WorksPhysics and Astronomy
Relativistic viscous hydrodynamics in the Density Frame for heavy-ion collisions
Abstract
We present a numerical study of relativistic viscous hydrodynamics in the Density Frame for boost-invariant flows in 2+1 dimensions. The Density Frame is a formulation of relativistic viscous hydrodynamics that is first order in time, requires no auxiliary fields, and has no non-hydrodynamic modes; the only parameters are the shear and bulk viscosities and the equation of state. We formulate the numerical method in terms of a discrete analog of the dissipative function, which guarantees that the discrete equations produce entropy on all length scales and leads to a remarkably robust algorithm. We test the method against the exact solution of kinetic theory in the relaxation time approximation with Gubser symmetry. In making this comparison, we clarify how the kinetic solution matches onto hydrodynamics: we organize the solution by the opacity and show how the Bjorken attractor passes onto the Gubser solution order by order in gradients. We then replace the hydrodynamic engine of the MUSIC code with the Density Frame, and compare the two for IP-Glasma initial conditions in Pb-Pb collisions at TeV. The Density Frame smooths sharp gradients more rapidly. The largest differences arise in the bulk-viscous sector, where the bulk pressure in the Density Frame responds instantaneously to the flow, while in the Müller-Israel-Stewart-based MUSIC it relaxes over a finite time. With shear viscosity alone, the multiplicity, mean transverse momentum, and elliptic flow of final-state hadrons from the two codes are very similar. The Density Frame is now ready for a complete Bayesian analysis of heavy-ion data.
Introduction
Hydrodynamics, transport coefficients, and the Density Frame
Nuclear collisions at the Relativistic Heavy Ion Collider (RHIC) and the Large Hadron Collider (LHC) exhibit remarkable patterns of collective flow, which are described by relativistic viscous hydrodynamics [1, 2, 3, 4, 5]. Through multistage simulations of heavy-ion collisions built around hydrodynamics, these flow patterns constrain the transport properties of the quark-gluon plasma and its equation of state. Bayesian comparisons between simulation and experiment now give increasingly quantitative estimates of the shear and bulk viscosities of QCD [6, 7, 8, 9, 10], and flow-like correlations are seen even in proton-nucleus and other small systems [11, 12].
There are, however, conceptual challenges with the hydrodynamic theories used in these simulations. Relativistic Navier–Stokes theory in the Landau or Eckart frame is unstable and acausal [13]. The standard remedy, due to Müller and to Israel and Stewart (MIS) [14, 15, 16], promotes the viscous stresses to dynamical fields which relax to their Navier–Stokes values over a time . Nearly all simulations of heavy-ion collisions use a variant of this approach, often called second-order hydrodynamics [17, 18]. The more recent formulation of Bemfica, Disconzi, Noronha and Kovtun (BDNK) [19, 20] adds no new fields, but its equations are second order in time and depend on additional frame parameters. In each case the extra freedom introduces non-hydrodynamic modes. These modes are needed to preserve causality, but they model the microscopic physics in a non-universal way and lie outside the realm of hydrodynamics. The associated model parameters, such as in MIS theory or the frame parameters of BDNK, can obscure which universal transport properties are actually being extracted from data, especially in small systems.
The Density Frame formulation of viscous hydrodynamics has no additional non-hydrodynamic modes or parameters, resolving these conceptual and practical issues [21, 22, 23, 24]. Given a foliation of spacetime, the conserved energy and momentum densities on each spatial slice, and , are the slow variables of the effective theory. They are evolved with the conservation laws, and all other quantities, including the spatial stress , are determined instantaneously by and through the gradient expansion. The theory is strictly first order in gradients, and the only parameters are the shear and bulk viscosities and the equation of state. The equations are not manifestly covariant, but they are covariant to a given order in the gradient expansion, provided a Lorentz transformation is accompanied by a change of hydrodynamic frame [21, 22].
Although it is not the focus of this work, the Density Frame has an added benefit: the equations fit neatly into the traditional framework of dissipative stochastic processes [25, 26, 27, 28]. Noise can be added with Metropolis steps [22, 23], an approach that has proved useful for simulating systems close to critical points [29, 30, 31, 32, 33] and ordinary stochastic fluids [34]. We have previously studied the Density Frame numerically for the stochastic relativistic advection-diffusion equation [22].
This work
The deterministic version of Density Frame Hydrodynamics was studied previously in dimensions, where we compared it with QCD kinetic theory, BDNK, and MIS (MUSIC) [24]. In this paper, we extend these numerical methods and we solve the Density Frame equations in dimensions in Bjorken coordinates, establishing the Density Frame as a fully developed tool for studying heavy-ion collisions. In formulating the method we use a traditional concept, the dissipative function [25, 26, 35], which is the positive semi-definite quadratic form in the strains that determines the rate of entropy production. We find that it has a discrete analog: the viscous stresses are derivatives of a discrete dissipative function, so the semi-discrete viscous step produces entropy in the same way as the continuum equations. This makes the method remarkably robust: we needed only a very small energy-density pedestal to handle the dilute edge of the plasma, a level of regulation that we generally found insufficient for MIS hydrodynamic solvers. The same methods can be used in other domains, for example for viscous hydrodynamics in general relativity [23].
In this paper, we test the numerical scheme against the exact solution of RTA kinetic theory with Gubser symmetry [36, 37, 38]. In making this comparison we clarify how the kinetic solution matches onto hydrodynamics, which has not been fully explained in the literature. We connect the Gubser kinetic solution to the opacity expansion introduced in recent kinetic simulations [39, 40, 41], and show how the Bjorken attractor [42, 43] passes smoothly onto the Gubser solution, order by order in gradients. We then use this understanding to compare the kinetic solution with our numerical Density Frame solution. The comparison also tests the approximate foliation independence (and by extension Lorentz invariance) of the Density Frame: in Gubser’s coordinates, with as the time variable, the fluid is at rest and the Density Frame solution is Gubser’s hydrodynamic solution. Our code, however, uses Bjorken as the time variable, and agrees with Gubser’s solution up to controlled gradient corrections.
Finally, we have constructed a practical tool—packaged, public, and documented—so that it can replace the hydrodynamic engine of existing heavy-ion codes [44]. As an example, we replaced the hydrodynamic engine of MUSIC [45, 46, 47] with the Density Frame and compared the two for IP-Glasma initial conditions [48]. The Density Frame rapidly smooths out very sharp gradients; this dissipation of fine structure seems to us more faithful to the physics of viscous hydrodynamics. The largest difference is in the bulk-viscous sector: in MIS simulations such as MUSIC the bulk pressure relaxes slowly near the crossover, while in the Density Frame it responds instantaneously. Building on these results, we show that simulations of heavy-ion collisions that replace MIS hydrodynamics with Density Frame lead to similar flow patterns and final hadronic observables. In the future, this framework can be used to investigate the sensitivity to non-hydrodynamic modes in small systems such as p+Pb and O+O collisions [49], and to address the theoretical error bars in current Bayesian extractions of transport coefficients.
The paper is organized as follows. Sect. II presents the Density Frame equations in Bjorken coordinates, the dissipative function, and the numerical method. Sect. III compares the method with the Gubser kinetic solution, clarifying the matching between the Bjorken attractor and three-dimensional hydrodynamics in the opacity expansion. Sect. IV compares the Density Frame with MUSIC for realistic initial conditions, and Sect. V concludes.
