality of emulator predictions is good for all the investigated observables. The RMS errors E\mathcal{E} presented here are on average ≈40%\approx40\% smaller compared to the the initial analysis with 900 training points sampled from the prior. Notably the emulator error is systematically overestimated for this set; this happens because in the hyperparameter tuning one has to account also for the edges of the prior, where the quality of predictions degrade due to sparsity of training points.

Observable⟨E⟩200\langle\mathcal{E} \rangle_{200}⟨H⟩200\langle\mathcal{H} \rangle_{200}⟨E⟩2760\langle\mathcal{E} \rangle_{2760}⟨H⟩2760\langle\mathcal{H} \rangle_{2760}⟨E⟩5023\langle\mathcal{E} \rangle_{5023}⟨H⟩5023\langle\mathcal{H} \rangle_{5023}⟨E⟩5440\langle\mathcal{E} \rangle_{5440}⟨H⟩5440\langle\mathcal{H} \rangle_{5440}
N(ch)N(\mathrm{ch})0.018-0.2610.014-0.3060.014-0.2930.017-0.567
N(π)N(\pi)0.024-0.1550.016-0.2750.016-0.255--
N(K)N(K)0.0270.0010.017-0.1550.017-0.186--
N(p)N(p)0.035-0.1850.030-0.1140.027-0.200--
⟨pT⟩(ch)\langle p_T \rangle(\mathrm{ch})------0.008-0.327
⟨pT⟩(π)\langle p_T \rangle(\pi)0.014-0.2220.015-0.0540.016-0.057--
⟨pT⟩(K)\langle p_T \rangle(K)0.010-0.3610.009-0.4340.009-0.328--
⟨pT⟩(p)\langle p_T \rangle(p)0.013-0.1640.010-0.3410.012-0.145--
v2v_20.029-0.2060.017-0.0960.0170.0400.017-0.094
v3v_30.053-0.1060.025-0.0660.026-0.1730.022-0.141
v4v_40.084-0.1200.036-0.1390.040-0.0010.034-0.257
NSC(4,2)\mathrm{NSC}(4,2)0.088-0.2030.052-0.1080.053-0.014--

Table VI. Summary of centrality-averaged emulator quality parameters for a test set of 20 points sampled from the initial posterior distribution. E\mathcal{E} reflects the emulator prediction error relative to the true value, while H\mathcal{H} measures the accuracy of the uncertainty estimate reported by the Gaussian process. The negative and positive values of H\mathcal{H} indicate overestimation and underestimation of error, respectively.

Closure tests

Closure tests serve as a consistency check for the full analysis procedure. For the target data, we take the model output at the same 20 points which were used for the GP emulator quality tests, and check if the analysis is able to recover the parameter values used to produce the data. Here the relative uncertainties associated with the target data are taken to be the same as in the corresponding experimental data, while the theoretical uncertainty can be considered negligible when comparing to model data. We summarize the results in Table VII.

ParameterP(Truth in credible interval)P(\text{Truth in credible interval})90% C.I.
20% C.I.68% C.I.90% C.I.avg % of prior
KsatK_{\mathrm{sat}}40%75%100%16.6%
σn\sigma_n45%85%95%18.0%
THT_H10%40%65%69.4%
(η/s)min⁡(\eta/s)_{\min}25%85%100%12.9%
PHP_H10%40%70%27.7%
Δ(η/s)H\Delta(\eta/s)_H5%60%85%36.0%
SQS_Q15%65%90%77.6%
Wmin⁡W_{\min}25%55%75%63.4%
(ζ/s)max⁡(\zeta/s)_{\max}10%35%90%44.6%
Tmax⁡ζ/sT_{\max}^{\zeta/s}15%45%85%53.4%
aζ/sa_{\zeta/s}20%60%85%75.7%
(ζ/s)width(\zeta/s)_{\mathrm{width}}20%65%90%77.1%
TchemT_{\mathrm{chem}}15%75%100%17.0%
CKn/(η/s)min⁡C_{\mathrm{Kn}}/(\eta/s)_{\min}5%60%85%23.9%
CR/(η/s)min⁡C_R/(\eta/s)_{\min}15%55%85%44.9%

Table VII. Probabilities for finding the closure test truth value in a specified credible interval after 20 trials. The last column shows the average width of the 90% credible interval as a fraction of the prior range of parameter values.

We obtain the narrowest credible intervals (<25%< 25\% of the prior width on average) for KsatK_{\mathrm{sat}}, σn\sigma_n, (η/s)min⁡(\eta/s)_{\min}, TchemT_{\mathrm{chem}}, and CKn/(η/s)min⁡C K_n/(\eta/s)_{\min}. These intervals are particularly conservative for KsatK_{\mathrm{sat}} and σn\sigma_n, which have their true values within 20% credible interval in 40% of the trials. On the other extreme is the parameter THT_H which has a very poor inference accuracy; the marginal posterior distribution is both wide (the 90% credible interval is ≈70%\approx70\% of the prior width on average) and imprecise (true value is within 90% credible interval in only 65% of the cases).

Notable Pearson correlations between model parameters and observables

Tables VIII-XI summarise notable Pearson correlations (∣r∣>0.3|r| > 0.3 in at least one centrality class) between the model parameters and observables for the posterior distribution from the analysis with 20% theoretical uncertainty. Upper and lower values in a cell represent the highest and lowest correlations, respectively, among all the investigated centrality classes for each observable.

Notable Pearson correlations between observables and transport coefficients at various temperatures

Tables XII and XIII summarise notable Pearson correlations (∣r∣>0.3|r| > 0.3 in at least one centrality class) between model parameters and values of specific shear viscosity η/s\eta/s and bulk viscosity ζ/s\zeta/s, respectively, at various temperatures, for the posterior distribution from the analysis with 20% theoretical uncertainty.