The Density Frame for 2+1D Bjorken Flow
In this section we first review the Density Frame formulation of relativistic viscous hydrodynamics, which is described in detail in our previous work [23, 24], and then present the numerical method. We focus on the extension to Bjorken coordinates, and initially describe the method in 1+1D for simplicity. With the right notation and concepts, the extension to 2+1D is straightforward. Relative to our previous work, the main new ingredient is to make maximal use of the dissipative function [25, 26, 35], which is the positive semi-definite quadratic form that appears in the entropy production. We will see that it is possible to write down the dissipative function in a discrete form, and that the viscous strains can be implemented by taking derivatives of this function at the discrete level. This approach is very effective and leads to a delightfully stable numerical method.
The equations of motion
Let us consider a special case of Bjorken flow with a non-zero flow velocity and spatial dependence only in the direction. The metric in Bjorken coordinates is
and the non-zero Christoffel symbols are and . The normal vector for the foliation is simply and the spatial metric is where roman indices run over the spatial coordinates . The volume element of the spatial metric is . In two dimensions, we will use for the transverse directions . The foliation provided by the Bjorken coordinates has extrinsic curvature , which has only one non-zero component . To highlight the role of the extrinsic curvature we will write instead of simply in the equations of motion.
The evolution variables in hydrodynamics are the energy and momentum densities
and the equations of energy-momentum conservation read
These equations are extended to two dimensions in an obvious way by including the direction and the corresponding momentum density . The explicit factors of represent the volume element, .
The conservation equations are closed by specifying a constitutive relation for the spatial stress tensor in terms of and their spatial derivatives. In ideal hydrodynamics this is conveniently done by introducing intermediate parameters (the inverse temperature and flow velocity), which are defined by algebraic relations to the conserved charge densities on the spatial slice:
The ideal stress tensor written here takes the functional form
This equation is to be understood as follows: is determined from the energy and momentum densities as specified by eq. (4), and subsequently is used to specify the spatial stress
It is important that is ultimately a function of .
In viscous hydrodynamics the spatial stress tensor receives derivative corrections
In the Density Frame the algebraic relations in eq. (4) define to all orders in the derivative expansion. The viscous strains in Bjorken coordinates are proportional to the spatial covariant derivatives of :
A general expression for the tensor is given in our companion paper and depends on the shear and bulk viscosities, the flow velocity of the fluid , and the speed of sound [23]. In dimensions with only dependence, the symmetries of the problem imply that the only relevant components of are
Here and . The determinant of this matrix is
which is positive for . For bulk viscosity alone the matrix has rank one, since bulk viscosity couples only to the expansion rate, and it is positive semi-definite.
The Density Frame and the more familiar Landau frame describe the same stress tensor, but split it differently into ideal and viscous parts. The Landau-frame energy density and flow velocity are defined by the eigenvalue problem
and the Landau-frame viscous stress, , is transverse to . The two frames differ at first order in gradients, so the Landau-frame temperature and flow velocity differ from the Density Frame by first-order corrections. When comparing with kinetic theory (Sect. III) and with MUSIC (Sect. IV), we construct the full from the Density Frame solution and transform it to the Landau frame. Existing particlization codes also assume Landau-frame variables, so this transformation should be made before freezeout.
Entropy production and the dissipative function
Before describing the numerical method, we show that the Density Frame equations produce entropy, and identify the quadratic form that controls this production, known as the dissipative function. This quadratic form is the central ingredient of the numerical method in Sect. II C. Using the equations of motion, it is easy to evaluate the entropy production. The entropy density on a constant slice is defined as in ideal hydrodynamics, , where the intermediate parameters are functions of the conserved charges. For a fluid cell of volume , with entropy and four-momentum , the thermodynamic identity reads [23]
Since , the entropy density evolves as
and upon using the equations of motion, we find
The right hand side is positive semi-definite as long as the matrix of transport coefficients is positive definite.
We will define the dissipative function as half the positive semi-definite quadratic form that appears in the entropy production:
Regarding the as a constant metric on the space of strains , the viscous strains are given by the functional derivative of with respect to :
When evaluating the rate of entropy production, i.e.
upon using the equations of motion and dropping the entropy flux at the boundaries, the ideal terms do not contribute, and the viscous terms can be written in terms of the dissipative function as
When devising a numerical method, we will write down the discrete version of the dissipative function and implement the discretized viscous strains by taking derivatives of this function.
Numerical method
As in our previous work we will use operator splitting together with Implicit-Explicit (IMEX) time integrators [50, 51, 52] to evolve the system. The variables are denoted generically1 and evolve schematically as
where and represent the ideal and viscous hydrodynamic evolution respectively. The ideal evolution is strongly non-linear and will be treated explicitly, while the viscous evolution will be treated implicitly.
In the explicit step, , the variables are evolved over a time interval using the equations of motion:
For the explicit step we use a standard Kurganov-Tadmor (KT) scheme of ideal hydrodynamics [45, 53, 54].
In the implicit step, , the variables are evolved with the viscous equations of motion,
At first order in , the implicit update has the schematic form
which in general is non-linear and must be solved with a Newton solver. Combining the explicit and implicit steps in sequence leads to a non-linear IMEX scheme. Higher-order IMEX schemes are implemented in PETSc [55], which we use both for time stepping and for solving the implicit non-linear equations, as in our previous work [24]. Our remaining task is to specify in detail the function and its Jacobian for the Newton solver.
To this end, we write down the discrete version of the dissipative function . Then taking derivatives of this function with respect to the variables at the -th lattice site, and , defines the discretization. We will call the vertex located halfway between the and lattice sites the vertex, with an analogous terminology for the vertex. A generic variable at the and vertices is denoted and , respectively. These variables will be defined as suitable averages of the variables at the neighboring lattice sites (see below). We then introduce the discrete dissipative function
where is the contribution to the dissipative function from the vertex
Then the discretized viscous strains are given by
and
respectively, which gives concrete meaning to eq. (23). As in the continuum, is a function of and and and the derivatives of with respect to and are to be evaluated by treating as a constant metric on the space of strains. In practice we evaluate at by averaging the conserved charges of the neighboring lattice site, using these averages to define , and finally evaluating .
There are several benefits to defining the discretization in this way. First we note that the difference structure of the conservation equations is automatically implemented by the definition of . For example, the viscous contribution to the momentum equation is given by the difference of the viscous stress at the neighboring lattice sites
Further, with this definition of the entropy production in the viscous step is automatically positive semi-definite in a semi-discrete sense. To see this, we note that in the viscous step only and change, at fixed , so the total rate of entropy production on the lattice, with , is given by
which follows from the definition of the entropy. Using the semi-discrete viscous equations of motion, eq. (23), the right hand side can be written as
Thus the total rate of entropy production in the viscous step is given by , as in the continuum.
Now we will write down the explicit form of the viscous strains by differentiating and introducing a compact notation for the finite differences. Let denote the relative position with respect to the -th lattice site. means for and for . We then introduce the following notation for the finite difference operators
for and respectively. Then the form of is given by
where we are summing over the repeated indices and . Differentiating with respect to and yields the appropriate finite difference contribution to the viscous strains at the -th and -th lattice sites from . Thus, the discrete equation for the time evolution is given in part by
The complete time evolution is found by summing over the contributions from all vertices (i.e. and for lattice site ) to yield in its complete form. To summarize, eq. (34) defines the function in the implicit equation .
The procedure is now easily extended to the 2+1D case. The two-dimensional lattice sites are labeled by the vector and the relative positions are labeled by the vector where and can take the values 0 and 1. means for where , , , and . We will write down the contribution of to the viscous strains. The finite difference operators are defined analogously to the 1+1D case,
The equations of motion are decorated with transverse indices
To write down an approximate Jacobian for implicit solver, we have to evaluate the derivatives of the right-hand side of eq. (36) with respect to the variables on the lattice . In keeping with the spirit of the dissipative function, we will treat as a constant, when taking these derivatives. The Jacobian of the discretized operator then involves the thermodynamic susceptibility defined by
Explicit formulas for this susceptibility matrix are given in our previous work [23, 24]. Using a compact notation eq. (34) can be written as
and thus the contribution of to the Jacobian at sites and is
Adding the contributions of this sort from all of the vertices, determines the full Jacobian matrix for the Newton solver, .
So far we have presented a fully non-linear IMEX scheme. A simpler alternative to a fully non-linear implicit step, known as Extrapolated IMEX (EIMEX), is to linearize the problem over a time interval [52]. We adopted this approach in our previous numerical work on the Density Frame [24] and found it to be as effective and robust as the fully non-linear method, while being easier to implement and faster too.
In this approach, we solve the linearized equation
where and is a time-independent approximate Jacobian matrix written down in the preceding paragraphs. This leads to the update
so that the increment is obtained by solving the linear system
The first order EIMEX scheme presented here, as well as its higher-order generalizations, are all implemented in PETSc [55]. We adopted a third-order EIMEX scheme studied in our previous work [24]. We used PETSc both for EIMEX time stepping and to solve the linear systems arising in the linearized treatment. In practice, the default linear solver in PETSc—a GMRES solver [56] with either an ILU or a Jacobi preconditioner—has proven to be efficient for these linear systems.
Comparison to the Gubser solution: an exact kinetic benchmark
To test Density Frame hydrodynamics we need a reference solution with transverse expansion in which the approach to hydrodynamics is under control. Gubser flow provides one. It is a conformal, boost-invariant and azimuthally symmetric flow with both longitudinal and radial expansion [57, 58], and the Boltzmann equation in the relaxation time approximation (RTA) can be solved exactly with the same symmetry [36, 37]. The kinetic solution and its relation to hydrodynamics have been studied extensively [38, 59, 60]. However, how the kinetic solution matches onto hydrodynamics has not been fully explained. Here we present it in a form suited to heavy-ion collisions, with the initial conditions set at , and organize it by the opacity [39, 40, 41, 61, 62]. We show how the Bjorken attractor [42, 43, 62, 63] passes onto the Gubser solution order by order in the gradient expansion, and use this to make a clean comparison with Density Frame hydrodynamics.
Gubser’s hydrodynamic solution has a single integration constant, which we take to be the central temperature , i.e. the temperature at and . This integration constant fixes the entire Gubser flow to all orders in the gradient expansion [64]. Our goal in this section is to relate to the initial conditions. The initial conditions have a momentum scale and a transverse size ,2 and the dynamics is set by one transport coefficient, . We work with kinetic theory, which is valid in the semi-classical limit where and , i.e. momenta are large compared with and the mean free path is long compared with the thermal wavelength.3 In this limit, with momenta measured in units of and lengths and times in units of , the solution of the kinetic equation depends on , and only through the opacity
which measures the number of relaxation times a particle experiences over a time (see below). The semi-classical limit is taken with held fixed, i.e. and with . We determine for all opacities, from free streaming to the hydrodynamic regime (Figure 1). Since fixes the hydrodynamic flow, this gives a complete prediction for Gubser flow in terms of the initial conditions and , and the full kinetic solution then provides a clean test of the Density Frame (Sect. III C).
Figure 1. Central temperature at in units of versus inverse opacity (left) and opacity (right). Solid: RTA kinetic theory. Dashed: first-order hydrodynamics, . Dotted: second-order hydrodynamics, , shown up to its minimum at . Dash-dotted: free streaming, . The arrows mark , the opacity used in Figure 2.
This section uses Milne coordinates , with the spacetime rapidity; the Gubser (de Sitter) technology is collected in App. A.
Kinetic theory and initial conditions
We solve the conformal Boltzmann equation in RTA,
where is the Landau velocity and is defined by the energy density, with . The constant is related to the shear viscosity by [37]. Classical kinetic theory requires , i.e. weak coupling; otherwise the system would relax faster than .
At the particles are produced with zero longitudinal momentum in the local frame () and with a transverse-momentum spectrum and density that depend on :
where is a modified Bessel function. The initial energy density is . The order-one constant fixes the normalization of , and is determined below by requiring that the final entropy be simply related to in the hydrodynamic regime. This profile makes the problem exactly solvable: it arises from a thermal distribution in Gubser’s de Sitter coordinates at , which then streams freely (App. A). It is also similar in form to the initial parton distributions of the Color Glass Condensate [62, 65], with playing the role of the saturation scale. With these initial conditions the kinetic equation reduces to an ordinary differential equation in one variable, which we solve with the moment method of Ref. [38] (App. A).
From free streaming to Gubser flow
Consider first a system that hydrodynamizes at a time . The process is a local one and happens independently at each point in the transverse plane. The effective temperature has three evolution stages [61], each with a constant of motion:
(a) For the particles stream freely along the beam axis at each transverse point. The energy per rapidity and area is constant, and the effective temperature falls as .
(b) For the system is a nearly ideal fluid undergoing Bjorken expansion at each transverse point. The entropy per rapidity and area is constant, and with ,
(c) For the radial expansion starts, and the fluid follows Gubser’s ideal solution,
with a dimensionless constant . Here is Gubser’s time variable; at early times and on the surface , so that , with the central temperature defined below.
Our task is to relate the three constants , and . Here and below, the subscript 1 refers to . Similarly, and , written with one argument, refer to .
We will study the central temperature, defined as the temperature at the center of the fireball at ,
In the hydrodynamic regime, where the Gubser solution applies, . The central temperature can be estimated by extrapolating either the free result or the Bjorken flow result to time . This leads us to define two estimates for the central temperature,
These estimates are parametrically valid in the free-streaming () and hydrodynamic () regimes, respectively. In fact, these estimates happen to reproduce the central temperature exactly in the corresponding limits (see below), though in general one would expect a numerical factor of order one. We will see shortly that for , is parametrically smaller than . will serve as our unit of temperature in what follows.
Viscous corrections to ideal Bjorken flow are controlled by the ratio of the relaxation time to the age of the system, . This motivates a definition of the hydrodynamization time,
For , the viscous corrections form a series in ,
The first coefficient, , follows from Navier–Stokes. The second depends on the second-order coefficients through (see (A8)); for RTA, and [66], and so .
For times , the temperature follows the Bjorken attractor [42, 43, 63, 67], an approximately universal function interpolating between free streaming and the Bjorken solution, . This interpolating function has the limits
We fix by requiring that as . Solving the kinetic equations numerically, we find4 . This effectively defines the relation between and ,
At , the two limiting temperatures and differ only by a factor .
The opacity compares the hydrodynamization time to the radius, and equals the number of relaxation times at :
Combining (62) with the definition of gives the unintuitive combination of parameters in5 eq. (51). In terms of , the expansion parameter of (60) is , and the Bjorken series reads
At the boundary of applicability, , the series is organized in powers of , which must match the subsequent evolution.