Similarly, tables XIV and XV summarise notable Pearson correlations between observables and viscous coefficients at different temperatures. Upper and lower values in a cell represent the highest and lowest correlations, respectively, among all the investigated centrality classes for each observable.

KsatK_{\mathrm{sat}}σn\sigma_nTHT_H(η/s)min⁡(\eta/s)_{\min}PHP_HΔ(η/s)H\Delta(\eta/s)_HSQS_QWmin⁡W_{\min}(ζ/s)max⁡(\zeta/s)_{\max}Tmax⁡ζ/sT_{\max}^{\zeta/s}aζ/sa_{\zeta/s}(ζ/s)width(\zeta/s)_{\mathrm{width}}TchemT_{\mathrm{chem}}CKnC_{\mathrm{Kn}}CRC_R
N(ch)N(\mathrm{ch})0.66 0.410.07 -0.48- -- -- -- -- -- -- -- -- -- --0.33 -0.49- -- -
N(π)N(\pi)0.64 0.070.39 -0.49- -- -- -- -- -- -- -- -- -- --0.37 -0.55- -- -
N(K)N(K)0.67 0.060.29 -0.47- -- -- -- -- -- -- -- -- -- --0.19 -0.46- -- -
N(p)N(p)- -- -- -- -- -- -- -- -- -- -- -- -0.70 -0.42- -- -
⟨pT⟩(π)\langle p_T\rangle(\pi)0.52 0.150.43 -0.14- -- -- -0.43 0.12- -- -- -0.30 0.12- -- -0.38 -0.34- -- -
⟨pT⟩(K)\langle p_T\rangle(K)- -0.41 -0.09- -0.60 0.34- -- -- -- -- -0.32 -0.01- -- -0.40 -0.270.39 0.14- -
⟨pT⟩(p)\langle p_T\rangle(p)- -- -- -0.67 0.45- -- -- -- -- -- -- -- -0.32 -0.250.53 0.220.38 -0.18
v2v_20.37 -0.18- -- -- -- -0.03 -0.31- -- -- -- -- -- -0.33 -0.10-0.03 -0.400.57 -0.29
v3v_3- -- -- -0.56 -0.33- -- -- -- -- -- -- -- -0.48 -0.210.39 -0.07- -
v4v_4- -- -- -0.65 -0.29- -- -- -0.01 -0.31- -- -- -- -0.55 -0.200.48 0.060.31 -0.31
NSC(4,2)NSC(4,2)0.30 -0.02-0.02 -0.35- -0.49 -0.09- --0.07 -0.42- -- -- -- -- -- -0.01 -0.31- -- -

Table VIII. Notable Pearson correlations (∣r∣>0.3|r| > 0.3) in at least one centrality class) between the model parameters and observables at 200 GeV Au+Au collisions for the posterior distribution from the analysis with 20% theoretical uncertainty. Upper and lower values in a cell represent the highest and lowest correlations, respectively, among all the investigated centrality classes for each observable.

Pb+Pb 2.76 TeVKsatK_{\mathrm{sat}}σn\sigma_nTHT_H(η/s)min⁡(\eta/s)_{\min}PHP_HΔ(η/s)H\Delta(\eta/s)_HSQS_QWmin⁡W_{\min}(ζ/s)max⁡(\zeta/s)_{\max}Tmax⁡ζ/sT_{\max}^{\zeta/s}aζ/sa_{\zeta/s}(ζ/s)width(\zeta/s)_{\mathrm{width}}TchemT_{\mathrm{chem}}CKnC_{\mathrm{Kn}}CRC_R
N(ch)N(\mathrm{ch})0.63 -0.030.55 -0.42- -- -- -- -- -- -- -- -- -- --0.20 -0.32- -- -
N(π)N(\pi)0.62 -0.160.62 -0.46- -- -- -- -- -- -- -- -- -- --0.21 -0.42- -- -
N(K)N(K)0.62 -0.180.59 -0.42- -- -- -- -- -- -- -- -- -- -- -- -- -
N(p)N(p)- -0.32 0.07- -- -- -- -- -- -0.09 -0.36- -- -- -0.82 -0.03- -- -
⟨pT⟩(π)\langle p_T\rangle(\pi)0.41 0.170.43 -0.05- -- -- -0.44 0.12- -- -- -- -- -- -0.40 -0.34- -- -
⟨pT⟩(K)\langle p_T\rangle(K)- -0.38 0.01- -0.56 0.39- -- -- -- --0.08 -0.31- -- -- -0.39 -0.260.33 0.17- -
⟨pT⟩(p)\langle p_T\rangle(p)- -- -- -0.66 0.53- -0.09 -0.32- -- --0.14 -0.31- -- -- -- -0.47 0.29- -
v2v_20.64 -0.230.30 -0.44- -- -- -- -- -- -- -- -- -- -- --0.06 -0.44- -
v3v_30.52 -0.18- -- --0.04 -0.47- -- -- -- -- -- -- -- -- --0.08 -0.39- -
v4v_40.33 -0.12- -- -0.06 -0.45- -- -- -- -- -- -- -- -- -- -- -
NSC(4,2)\mathrm{NSC}(4,2)- --0.21 -0.47- -0.40 -0.09- --0.09 -0.45- -- -- -- -- -- --0.22 -0.320.21 -0.46- -

Table IX. Same as table VIII but for 2.76 TeV Pb+Pb collisions.