For , the radial expansion matters. Gubser’s ideal solution is then the leading term, and the viscous corrections form an expansion in [57, 64, 68]. Through second order (App. A),
with the flow of (56). Here is Gubser’s first-order correction, a hypergeometric function [57], and is elementary; both are given in App. A [Eqs. (74) and (75)]. Both vanish at , and, in general, the central temperature is to all orders in the hydrodynamic gradient expansion. At early times, this expansion must reproduce the Bjorken series (60) at every transverse point. For the ideal solution, this works because and have the same profile, and the first- and second-order terms reproduce and . The constant , however, has to be re-matched at each order. Matching at gives
with
Unlike , the coefficient depends only on . The opacity parameter equals only at lowest order in the gradient expansion.
In the opposite limit, , the initial condition streams freely through the radial expansion, giving [37]
Thus, the central temperature in the free regime is parametrically smaller than :
Here we have used (62) and the definition of . Without longitudinal work [69], the free-streaming temperature falls more slowly than in ideal hydrodynamics, as rather than . For the time comes well before , so extrapolating the hydrodynamic behavior back to overestimates the temperature.
Figure 1 shows the central temperature from the kinetic solution as a function of opacity, together with the hydrodynamic expansion (66) and the free-streaming result (69). The kinetic result interpolates smoothly between the two limits. The figure is therefore a complete prediction for the central temperature , and hence for Gubser flow in the hydrodynamic regime, which is completely fixed by . Specifically, given , and , with the initial energy density and , the opacity follows from (51), and the figure gives in units of . The opacity alone thus determines how close the central temperature comes to the hydrodynamic limit.
Density frame and the kinetic solution
In the Density Frame the only evolution variables are the conserved densities and . The Density Frame energy density and velocity are defined by and , with no viscous part in . The viscous stress appears only in the spatial components, , and is fixed at each time by the constitutive relation. It is not an independent variable [23].
To compare with kinetic theory, we construct the full stress tensor from the Density Frame solution and transform it to the Landau frame, as described at the end of Sect. II A [see eq. (13)].
At an initial time , we take from the kinetic solution only the conserved densities and . These are the Density Frame variables, and everything else follows from the constitutive relations. As a result, the Density Frame viscous stress at is not the kinetic one: it takes the value set by the gradients of the initial , while the kinetic stress remembers the earlier evolution. In Israel–Stewart-type theories, is an independent variable and could be initialized from kinetic theory; the Density Frame has no such freedom.
This sets the right way to compare. We fix and increase . The mismatch at then shrinks as grows, and the error of the comparison shrinks with it. If instead were held fixed, the system would always be initialized at the same stage of hydrodynamization, and the error would not decrease. Avoiding this initialization error is the main reason for comparing with the full kinetic solution, which provides a consistent initial state at any .
We initialize at and evolve to on a grid covering . Since , at . Figure 2 compares the two at , both shown in the Landau frame. We note the following. (i) The differences in the energy density (left panel) are small, hard to see in the figure, and of order . (ii) The dashed lines in this panel indicate the central energy density at in ideal hydrodynamics, where , and the corresponding first-order viscous reduction, . This shift reflects the arrows in Figure 1. (iii) The right panel shows the shear stress, which is smaller than the energy density by a factor of order . Its deviation from the kinetic result is of order (i.e. ). We have studied a range of , and all components of the stress tensor along different directions in the transverse plane, with similar results.
Figure 2. Gubser flow with . slices of the energy density (left) and the Landau-frame shear stress (right), in units of , at (dark to light). Thick colored lines: RTA kinetic theory. Thin black lines: Density Frame hydrodynamics, initialized with the kinetic and at and transformed to the Landau frame. Dashed lines (left): the central value at for ideal hydrodynamics, , and for Navier–Stokes, , corresponding to the temperature shift marked by the arrows in Figure 1.
These comparisons also test the approximate foliation independence, and by extension the Lorentz invariance, of the Density Frame. With as the time variable the fluid is at rest, and the Density Frame solution is Gubser’s hydrodynamic solution. Our code instead uses slices, on which the fluid flows radially, and it reproduces the kinetic solution, and hence Gubser’s hydrodynamic solution, up to higher-order gradient corrections.
Comparison to MUSIC for typical heavy-ion initial conditions
We compare the 2+1D Density Frame hydrodynamic evolution with the Israel–Stewart (more specifically, DNMR [18]) implementation in MUSIC [45, 46, 47] using realistic initial conditions for heavy-ion collisions. Both simulations are initialized with the same IP-Glasma event [48, 70] at a proper time of fm and evolved in Milne coordinates. The IP-Glasma + MUSIC framework has been extensively used to model heavy-ion collisions and successfully describe a broad range of experimental observables. For both simulations, we use a QCD equation of state constructed by smoothly matching lattice QCD results at high temperatures [71] to a hadron resonance gas at low temperatures. The shear viscosity-to-entropy density ratio is held constant at , while the bulk viscosity is initially set to zero.
In the second-order viscous hydrodynamic framework employed by MUSIC, the shear-stress tensor and bulk viscous pressure are independent dynamical fields that require explicit initialization. While these quantities were often initialized to zero, the current practice when coupling IP-Glasma to MUSIC is to initialize them by matching the energy-momentum tensor across the transition, thereby ensuring energy-momentum conservation. Specifically, the IP-Glasma energy-momentum tensor is decomposed through Landau matching to extract the local energy density and flow velocity, while its traceless component transverse to the fluid velocity is used to initialize the shear-stress tensor. However, because IP-Glasma is conformal, whereas MUSIC typically uses a nonconformal QCD equation of state, a mismatch arises between their equilibrium pressures. To compensate for this difference while conserving energy and momentum at initialization, the bulk viscous pressure is initialized accordingly, generally resulting in a positive value. Such a positive bulk pressure enhances the effective isotropic pressure and can contribute to the fluid expansion, in contrast to the negative bulk pressure typically associated with an expanding viscous fluid, which acts to resist expansion. There have long been concerns about this bulk pressure initialization, which can have phenomenological implications [72]. In Density Frame hydrodynamics, by contrast, the shear-stress tensor and bulk viscous pressure are not independent dynamical variables and therefore do not require separate initialization. Instead, only the energy density and flow velocity are extracted from IP-Glasma and used as the Density Frame energy density and flow velocity,6 while the viscous contributions are determined by their Navier–Stokes constitutive relations.
An important difference between the two implementations lies in their treatment of low-density regions, where numerical noise can generate large, spurious gradients and lead to instabilities. In MUSIC, these instabilities are controlled by applying regulators in such regions. In contrast, the Density Frame implementation requires only a small energy-density pedestal. At the initial time, cells with energy densities below fm are assigned this minimum value, with no further regulation applied during the subsequent evolution.
Hydrodynamic fields
To facilitate a direct comparison with MUSIC, we transform the Density Frame energy-momentum tensor to the Landau frame at selected times. Figure 3 compares the transverse temperature profiles obtained from Density Frame hydrodynamics and MUSIC at different proper times. Although the overall temperature distributions remain similar, the Density Frame formulation dissipates small-scale structures more rapidly, producing noticeably smoother profiles. This difference is particularly evident in the one-dimensional slices at the Cartesian transverse coordinate , shown in Figure 4. A similar smoothing of small-scale fluctuations is observed in the transverse velocity component , as illustrated in Figure 5 and Figure 6.
Figure 3. Evolution of the temperature () profile for a single IP-Glasma event using Density Frame hydrodynamics (left) and MUSIC (right). The temperature is shown at proper time fm (top), fm (middle) and fm (bottom). Both simulations use the same initial temperatures initialized at fm with shear viscosity and zero bulk viscosity.