Pb+Pb 5.02 TeVKsatK_{\mathrm{sat}}σn\sigma_nTHT_H(η/s)min⁡(\eta/s)_{\min}PHP_HΔ(η/s)H\Delta(\eta/s)_HSQS_QWmin⁡W_{\min}(ζ/s)max⁡(\zeta/s)_{\max}Tmax⁡ζ/sT_{\max}^{\zeta/s}aζ/sa_{\zeta/s}(ζ/s)width(\zeta/s)_{\mathrm{width}}TchemT_{\mathrm{chem}}CKnC_{\mathrm{Kn}}CRC_R
N(ch)N(\mathrm{ch})0.59 -0.050.60 -0.41- -- -- -- -- -- -- -- -- -- -- -- -- -
N(π)N(\pi)0.58 -0.170.66 -0.45- -- -- -- -- -- -- -- -- -- --0.18 -0.40- -- -
N(K)N(K)0.58 -0.200.64 -0.40- -- -- -- -- -- -- -- -- -- -- -- -- -
N(p)N(p)0.13 -0.340.45 0.08- -- -- -- -- -- -0.07 -0.33- -- -- -0.82 0.10- -- -
⟨pT⟩(π)\langle p_T\rangle(\pi)0.39 0.220.41 -0.04- -- -- -0.46 0.12- -- -- -0.31 -0.01- -- -0.37 -0.31- -- -
⟨pT⟩(K)\langle p_T\rangle(K)- -0.36 -0.02- -0.56 0.40- -- -- -- --0.07 -0.370.30 -0.10- -- -0.39 -0.300.33 0.16- -
⟨pT⟩(p)\langle p_T\rangle(p)- -- -- -0.66 0.53- -0.12 -0.32- -- --0.09 -0.35- -- -- -- -0.56 0.33- -
v2v_20.63 -0.120.45 -0.35- --0.15 -0.33- --0.20 -0.34- -0.30 0.19- -- -- -- -- --0.12 -0.56- -
v3v_30.48 -0.270.27 -0.32- -0.01 -0.56- -- -- -- -- -- -- -- -- --0.11 -0.38- -
v4v_40.31 -0.190.03 -0.34- -0.30 -0.40- -- -- -- -- -- -- -- -- -0.37 -0.22- -
NSC(4,2)\mathrm{NSC}(4,2)- --0.26 -0.48- -- -- --0.29 -0.48- -- -- -- -- -- -- -- -- -

Table X. Same as table VIII but for 5.02 TeV Pb+Pb collisions.

Xe+Xe 5.44 TeVKsatK_{\mathrm{sat}}σn\sigma_nTHT_H(η/s)min⁡(\eta/s)_{\min}PHP_HΔ(η/s)H\Delta(\eta/s)_HSQS_QWmin⁡W_{\min}(ζ/s)max⁡(\zeta/s)_{\max}Tmax⁡ζ/sT_{\max}^{\zeta/s}aζ/sa_{\zeta/s}(ζ/s)width(\zeta/s)_{\mathrm{width}}TchemT_{\mathrm{chem}}CKnC_{\mathrm{Kn}}CRC_R
N(ch)N(\mathrm{ch})0.60 -0.260.74 -0.43- -- -- -- -- -- -- -- -- -- -- -- -- -
⟨pT⟩(ch)\langle p_T\rangle(\mathrm{ch})0.33 0.090.43 0.10- -- -- -- -- -- --0.09 -0.31- -- -- -0.49 0.01- -- -
v2v_20.67 0.120.24 -0.43- --0.11 -0.470.30 0.03-0.17 -0.36- -- -- -- -- -- -- --0.43 -0.61- -
v3v_30.52 0.060.06 -0.31- --0.29 -0.58- -- -- -- -- -- -- -- -- --0.27 -0.41- -
v4v_4- -- -- --0.34 -0.51- -- -- -- -- -- -- -- -- -- -- -

Table XI. Same as table VIII but for 5.44 TeV Xe+Xe collisions.

(η/s)(T)(\eta/s)(T) at T=XT = X MeV
Parameter100150200250300
KsatK_{\mathrm{sat}}----0.32-0.46
σn\sigma_n-----
THT_H-----
(η/s)min⁡(\eta/s)_{\min}0.310.970.93--
PHP_H-----
Δ(η/s)H\Delta(\eta/s)_H0.81--0.32-
SQS_Q-----
Wmin⁡W_{\min}----0.38-0.60
(ζ/s)max⁡(\zeta/s)_{\max}---0.30--
Tmax⁡ζ/sT_{\max}^{\zeta/s}-----
aζ/sa_{\zeta/s}-----
(ζ/s)width(\zeta/s)_{\mathrm{width}}-----
TchemT_{\mathrm{chem}}-----
CKnC_{\mathrm{Kn}}0.510.660.660.41-
CRC_R0.31----

Table XII. Notable Pearson correlations between η/s\eta/s at various temperatures and parameters for the posterior distribution from the analysis with 20% theoretical uncertainty.

(ζ/s)(T)(\zeta/s)(T) at T=XT = X MeV
Parameter100150200250300
KsatK_{\mathrm{sat}}-----
σn\sigma_n--0.37-0.32--
THT_H-----
(η/s)min⁡(\eta/s)_{\min}---0.30--
PHP_H-----
Δ(η/s)H\Delta(\eta/s)_H-----
SQS_Q-----
Wmin⁡W_{\min}-----
(ζ/s)max⁡(\zeta/s)_{\max}--0.530.690.56
Tmax⁡ζ/sT_{\max}^{\zeta/s}0.46--0.350.660.84
aζ/sa_{\zeta/s}0.630.44---
(ζ/s)width(\zeta/s)_{\mathrm{width}}0.510.56-0.470.340.58
TchemT_{\mathrm{chem}}-0.37-0.50---
CKnC_{\mathrm{Kn}}-----
CRC_R-----

Table XIII. Notable Pearson correlations between ζ/s\zeta/s at various temperatures and parameters for the posterior distribution from the analysis with 20% theoretical uncertainty.