Figure 4. Temperature profiles along the transverse slice () obtained with Density Frame hydrodynamics (solid blue) and MUSIC (dashed orange) at proper times , , and fm. Both simulations are initialized with the same IP-Glasma event at fm and have and .
Figure 5. Evolution of the -component of transverse velocity () profile for a single IP-Glasma event using Density Frame hydrodynamics (left) and MUSIC (right). The velocity field is shown at proper time fm (top), 1 fm (middle) and 4 fm (bottom). Both simulations use the same initial velocities initialized at fm with shear viscosity and zero bulk viscosity.
Figure 6. The -component of the transverse velocity along the transverse slice () obtained with Density Frame hydrodynamics (solid blue) and MUSIC (dashed orange) at proper times , , and fm. Both simulations are initialized with the same IP-Glasma event at fm and have and .
Figure 7 compares the shear-viscous stress component , normalized by the maximum pressure in each setup at each proper time, along the transverse slice . We also show MUSIC results with the shear relaxation time reduced by a factor of two, from its default value of to . As with the temperature and velocity profiles, the Density Frame formulation produces a smoother shear-stress profile, with small-scale fluctuations dissipating more rapidly than in MUSIC. Although the hy-
Figure 7. Shear-viscous stress component , normalized by the maximum pressure in each simulation at the corresponding time, along the transverse slice (), obtained with Density Frame hydrodynamics (solid blue) and MUSIC (dashed orange) at proper times , 1.0, 2.0 and 4.0 fm. Results from MUSIC with the shear relaxation time reduced by a factor of two, are also shown (dash-dotted green). Both simulations are initialized with the same IP-Glasma event at fm and have and .
drodynamic evolution begins at fm, differences between the two formulations are already visible at fm. In MUSIC, the shear-stress tensor is initialized using values from IP-Glasma and subsequently evolves according to the Israel–Stewart relaxation equations. In contrast, the Density Frame formulation has no independent viscous degrees of freedom, and the shear stress is determined instantaneously by the local hydrodynamic gradients through the constitutive relations. Consequently, the Density Frame shear stress responds immediately to changes in the hydrodynamic fields, whereas the MUSIC shear stress approaches its first-order constitutive value over a finite relaxation time. Reducing accelerates this relaxation, bringing the MUSIC results closer to those of the Density Frame, which we show in Figure 7.
The difference between the two formulations is even more pronounced in the bulk-viscous sector, as shown in Figure 8. For this comparison, we introduce a temperature-dependent bulk viscosity parametrized in Ref. [73] and show MUSIC results with both the default bulk relaxation time given in [74], and twice its value. As discussed above, in the IP-Glasma + MUSIC framework, the bulk pressure is initialized to account for the mismatch between the equations of state used in the two stages. In the Density Frame, the bulk pressure is determined instantaneously by the local hydrodynamic gradients through the constitutive relations. It therefore takes its first-order hydrodynamic value , which is negative for the expanding system considered here and acts against the expansion, as is generally expected. The MUSIC bulk pressure gradually approaches the Density Frame result over a finite relaxation time, with a larger producing a more delayed response.
Figure 8. Bulk stress , normalized by the maximum pressure in each simulation at the corresponding time, along the transverse slice (), obtained with Density Frame hydrodynamics (solid blue) and MUSIC (dashed orange) at proper times , , , , and fm. Results from MUSIC with twice the bulk relaxation time are also shown (dotted red). Both simulations are initialized with the same IP-Glasma event at fm and have and a temperature dependent .
Observables
Finally, we compare the two hydrodynamic formulations at the level of experimental observables in Pb–Pb collisions at TeV. For this comparison, we simulate 775 IP-Glasma events spanning the 0–50% centrality range, with the hydrodynamic evolution initialized at fm. Both formulations use identical model parameters, taken from Ref. [73], rather than parameters calibrated separately to data. We consider both the temperature-dependent bulk viscosity parametrization employed there and a shear-only
scenario in which the bulk viscosity is set to zero. Figure 9 shows the charged-particle multiplicity , mean transverse momentum , and elliptic flow coefficient as functions of centrality. Despite the differences in the evolution of the local hydrodynamic fields, the two formulations yield broadly similar results, particularly in the absence of bulk viscosity. Including bulk viscosity leads to somewhat larger differences in , consistent with the more pronounced discrepancies in the bulk-pressure evolution discussed above.
Figure 9. Comparison of charged-particle multiplicity (left), mean transverse momentum (middle) and elliptic flow (right) as functions of centrality for Pb-Pb collisions at TeV. Results from Density Frame hydrodynamics (blue) and MUSIC (orange) are shown with shear viscosity . Solid lines correspond to simulations with temperature-dependent bulk viscosity , while dashed lines correspond to zero bulk viscosity.
In more detail, with shear viscosity alone the elliptic flow from the two codes agrees within the statistical errors, while the multiplicity and mean transverse momentum in the Density Frame are lower by about 5% and 2–3%, respectively. These two shifts are not independent, since the mean transverse momentum and the multiplicity are correlated. The non-hydrodynamic modes of MIS therefore have little effect on these observables, consistent with previous Bayesian analyses, which found only weak constraints on the shear relaxation time [7, 8, 75]. With bulk viscosity, in peripheral collisions is about 10% lower in the Density Frame. Even so, the differences between the two codes are generally smaller than the effect of the bulk viscosity itself (compare the solid and dashed curves). It is possible that certain choices of specific bulk viscosity could lead to larger differences between MIS and Density Frame. The bulk viscosity from Ref. [73], used in the present calculations, is relatively broad in temperature. If a narrower peak is used (e.g. Ref. [76]), especially a peak close to the particlization temperature, it leaves less time for MIS to relax to the Navier-Stokes bulk pressure, and difference with Density Frame could be more significant. More studies will be needed to resolve this question.
We emphasize that the two formulations are compared here using identical model parameters, rather than parameters independently calibrated to experimental data. The differences in final state observables between Density Frame and MIS could in principle be reduced further by re-adjusting the model parameters. The current results suggest that when MIS and Density Frame are both calibrated to experimental data, they can have a similar agreement, although with different constraints on the model parameters.
Conclusions and Outlook
We have presented a simulation of viscous hydrodynamics in the Density Frame for heavy-ion collisions in 2+1 dimensions, with a public, documented code [44] that is suitable for a complete Bayesian analysis of heavy-ion data. We also provide a modified version of MUSIC that incorporates the Density Frame hydrodynamic solver as an alternative evolution engine [77]. We compared the Density Frame results with the exact solution of kinetic theory with Gubser symmetry (Sect. III and App. A), for which we presented a new analysis of how the kinetic solution matches onto hydrodynamics. We also compared Density Frame with MIS-based MUSIC for IP-Glasma initial conditions (Sect. IV), which are known to be numerically challenging. The Density Frame simulations are quite robust and readily evolve the IP-Glasma initial conditions: the only regulator is a small minimal energy density, imposed at the initial time. The reason for this stability is that, as in the continuum, the viscous strains follow from variational derivatives of a discrete dissipative function, the positive semi-definite quadratic form that determines the entropy production (Sect. II B). Consequently, the semi-discrete viscous step produces entropy on all length scales.