Au+Au 200 GeV; (η/s)(T)(\eta/s)(T) at T=XT = X MeVPb+Pb 2.76 TeV; (η/s)(T)(\eta/s)(T) at T=XT = X MeVPb+Pb 5.02 TeV; (η/s)(T)(\eta/s)(T) at T=XT = X MeVXe+Xe 5.44 TeV; (η/s)(T)(\eta/s)(T) at T=XT = X MeV
100150200250300100150200250300100150200250300100150200250300
N(ch)N(\mathrm{ch})-------------------
-------------------
N(π)N(\pi)-------------------
-------------------
N(K)N(K)-------------------
-------------------
N(p)N(p)-------------------
-------------------
⟨pT⟩(ch)\langle p_T\rangle(\mathrm{ch})-----------------0.30-
-----------------0.25-
⟨pT⟩(π)\langle p_T\rangle(\pi)0.35----0.33----0.37---------
0.15----0.14----0.09---------
⟨pT⟩(K)\langle p_T\rangle(K)-0.600.59---0.560.56---0.550.550.31------
-0.330.35---0.380.39---0.380.410.18------
⟨pT⟩(p)\langle p_T\rangle(p)-0.680.65---0.650.640.31--0.650.660.38------
-0.450.46---0.520.530.22--0.530.490.17------
v2v_2-0.02-----0.16--0.06-0.12-0.09-0.31-0.15-0.23-0.23-0.19-0.33-0.11-0.23-0.46-0.27
-0.47-----0.44--0.32-0.43-0.38-0.52-0.34-0.42-0.53-0.44-0.50-0.47-0.54-0.60-0.48
v3v_3-0.530.560.32---0.05-0.13-0.190.01-0.03-0.09-0.120.13--0.27-0.40-0.43-0.17
--0.33-0.29-0.04---0.46-0.48-0.48-0.41--0.58-0.55-0.57-0.46--0.56-0.64-0.62-0.43
v4v_40.330.650.630.33--0.050.070.07-0.260.260.320.220.18--0.32-0.39-0.08-
-0.25-0.31-0.24-0.03---0.44-0.49-0.39--0.32-0.40-0.47-0.47-0.35--0.49-0.53-0.39-
NSC(4,2)\mathrm{NSC}(4,2)-0.040.440.44---0.040.380.36---0.30---------
-0.30-0.10-0.14---0.52-0.14-0.08---0.47---------

Table XIV. Notable Pearson correlations (∣r∣>0.3|r| > 0.3 in at least one centrality class) between specific shear viscosity η/s\eta/s at various temperatures and observables, for the posterior distribution from the analysis with 20% theoretical uncertainty. Upper and lower values in a cell represent the highest and lowest correlations, respectively, among all the investigated centrality classes for each observable.

Au+Au 200 GeVAu+Au 200 GeVAu+Au 200 GeVAu+Au 200 GeVAu+Au 200 GeVPb+Pb 2.76 TeVPb+Pb 2.76 TeVPb+Pb 2.76 TeVPb+Pb 2.76 TeVPb+Pb 2.76 TeVPb+Pb 5.02 TeVPb+Pb 5.02 TeVPb+Pb 5.02 TeVPb+Pb 5.02 TeVPb+Pb 5.02 TeVXe+Xe 5.44 TeVXe+Xe 5.44 TeVXe+Xe 5.44 TeVXe+Xe 5.44 TeVXe+Xe 5.44 TeV
(ζ/s)(T)(\zeta/s)(T) at T=XT=X MeV(ζ/s)(T)(\zeta/s)(T) at T=XT=X MeV(ζ/s)(T)(\zeta/s)(T) at T=XT=X MeV(ζ/s)(T)(\zeta/s)(T) at T=XT=X MeV(ζ/s)(T)(\zeta/s)(T) at T=XT=X MeV(ζ/s)(T)(\zeta/s)(T) at T=XT=X MeV(ζ/s)(T)(\zeta/s)(T) at T=XT=X MeV(ζ/s)(T)(\zeta/s)(T) at T=XT=X MeV(ζ/s)(T)(\zeta/s)(T) at T=XT=X MeV(ζ/s)(T)(\zeta/s)(T) at T=XT=X MeV(ζ/s)(T)(\zeta/s)(T) at T=XT=X MeV(ζ/s)(T)(\zeta/s)(T) at T=XT=X MeV(ζ/s)(T)(\zeta/s)(T) at T=XT=X MeV(ζ/s)(T)(\zeta/s)(T) at T=XT=X MeV(ζ/s)(T)(\zeta/s)(T) at T=XT=X MeV(ζ/s)(T)(\zeta/s)(T) at T=XT=X MeV(ζ/s)(T)(\zeta/s)(T) at T=XT=X MeV(ζ/s)(T)(\zeta/s)(T) at T=XT=X MeV(ζ/s)(T)(\zeta/s)(T) at T=XT=X MeV(ζ/s)(T)(\zeta/s)(T) at T=XT=X MeV
100150200250300100150200250300100150200250300100150200250300
N(ch)N(\mathrm{ch})- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -
N(π)N(\pi)- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -
N(K)N(K)- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -
N(p)N(p)0.31 -0.190.51 -0.24- -- -- -0.32 -0.220.41 -0.35- -0.09 -0.30- -0.30 -0.32- -- -- -- -- -- -- -- -- -
⟨pT⟩(ch)\langle p_T\rangle(\mathrm{ch})- -- -- -- -- -- -- -- -- -- -- -- -- -- -- --0.12 -0.32-0.24 -0.58-0.41 -0.51- -- -
⟨pT⟩(π)\langle p_T\rangle(\pi)- -- -0.00-0.16 -0.54- -0.44- -0.07 -0.340.03 -0.58-0.23 -0.41- -- --0.02 -0.53-0.22 -0.46- -- -- -- -- -- -- -
⟨pT⟩(K)\langle p_T\rangle(K)- --0.06 -0.53-0.23 -0.46- -- -- --0.05 -0.53-0.29 -0.47- -- --0.01 -0.51-0.28 -0.470.01 -0.35- -- -- -- -- -- -
⟨pT⟩(p)\langle p_T\rangle(p)- -0.03 -0.46-0.14 -0.44- -- -- -0.05 -0.39-0.20 -0.43- -- --0.10 -0.41-0.18 -0.39-0.02 -0.33- -- -- -- -- -- -
v2v_2- -- -0.09 -0.31- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -
v3v_3- --0.01 -0.34- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -
v4v_4- -0.08 -0.420.22 -0.32- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -
NSC(4,2)NSC(4,2)- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -

Table XV. Like table XIV but for ζ/s\zeta/s instead of η/s\eta/s.