The comparison with MIS illustrates this. Small-scale structure in the initial conditions is dissipated much more quickly in the Density Frame (see Figure 3 and Figure 5). In MUSIC the viscous stresses at short wavelengths are non-hydrodynamic modes, which relax over the time , so structures smaller than a mean free path survive for a time of order instead of being damped on their own viscous time scale. At later times, however, the two codes give very similar temperature and flow velocity (Figure 4 and Figure 6), and the shear stresses are largely compatible once MUSIC has relaxed (Figure 7). The largest difference is in the bulk-viscous sector. In the Density Frame the bulk pressure adjusts immediately to the underlying flow, while in MIS hydrodynamics it relaxes over the time , which can become long in the crossover region. As a result, the bulk pressure in MIS lags behind that of the density frame (see Figure 8). Since the initialization of the bulk pressure is rather ad-hoc, the bulk evolution of Density Frame provides a valuable scenario to consider, and can help improve theoretical uncertainties on phenomenological constraints of QCD’s bulk viscosity from heavy-ion data.
These differences between Density Frame and MIS have only a modest effect on the final observables (Sect. IV B) in our simulations. Without bulk viscosity the two codes agree closely, and with it the differences are smaller than the effect of the bulk viscosity itself (Figure 9). In general, the difference in observables between MIS and Density Frame will depend on the exact choice of temperature dependence of the specific shear and bulk viscosity, which we will explore in future work.
The Density Frame is causal in the ordinary sense: the response follows the source in time. It is not causal in the relativistic sense, where the response is confined to the forward light cone. Relativistic causality generally requires short-distance modes that enforce the required analytic structure [78, 79, 80]. Our comparisons with kinetic theory and MIS suggest that such modes do not play a significant role in the evolution, at least for the observables considered here. Indeed, the Gubser comparison shows that foliation independence, and hence Lorentz invariance, holds to the expected order in gradients (Sect. III C).
The Density Frame is now ready for a complete Bayesian analysis of heavy-ion data. Since it has no relaxation times or other parameters beyond the viscosities and the equation of state, such an analysis would constrain the transport coefficients of QCD directly. In the future, the same framework can be used to search for non-hydrodynamic modes in small systems such as p+Pb and O+O collisions, to address the theoretical error bars in current Bayesian extractions, and, with noise added through Metropolis steps [22, 34], to study stochastic hydrodynamics in heavy-ion collisions.
Acknowledgments
We thank J. Bhambure and D. Almaalol for their contributions to the initial stages of this work. D.T. is supported by the U.S. Department of Energy, Office of Science, Office of Nuclear Physics, grant No. DE-FG-02-88ER40388. M.S. and J.-F.P. are supported by Vanderbilt University and by the U.S. Department of Energy, Office of Science under Award Number DE-SC-0024347. This research used resources of the National Energy Research Scientific Computing Center (NERSC), a Department of Energy User Facility using NERSC award NP-ERCAP 35438. We acknowledge the use of Claude (Anthropic) and ChatGPT (OpenAI) in preparing this manuscript and in developing analysis codes.
Matching the Gubser and Bjorken flows
This appendix collects known results in one notation. The de Sitter formulation of Gubser flow and its ideal and first-order viscous solutions are due to Gubser [57], with subsequent work by Gubser and Yarom [58]. The exact solution of the RTA Boltzmann equation with this symmetry is due to Denicol, Heinz, Martinez, Noronha and Strickland [36, 37], and we solve it with the moment method of Dash and Roy [38]. Second-order hydrodynamic Gubser flow has been studied in Israel–Stewart theory [68], in second-order conformal hydrodynamics [81, 82, 83], and through the slow-roll expansion [64]. Our closed-form second-order result below is a convenient restatement of these. With the formulas in hand, we find the precise relation between the Gubser solution and the late-time Bjorken attractor through second order in the gradient expansion.
The flow expands both along the beam and radially in Milne coordinates, but it is at rest in another set of coordinates. For a conformal system we may rescale the metric, . In the coordinates , with
the rescaled metric is de Sitter space times a line, . Our corresponds to in the notation of Gubser and most of the subsequent literature [36, 37, 57, 58]. With our initial conditions, everything depends only on . To return to Milne coordinates we use , and . The fluid at rest becomes the flow , , with .
The kinetic problem simplifies in the same way. The distribution function depends on and on two conserved combinations of momenta, , and the relaxation time is . We solve for the moments of Ref. [38]. The first is the energy density, ; the second gives the pressure anisotropy, , and the shear stress, . We keep 160 moments.
The initial state is thermal in de Sitter coordinates at a very early time , f=e^{-p^\hat{}/\hat{T}_S}, and then streams freely. This gives , with and . A short calculation shows that the distribution function approaches the Bessel function form in (53). In practice we start at , and we have checked that our results are insensitive to the precise value.
Hydrodynamics simplifies in the same way. The fluid is at rest in de Sitter coordinates, so energy conservation becomes one equation for [37, 68],
The first term is the ideal expansion and the second is the viscous correction. For the stress we use second-order conformal hydrodynamics [17]. Gubser flow has no vorticity, so only two second-order coefficients appear, and . In a conformal theory both are proportional to , and we write and . For a classical ideal gas with the RTA collision integral adopted here, and [66]. On the static flow, with and , the stress is
The first term is Navier–Stokes. In the bracket, the term comes from the time derivative of , and the term from the terms quadratic in .
We solve (A2) as a series in , where and . As in the text, we write
and require each correction to vanish at . Then is the temperature at at every order. At first order, . Integrating gives Gubser’s result [57],
The second order is simpler than the first [81, 82]. Feeding back into the Navier–Stokes term gives , which cancels the same term from expanding the logarithm in (A2). What is left integrates to powers of :
We checked the result against a numerical solution of (A2).
It remains to fix by matching to the Bjorken regime. At early times at , is large. There grows as up to a constant, , with given in eq. (59), and of only the term grows, so
with
The pieces of the constant term in (A6) cancel, so depends only on . The early-time solution therefore has the form of viscous Bjorken flow,
with . Expanding to second order in at fixed reproduces (A7); there are no cross terms. Since , the prefactor defines the ideal Bjorken temperature at , , and with it . The bracket is therefore the second-order Bjorken series (60), with the coefficients and of (A8). The prefactor depends only on , and by construction . Solving for gives
Here we have re-expanded in the opacity rather than . It is notable that, through second order, the central temperature depends only on , while the early-time Bjorken flow depends only on [17].
References
References
- [1]U. Heinz and R. Snellings, “Collective flow and viscosity in relativistic heavy-ion collisions,” Ann. Rev. Nucl. Part. Sci. 63 (2013) 123–151, arXiv:1301.2826 [nucl-th].DOI
- [2]C. Gale, S. Jeon, and B. Schenke, “Hydrodynamic Modeling of Heavy-Ion Collisions,” Int. J. Mod. Phys. A 28 (2013) 1340011, arXiv:1301.5893 [nucl-th].DOI
- [3]D. A. Teaney, “Viscous Hydrodynamics and the Quark Gluon Plasma,” in Quark-gluon plasma 4, R. C. Hwa and X.-N. Wang, eds., pp. 207–266. World Scientific, 2010. arXiv:0905.2433 [nucl-th].DOI
- [4]W. Busza, K. Rajagopal, and W. van der Schee, “Heavy Ion Collisions: The Big Picture, and the Big Questions,” Ann. Rev. Nucl. Part. Sci. 68 (2018) 339–376, arXiv:1802.04801 [hep-ph].DOI
- [5]U. Heinz and B. Schenke, “Hydrodynamic description of the quark-gluon plasma,” in Quark-Gluon Plasma: 50 Years and Beyond, T. Nayak, M. van Leeuwen, S. Bass, C. Ratti, and J. Dunlop, eds., pp. 165–199. Springer Nature Switzerland, 2026.arxiv.org/abs/2412.19393
- [6]J. E. Bernhard, J. S. Moreland, and S. A. Bass, “Bayesian estimation of the specific shear and bulk viscosity of quark–gluon plasma,” Nature Phys. 15 no. 11, (2019) 1113–1117.DOI
- [7]G. Nijs, W. van der Schee, U. Gürsoy, and R. Snellings, “Transverse Momentum Differential Global Analysis of Heavy-Ion Collisions,” Phys. Rev. Lett. 126 no. 20, (2021) 202301, arXiv:2010.15130 [nucl-th].DOI
- [8]JETSCAPE Collaboration, D. Everett et al., “Multisystem Bayesian constraints on the transport coefficients of QCD matter,” Phys. Rev. C 103 no. 5, (2021) 054904, arXiv:2011.01430 [hep-ph].