References

  1. [1]A. Bazavov et al. [HotQCD], Phys. Rev. D 90, 094503 (2014) doi:10.1103/PhysRevD.90.094503 [arXiv:1407.6387 [hep-lat]].DOI
  2. [2]A. Bazavov, P. Petreczky and J. H. Weber, Phys. Rev. D 97, no.1, 014510 (2018) doi:10.1103/PhysRevD.97.014510 [arXiv:1710.05024 [hep-lat]].DOI
  3. [3]S. Borsanyi, Z. Fodor, C. Hoelbling, S. D. Katz, S. Krieg and K. K. Szabo, Phys. Lett. B 730, 99-104 (2014) doi:10.1016/j.physletb.2014.01.007 [arXiv:1309.5258 [hep-lat]].DOI
  4. [4]S. Borsanyi, G. Endrodi, Z. Fodor, A. Jakovac, S. D. Katz, S. Krieg, C. Ratti and K. K. Szabo, JHEP 11, 077 (2010) doi:10.1007/JHEP11(2010)077 [arXiv:1007.2580 [hep-lat]].DOI
  5. [5]J. S. Moreland, J. E. Bernhard and S. A. Bass, Phys. Rev. C 92, no.1, 011901 (2015) doi:10.1103/PhysRevC.92.011901 [arXiv:1412.4708 [nucl-th]].DOI
  6. [6]B. Schenke, P. Tribedy and R. Venugopalan, Phys. Rev. Lett. 108, 252301 (2012) doi:10.1103/PhysRevLett.108.252301 [arXiv:1202.6646 [nucl-th]].DOI
  7. [7]C. Gale, S. Jeon, B. Schenke, P. Tribedy and R. Venugopalan, Phys. Rev. Lett. 110, no.1, 012302 (2013) doi:10.1103/PhysRevLett.110.012302 [arXiv:1209.6330 [nucl-th]].DOI
  8. [8]B. Schenke, S. Jeon and C. Gale, Phys. Rev. C 82, 014903 (2010) doi:10.1103/PhysRevC.82.014903 [arXiv:1004.1408 [hep-ph]].DOI
  9. [9]K. J. Eskola, K. Kajantie, P. V. Ruuskanen and K. Tuominen, Nucl. Phys. B 570, 379-389 (2000) doi:10.1016/S0550-3213(99)00720-8 [arXiv:hep-ph/9909456 [hep-ph]].DOI
  10. [10]K. J. Eskola, K. Kajantie and K. Tuominen, Phys. Lett. B 497, 39-43 (2001) doi:10.1016/S0370-2693(00)01341-1 [arXiv:hep-ph/0009246 [hep-ph]].DOI
  11. [11]R. Paatelainen, K. J. Eskola, H. Holopainen and K. Tuominen, Phys. Rev. C 87, no.4, 044904 (2013) doi:10.1103/PhysRevC.87.044904 [arXiv:1211.0461 [hep-ph]].DOI
  12. [12]R. Paatelainen, K. J. Eskola, H. Niemi and K. Tuominen, Phys. Lett. B 731, 126-130 (2014) doi:10.1016/j.physletb.2014.02.018 [arXiv:1310.3105 [hep-ph]].
  13. [13]H. Niemi, K. J. Eskola and R. Paatelainen, Phys. Rev. C 93, no.2, 024907 (2016) doi:10.1103/PhysRevC.93.024907 [arXiv:1505.02677 [hep-ph]].DOI
  14. [14]M. Kuha, J. Auvinen, K. J. Eskola, H. Hirvonen, Y. Kanakubo and H. Niemi, Phys. Rev. C 111, no.5, 054914 (2025) doi:10.1103/PhysRevC.111.054914 [arXiv:2406.17592 [hep-ph]].DOI
  15. [15]H. Hirvonen, M. Kuha, J. Auvinen, K. J. Eskola, Y. Kanakubo and H. Niemi, Phys. Rev. C 110, no.3, 034911 (2024) doi:10.1103/PhysRevC.110.034911 [arXiv:2407.01338 [hep-ph]].DOI
  16. [16]T. Pierog and K. Werner, Nucl. Phys. B Proc. Suppl. 196, 102-105 (2009) doi:10.1016/j.nuclphysbps.2009.09.017 [arXiv:0905.1198 [hep-ph]].DOI
  17. [17]T. Pierog, I. Karpenko, J. M. Katzy, E. Yatsenko and K. Werner, Phys. Rev. C 92, no.3, 034906 (2015) doi:10.1103/PhysRevC.92.034906 [arXiv:1306.0121 [hep-ph]].DOI
  18. [18]K. Werner, B. Guiot, I. Karpenko and T. Pierog, Phys. Rev. C 89, no.6, 064903 (2014) doi:10.1103/PhysRevC.89.064903 [arXiv:1312.1233 [nucl-th]].DOI
  19. [19]K. Werner, Phys. Rev. C 108, no.6, 064903 (2023) doi:10.1103/PhysRevC.108.064903 [arXiv:2301.12517 [hep-ph]].DOI
  20. [20]K. Werner, Phys. Rev. C 109, no.1, 014910 (2024) doi:10.1103/PhysRevC.109.014910 [arXiv:2306.10277 [hep-ph]].DOI
  21. [21]K. Werner and B. Guiot, Phys. Rev. C 108, no.3, 034904 (2023) doi:10.1103/PhysRevC.108.034904 [arXiv:2306.02396 [hep-ph]].DOI
  22. [22]A. Kurkela, A. Mazeliauskas, J. F. Paquet, S. Schlichting and D. Teaney, Phys. Rev. C 99, no.3, 034910 (2019) doi:10.1103/PhysRevC.99.034910 [arXiv:1805.00961 [hep-ph]].DOI
  23. [23]A. Kurkela, R. Törnkvist and K. Zapp, Eur. Phys. J. C 84, no.1, 74 (2024) doi:10.1140/epjc/s10052-024-12424-2 [arXiv:2211.15454 [hep-ph]].DOI
  24. [24]S. A. Bass, M. Belkacem, M. Bleicher, M. Brandstetter, L. Bravina, C. Ernst, L. Gerland, M. Hofmann, S. Hofmann and J. Konopka, et al. Prog. Part. Nucl. Phys. 41, 255-369 (1998) doi:10.1016/S0146-6410(98)00058-1 [arXiv:nucl-th/9803035 [nucl-th]].DOI
  25. [25]M. Bleicher, E. Zabrodin, C. Spieles, S. A. Bass, C. Ernst, S. Soff, L. Bravina, M. Belkacem, H. Weber and H. Stoecker, et al. J. Phys. G 25, 1859-1896 (1999) doi:10.1088/0954-3899/25/9/308 [arXiv:hep-ph/9909407 [hep-ph]].DOI
  26. [26]J. Weil et al. [SMASH], Phys. Rev. C 94, no.5, 054905 (2016) doi:10.1103/PhysRevC.94.054905 [arXiv:1606.06642 [nucl-th]].DOI
  27. [27]H. Hirvonen, K. J. Eskola and H. Niemi, Phys. Rev. C 106, no.4, 044913 (2022) doi:10.1103/PhysRevC.106.044913 [arXiv:2206.15207 [hep-ph]].DOI
  28. [28]U. Heinz and B. Schenke, [arXiv:2412.19393 [nucl-th]].arxiv.org/abs/2412.19393
  29. [29]J. Novak, K. Novak, S. Pratt, J. Vredevoogd, C. Coleman-Smith and R. Wolpert, Phys. Rev. C 89, no.3, 034917 (2014) doi:10.1103/PhysRevC.89.034917 [arXiv:1303.5769 [nucl-th]].DOI
  30. [30]J. E. Bernhard, J. S. Moreland and S. A. Bass, Nature Phys. 15, no.11, 1113-1117 (2019) doi:10.1038/s41567-019-0611-8DOI
  31. [31]J. Auvinen, K. J. Eskola, P. Huovinen, H. Niemi, R. Paatelainen and P. Petreczky, Phys. Rev. C 102, no.4, 044911 (2020) doi:10.1103/PhysRevC.102.044911 [arXiv:2006.12499 [nucl-th]].DOI
  32. [32]D. Everett et al. [JETSCAPE], Phys. Rev. Lett. 126, no.24, 242301 (2021) doi:10.1103/PhysRevLett.126.242301 [arXiv:2010.03928 [hep-ph]].DOI
  33. [33]D. Everett et al. [JETSCAPE], Phys. Rev. C 103, no.5, 054904 (2021) doi:10.1103/PhysRevC.103.054904 [arXiv:2011.01430 [hep-ph]].DOI
  34. [34]R. Ehlers et al. [JETSCAPE], Phys. Rev. C 111, no.5, 054913 (2025) doi:10.1103/PhysRevC.111.054913 [arXiv:2408.08247 [hep-ph]].DOI
  35. [35]G. Nijs, W. van der Schee, U. Gürsoy and R. Snellings, Phys. Rev. Lett. 126, no.20, 202301 (2021) doi:10.1103/PhysRevLett.126.202301 [arXiv:2010.15130 [nucl-th]].DOI
  36. [36]G. Nijs, W. van der Schee, U. Gürsoy and R. Snellings, Phys. Rev. C 103, no.5, 054909 (2021) doi:10.1103/PhysRevC.103.054909 [arXiv:2010.15134 [nucl-th]].DOI
  37. [37]G. Nijs and W. van der Schee, Phys. Rev. Lett. 129, no.23, 232301 (2022) doi:10.1103/PhysRevLett.129.232301 [arXiv:2206.13522 [nucl-th]].DOI
  38. [38]G. Nijs and W. van der Schee, [arXiv:2304.06191 [nucl-th]].arxiv.org/abs/2304.06191
  39. [39]G. Giacalone, G. Nijs and W. van der Schee, Phys. Rev. Lett. 131, no.20, 20 (2023) doi:10.1103/PhysRevLett.131.202302 [arXiv:2305.00015 [nucl-th]].DOI
  40. [40]H. Mäntysaari, B. Schenke, C. Shen and W. Zhao, Phys. Lett. B 833, 137348 (2022) doi:10.1016/j.physletb.2022.137348 [arXiv:2202.01998 [hep-ph]].DOI
  41. [41]M. R. Heffernan, C. Gale, S. Jeon and J. F. Paquet, Phys. Rev. C 109, no.6, 065207 (2024) doi:10.1103/PhysRevC.109.065207 [arXiv:2302.09478 [nucl-th]].DOI
  42. [42]M. R. Heffernan, C. Gale, S. Jeon and J. F. Paquet, Phys. Rev. Lett. 132, no.25, 252301 (2024) doi:10.1103/PhysRevLett.132.252301 [arXiv:2306.09619 [nucl-th]].DOI
  43. [43]S. A. Jahan, H. Roch and C. Shen, Phys. Rev. C 110, no.5, 5 (2024) doi:10.1103/PhysRevC.110.054905 [arXiv:2408.00537 [nucl-th]].DOI
  44. [44]J. E. Parkkila, A. Onnerstad and D. J. Kim, Phys. Rev. C 104, no.5, 054904 (2021) doi:10.1103/PhysRevC.104.054904 [arXiv:2106.05019 [hep-ph]].DOI
  45. [45]J. E. Parkkila, A. Onnerstad, S. F. Taghavi, C. Mordasini, A. Bilandzic, M. Virta and D. J. Kim, Phys. Lett. B 835, 137485 (2022) doi:10.1016/j.physletb.2022.137485 [arXiv:2111.08145 [hep-ph]].DOI