- [9]M. R. Heffernan, C. Gale, S. Jeon, and J.-F. Paquet, “Bayesian quantification of strongly interacting matter with color glass condensate initial conditions,” Phys. Rev. C 109 no. 6, (2024) 065207, arXiv:2302.09478 [nucl-th].DOI
- [10]J.-F. Paquet, “Applications of emulation and Bayesian methods in heavy-ion physics,” J. Phys. G 51 no. 10, (2024) 103001, arXiv:2310.17618 [nucl-th].DOI
- [11]CMS Collaboration, V. Khachatryan et al., “Evidence for Collective Multiparticle Correlations in p-Pb Collisions,” Phys. Rev. Lett. 115 no. 1, (2015) 012301, arXiv:1502.05382 [nucl-ex].
- [12]PHENIX Collaboration, A. Adare et al., “Measurements of elliptic and triangular flow in high-multiplicity ³He+Au collisions at √sNN = 200 GeV,” Phys. Rev. Lett. 115 no. 14, (2015) 142301, arXiv:1507.06273 [nucl-ex].
- [13]W. A. Hiscock and L. Lindblom, “Generic instabilities in first-order dissipative relativistic fluid theories,” Phys. Rev. D 31 (1985) 725–733.pubmed.ncbi.nlm.nih.gov/9955752
- [14]I. Müller, “Zum paradoxon der wärmeleitungstheorie,” Zeitschrift für Physik 198 no. 4, (Aug, 1967) 329–344. https://doi.org/10.1007/BF01326412.
- [15]W. Israel, “Nonstationary irreversible thermodynamics: A Causal relativistic theory,” Annals Phys. 100 (1976) 310–331.DOI
- [16]W. Israel and J. M. Stewart, “Transient relativistic thermodynamics and kinetic theory,” Annals Phys. 118 (1979) 341–372.DOI
- [17]R. Baier, P. Romatschke, D. T. Son, A. O. Starinets, and M. A. Stephanov, “Relativistic viscous hydrodynamics, conformal invariance, and holography,” JHEP 04 (2008) 100, arXiv:0712.2451 [hep-th].DOI
- [18]G. S. Denicol, H. Niemi, E. Molnar, and D. H. Rischke, “Derivation of transient relativistic fluid dynamics from the Boltzmann equation,” Phys. Rev. D 85 (2012) 114047, arXiv:1202.4551 [nucl-th]. [Erratum: Phys.Rev.D 91, 039902 (2015)].DOI
- [19]F. S. Bemfica, M. M. Disconzi, and J. Noronha, “Causality and existence of solutions of relativistic viscous fluid dynamics with gravity,” Phys. Rev. D 98 no. 10, (2018) 104064, arXiv:1708.06255 [gr-qc].DOI
- [20]P. Kovtun, “First-order relativistic hydrodynamics is stable,” JHEP 10 (2019) 034, arXiv:1907.08191 [hep-th].DOI
- [21]J. Armas and A. Jain, “Effective field theory for hydrodynamics without boosts,” SciPost Phys. 11 no. 3, (2021) 054, arXiv:2010.15782 [hep-th].DOI
- [22]G. Başar, J. Bhambure, R. Singh, and D. Teaney, “Stochastic relativistic advection diffusion equation from the Metropolis algorithm,” Phys. Rev. C 110 no. 4, (2024) 044903, arXiv:2403.04185 [nucl-th].DOI
- [23]J. Bhambure, R. Singh, and D. Teaney, “Stochastic relativistic viscous hydrodynamics from the Metropolis algorithm,” Phys. Rev. C 111 no. 6, (2025) 064909, arXiv:2412.10306 [nucl-th].arxiv.org/abs/2412.10306
- [24]J. Bhambure, A. Mazeliauskas, J.-F. Paquet, R. Singh, M. Singh, D. Teaney, and F. Zhou, “Relativistic viscous hydrodynamics in the density frame: Numerical tests and comparisons,” Phys. Rev. C 111 no. 6, (2025) 064910, arXiv:2412.10303 [nucl-th].arxiv.org/abs/2412.10303
- [25]L. D. Landau and E. M. Lifshitz, Statistical Physics, Part 1, vol. 5 of Course of Theoretical Physics. Pergamon Press, Oxford, 3rd ed., 1980.
- [26]L. D. Landau and E. M. Lifshitz, Fluid Mechanics: Course of Theoretical Physics, vol. 6. Elsevier, 2013.
- [27]R. F. Fox and G. E. Uhlenbeck, “Contributions to Non-Equilibrium Thermodynamics. I. Theory of Hydrodynamical Fluctuations,” Phys. Fluids 13 (1970) 1893.DOI
- [28]R. F. Fox and G. E. Uhlenbeck, “Contributions to Nonequilibrium Thermodynamics. II. Fluctuation Theory for the Boltzmann Equation,” Phys. Fluids 13 no. 12, (1970) 2881–2890.DOI
- [29]A. Florio, E. Grossi, A. Soloviev, and D. Teaney, “Dynamics of the O(4) critical point in QCD,” Phys. Rev. D 105 no. 5, (2022) 054512, arXiv:2111.03640 [hep-lat].DOI
- [30]A. Florio, E. Grossi, and D. Teaney, “Dynamics of the O(4) critical point in QCD: Critical pions and diffusion in model G,” Phys. Rev. D 109 no. 5, (2024) 054037, arXiv:2306.06887 [hep-lat].arxiv.org/abs/2306.06887
- [31]C. Chattopadhyay, J. Ott, T. Schaefer, and V. Skokov, “Dynamic scaling of order parameter fluctuations in model B,” Phys. Rev. D 108 no. 7, (2023) 074004, arXiv:2304.07279 [nucl-th].arxiv.org/abs/2304.07279
- [32]C. Chattopadhyay, J. Ott, T. Schaefer, and V. V. Skokov, “Simulations of Stochastic Fluid Dynamics near a Critical Point in the Phase Diagram,” Phys. Rev. Lett. 133 no. 3, (2024) 032301, arXiv:2403.10608 [nucl-th].DOI
- [33]C. Chattopadhyay, J. Ott, T. Schaefer, and V. V. Skokov, “Critical fluid dynamics in two and three dimensions,” Phys. Rev. D 111 no. 3, (2025) 034026, arXiv:2411.15994 [nucl-th].arxiv.org/abs/2411.15994
- [34]M. Harhoff, C. Riedel, J. Roth, and S. Schlichting, “Regularization of stochastic fluid dynamics and non-perturbative renormalization of the shear viscosity of incompressible fluids,” arXiv:2610.23442 [hep-ph].