  46. [47]N. Götz, I. Karpenko and H. Elfner, Phys. Rev. C 112, no.1, 014910 (2025) doi:10.1103/rzmI-rjzx [arXiv:2503.10181 [nucl-th]].arxiv.org/abs/2503.10181
  47. [48]J. F. Paquet, J. Phys. G 51, no.10, 103001 (2024) doi:10.1088/1361-6471/ad6a2b [arXiv:2310.17618 [nucl-th]].DOI
  48. [49]Rasmussen, C. E. and Williams, C. K. I., “Gaussian Processes for Machine Learning,” The MIT Press, 2005. doi:10.7551/mitpress/3206.001.0001DOI
  49. [50]J. E. Bernhard, PhD thesis, Duke University, 2018 [arXiv:1804.06469 [nucl-th]].arxiv.org/abs/1804.06469
  50. [51]D. S. Everett, PhD thesis, The Ohio State University, 2021 [arXiv:2107.11362 [hep-ph]].arxiv.org/abs/2107.11362
  51. [52]H. Hirvonen, K. J. Eskola and H. Niemi, Phys. Rev. C 108, no.3, 034905 (2023) doi:10.1103/PhysRevC.108.034905 [arXiv:2303.04517 [hep-ph]].DOI
  52. [53]H. Hirvonen, K. J. Eskola and H. Niemi, EPJ Web Conf. 296, 02002 (2024) doi:10.1051/epjconf/202429602002 [arXiv:2404.02602 [hep-ph]].DOI
  53. [54]J. A. Sun, L. Yan, C. Gale and S. Jeon, Phys. Rev. C 112, no.5, L051903 (2025) doi:10.1103/6skc-x5zj [arXiv:2410.13069 [nucl-th]].
  54. [55]M. Omana Kuttan, K. Zhou, J. Steinheimer and H. Stoecker, Phys. Rev. C 112, no.5, 5 (2025) doi:10.1103/wyq5-hlp5 [arXiv:2502.16330 [hep-ph]].
  55. [56]D. Stewart and J. Putschke, Phys. Rev. C 113, no.1, 014904 (2026) doi:10.1103/jsh8-6tm7 [arXiv:2507.23598 [nucl-th]].DOI
  56. [57]K. J. Eskola and K. Tuominen, Phys. Lett. B 489, 329-336 (2000) doi:10.1016/S0370-2693(00)00946-1 [arXiv:hep-ph/0002008 [hep-ph]].DOI
  57. [58]K. J. Eskola and K. Tuominen, Phys. Rev. D 63, 114006 (2001) doi:10.1103/PhysRevD.63.114006 [arXiv:hep-ph/0010319 [hep-ph]].DOI
  58. [59]K. J. Eskola, K. Kajantie and J. Lindfors, Nucl. Phys. B 323, 37-52 (1989) doi:10.1016/0550-3213(89)90586-5DOI
  59. [60]L. V. Gribov, E. M. Levin and M. G. Ryskin, Phys. Rept. 100, 1-150 (1983) doi:10.1016/0370-1573(83)90022-4DOI
  60. [61]I. Helenius, K. J. Eskola, H. Honkanen and C. A. Salgado, JHEP 07, 073 (2012) doi:10.1007/JHEP07(2012)073 [arXiv:1205.5359 [hep-ph]].DOI
  61. [62]P. Möller, A. J. Sierk, T. Ichikawa and H. Sagawa, Atom. Data Nucl. Data Tabl. 109-110, 1-204 (2016) doi:10.1016/j.adt.2015.10.002 [arXiv:1508.06294 [nucl-th]].DOI
  62. [63]H. Niemi, K. J. Eskola, R. Paatelainen and K. Tuominen, Phys. Rev. C 93, no.1, 014912 (2016) doi:10.1103/PhysRevC.93.014912 [arXiv:1511.04296 [hep-ph]].DOI
  63. [64]K. J. Eskola, H. Niemi, R. Paatelainen and K. Tuominen, Phys. Rev. C 97, no.3, 034911 (2018) doi:10.1103/PhysRevC.97.034911 [arXiv:1711.09803 [hep-ph]].DOI
  64. [65]P. Huovinen and P. Petreczky, Nucl. Phys. A 837, 26-53 (2010) doi:10.1016/j.nuclphysa.2010.02.015 [arXiv:0912.2541 [hep-ph]].DOI
  65. [66]P. Huovinen, Eur. Phys. J. A 37, 121-128 (2008) doi:10.1140/epja/i2007-10611-3 [arXiv:0710.4379 [nucl-th]].DOI
  66. [67]W. Israel and J. M. Stewart, Annals Phys. 118, 341-372 (1979) doi:10.1016/0003-4916(79)90130-1 G. S. Denicol, T. Koide and D. H. Rischke, Phys. Rev. Lett. 105, 162501 (2010) doi:10.1103/PhysRevLett.105.162501 [arXiv:1004.5013 [nucl-th]].
  67. [69]G. S. Denicol, H. Niemi, E. Molnar and D. H. Rischke, Phys. Rev. D 85, 114047 (2012) [erratum: Phys. Rev. D 91, no.3, 039902 (2015)] doi:10.1103/PhysRevD.85.114047 [arXiv:1202.4551 [nucl-th]].DOI
  68. [70]E. Molnár, H. Niemi, G. S. Denicol and D. H. Rischke, Phys. Rev. D 89, no.7, 074010 (2014) doi:10.1103/PhysRevD.89.074010 [arXiv:1308.0785 [nucl-th]].DOI
  69. [71]G. S. Denicol, S. Jeon and C. Gale, Phys. Rev. C 90, no.2, 024912 (2014) doi:10.1103/PhysRevC.90.024912 [arXiv:1403.0962 [nucl-th]].DOI
  70. [72]K. Paech and S. Pratt, Phys. Rev. C 74, 014901 (2006) [erratum: Phys. Rev. C 93, no.5, 059902 (2016)] doi:10.1103/PhysRevC.74.014901 [arXiv:nucl-th/0604008 [nucl-th]].DOI
  71. [73]K. Dusling and T. Schäfer, Phys. Rev. C 85, 044909 (2012) doi:10.1103/PhysRevC.85.044909 [arXiv:1109.5181 [hep-ph]].DOI
  72. [74]J. B. Rose, J. M. Torres-Rincon and H. Elfner, J. Phys. G 48, no.1, 015005 (2020) doi:10.1088/1361-6471/abbc5c [arXiv:2005.03647 [hep-ph]].DOI
  73. [75]H. Hirvonen, PhD thesis, University of Jyväskylä, 2024
  74. [76]G. Huang, Z. Liu, L. van der Maaten, and K. Q. Weinberger, Proceedings of 2017 IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2261-2269 (2017).
  75. [77]D. P. Kingma and J. Ba, [arXiv:1412.6980 [cs.LG]].arxiv.org/abs/1412.6980
  76. [78]H. Hirvonen, J. Auvinen, H. Niemi and K. J. Eskola, Zenodo (2026). doi: 10.5281/zenodo.20499599.
  77. [79]F. Pedregosa, G. Varoquaux, A. Gramfort, V. Michel, B. Thirion, O. Grisel, M. Blondel, P. Prettenhofer, R. Weiss, V. Dubourg, J. Vanderplas, A. Passos, D. Cournapeau, M. Brucher, M. Perrot, E. Duchesnay, Journal of Machine Learning Research 12, 2825 (2011).
  78. [80]S. S. Adler et al. [PHENIX], Phys. Rev. C 71, 034908 (2005) [erratum: Phys. Rev. C 71, 049901 (2005)] doi:10.1103/PhysRevC.71.034908 [arXiv:nucl-ex/0409015 [nucl-ex]].
  79. [81]K. Aamodt et al. [ALICE], Phys. Rev. Lett. 106, 032301 (2011) doi:10.1103/PhysRevLett.106.032301 [arXiv:1012.1657 [nucl-ex]].DOI
  80. [82]J. Adam et al. [ALICE], Phys. Rev. Lett. 116, no.22, 222302 (2016) doi:10.1103/PhysRevLett.116.222302 [arXiv:1512.06104 [nucl-ex]].DOI
  81. [83]S. Acharya et al. [ALICE], Phys. Lett. B 788, 166-179 (2019) doi:10.1016/j.physletb.2018.10.052 [arXiv:1805.04399 [nucl-ex]].DOI
  82. [84]S. S. Adler et al. [PHENIX], Phys. Rev. C 69, 034909 (2004) doi:10.1103/PhysRevC.69.034909 [arXiv:nucl-ex/0307022 [nucl-ex]].DOI
  83. [85]B. Abelev et al. [ALICE], Phys. Rev. C 88, 044910 (2013) doi:10.1103/PhysRevC.88.044910 [arXiv:1303.0737 [hep-ex]].DOI
  84. [86]S. Acharya et al. [ALICE], Phys. Rev. C 101, no.4, 044907 (2020) doi:10.1103/PhysRevC.101.044907 [arXiv:1910.07678 [nucl-ex]].DOI
  85. [87]L. Adamczyk et al. [STAR], Phys. Rev. C 98, no.3, 034918 (2018) doi:10.1103/PhysRevC.98.034918 [arXiv:1701.06496 [nucl-ex]].DOI
  86. [88]S. Acharya et al. [ALICE], JHEP 07, 103 (2018) doi:10.1007/JHEP07(2018)103 [arXiv:1804.02944 [nucl-ex]].DOI
  87. [89]S. Acharya et al. [ALICE], Phys. Lett. B 784, 82-95 (2018) doi:10.1016/j.physletb.2018.06.059 [arXiv:1805.01832 [nucl-ex]].DOI
  88. [90]J. Adam et al. [STAR], Phys. Lett. B 783, 459-465 (2018) doi:10.1016/j.physletb.2018.05.076 [arXiv:1803.03876 [nucl-ex]].DOI
  89. [91]J. Adam et al. [ALICE], Phys. Rev. Lett. 117, 182301 (2016) doi:10.1103/PhysRevLett.117.182301 [arXiv:1604.07663 [nucl-ex]].DOI
  90. [92]S. Acharya et al. [ALICE], Phys. Lett. B 818, 136354 (2021) doi:10.1016/j.physletb.2021.136354 [arXiv:2102.12180 [nucl-ex]].DOI
  91. [93]D. Foreman-Mackey, D. W. Hogg, D. Lang and J. Goodman, Publ. Astron. Soc. Pac. 125, 306-312 (2013) doi:10.1086/670067 [arXiv:1202.3665 [astro-ph.IM]].
  92. [94]S. Chekanov et al. [ZEUS], Eur. Phys. J. C 24, 345-360 (2002) doi:10.1007/s10052-002-0953-7 [arXiv:hep-ex/0201043 [hep-ex]].DOI
  93. [95]H. Roch, S. A. Jahan and C. Shen, Phys. Rev. C 110, no.4, 4 (2024) doi:10.1103/PhysRevC.110.044904 [arXiv:2405.12019 [nucl-th]].DOI

Paper details

vlnc:W_myjpwvgg
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC BY 4.0
Usage

views · 0

downloads · 0

A view counts once per browser per day, a download each time the PDF is requested.

  • Physics and Astronomy

Not recorded for this paper

View statement
Looking for related papers…
Contents