- [35]I. Dzyaloshinskii and G. Volovick, “Poisson brackets in condensed matter physics,” Annals of Physics 125 no. 1, (1980) 67–97. https://www.sciencedirect.com/science/article/pii/0003491680901190.DOI
- [36]G. S. Denicol, U. W. Heinz, M. Martinez, J. Noronha, and M. Strickland, “New Exact Solution of the Relativistic Boltzmann Equation and its Hydrodynamic Limit,” Phys. Rev. Lett. 113 no. 20, (2014) 202301, arXiv:1408.5646 [hep-ph].DOI
- [37]G. S. Denicol, U. W. Heinz, M. Martinez, J. Noronha, and M. Strickland, “Studying the validity of relativistic hydrodynamics with a new exact solution of the Boltzmann equation,” Phys. Rev. D 90 no. 12, (2014) 125026, arXiv:1408.7048 [hep-ph].DOI
- [38]A. Dash and V. Roy, “Hydrodynamic attractors for Gubser flow,” Phys. Lett. B 806 (2020) 135481, arXiv:2001.10756 [nucl-th].DOI
- [39]A. Kurkela, U. A. Wiedemann, and B. Wu, “Flow in AA and pA as an interplay of fluid-like and non-fluid like excitations,” Eur. Phys. J. C 79 no. 11, (2019) 965, arXiv:1905.05139 [hep-ph].DOI
- [40]V. E. Ambrus, S. Schlichting, and C. Werthmann, “Establishing the Range of Applicability of Hydrodynamics in High-Energy Collisions,” Phys. Rev. Lett. 130 no. 15, (2023) 152301, arXiv:2211.14356 [hep-ph].DOI
- [41]V. E. Ambrus, S. Schlichting, and C. Werthmann, “Opacity dependence of transverse flow, preequilibrium, and applicability of hydrodynamics in heavy-ion collisions,” Phys. Rev. D 107 no. 9, (2023) 094013, arXiv:2211.14379 [hep-ph].DOI
- [42]M. P. Heller and M. Spalinski, “Hydrodynamics Beyond the Gradient Expansion: Resurgence and Resummation,” Phys. Rev. Lett. 115 no. 7, (2015) 072501, arXiv:1503.07514.DOI
- [43]G. Giacalone, A. Mazeliauskas, and S. Schlichting, “Hydrodynamic attractors, initial state energy and particle production in relativistic nuclear collisions,” Phys. Rev. Lett. 123 no. 26, (2019) 262301, arXiv:1908.02866 [hep-ph].DOI
- [44]J.-F. Paquet, M. Singh, and D. Teaney, “DFHydro: Density Frame hydrodynamics in 2+1 dimensions.” https://github.com/derek-teaney/DFHydro, 2026. GitHub repository.
- [45]B. Schenke, S. Jeon, and C. Gale, “(3+1)D hydrodynamic simulation of relativistic heavy-ion collisions,” Phys. Rev. C 82 (2010) 014903, arXiv:1004.1408 [hep-ph].DOI
- [46]B. Schenke, S. Jeon, and C. Gale, “Elliptic and triangular flow in event-by-event (3+1)D viscous hydrodynamics,” Phys. Rev. Lett. 106 (2011) 042301, arXiv:1009.3244 [hep-ph].DOI
- [47]J.-F. Paquet, C. Shen, G. S. Denicol, M. Luzum, B. Schenke, S. Jeon, and C. Gale, “Production of photons in relativistic heavy-ion collisions,” Phys. Rev. C 93 no. 4, (2016) 044906, arXiv:1509.06738 [hep-ph].DOI
- [48]B. Schenke, P. Tribedy, and R. Venugopalan, “Fluctuating Glasma initial conditions and flow in heavy ion collisions,” Phys. Rev. Lett. 108 (2012) 252301, arXiv:1202.6646 [nucl-th].DOI
- [49]J. Brewer, A. Mazeliauskas, and W. van der Schee, “Opportunities of OO and pO collisions at the LHC,” in Opportunities of OO and pO collisions at the LHC. 3, 2021. arXiv:2103.01939 [hep-ph].arxiv.org/abs/2103.01939
- [50]U. M. Ascher, S. J. Ruuth, and R. J. Spiteri, “Implicit-explicit runge-kutta methods for time-dependent partial differential equations,” Applied Numerical Mathematics 25 no. 2, (1997) 151–167. https://www.sciencedirect.com/science/article/pii/S0168927497000561. Special Issue on Time Integration.DOI
- [51]L. Pareschi and G. Russo, “Implicit-explicit runge-kutta schemes and applications to hyperbolic systems with relaxation,” Journal of Scientific Computing 25 no. 1, (Oct, 2005) 129–155. https://doi.org/10.1007/s10915-004-4636-4.
- [52]E. M. Constantinescu and A. Sandu, “Extrapolated implicit-explicit time stepping,” SIAM Journal on Scientific Computing 31 no. 6, (2010) 4452–4477.DOI
- [53]A. Kurganov and E. Tadmor, “New High-Resolution Central Schemes for Nonlinear Conservation Laws and Convection–Diffusion Equations,” J. Comput. Phys. 160 (2000) 241–282.DOI
- [54]L. Del Zanna and N. Bucciantini, “An efficient shock-capturing central-type scheme for multidimensional relativistic flows. I. hydrodynamics,” Astron. Astrophys. 390 (2002) 1177, arXiv:astro-ph/0205290.DOI
- [55]S. Balay et al., “PETSc Web page.” https://petsc.org/, 2023. https://petsc.org/.
- [56]L. Trefethen and D. Bau, Numerical Linear Algebra. Other Titles in Applied Mathematics. Society for Industrial and Applied Mathematics, 1997. https://books.google.com/books?id=4Mou5YprRD_kC.DOI
- [57]S. S. Gubser, “Symmetry constraints on generalizations of Bjorken flow,” Phys. Rev. D 82 (2010) 085027, arXiv:1006.0006 [hep-th].DOI
- [58]S. S. Gubser and A. Yarom, “Conformal hydrodynamics in Minkowski and de Sitter spacetimes,” Nucl. Phys. B 846 (2011) 469–511, arXiv:1012.1314 [hep-th].DOI
- [59]C. Chattopadhyay, U. Heinz, S. Pal, and G. Vujanovic, “Higher order and anisotropic hydrodynamics for Bjorken and Gubser flows,” Phys. Rev. C 97 no. 6, (2018) 064909, arXiv:1801.07755 [nucl-th].DOI
- [60]M. Strickland, M. Nopoush, and R. Ryblewski, “Anisotropic hydrodynamics for conformal Gubser flow,” Nucl. Phys. A 956 (2016) 268–271, arXiv:1512.07334.DOI
- [61]M. Arslandok et al., “Hot QCD White Paper,” 3, 2023.arxiv.org/abs/2303.17254
- [62]S. Schlichting and D. Teaney, “The First fm/c of Heavy-Ion Collisions,” Ann. Rev. Nucl. Part. Sci. 69 (2019) 447–476, arXiv:1908.02113 [nucl-th].DOI
- [63]M. Strickland, “The non-equilibrium attractor for kinetic theory in relaxation time approximation,” JHEP 12 (2018) 128, arXiv:1809.01200.DOI
- [64]G. S. Denicol and J. Noronha, “Hydrodynamic attractor and the fate of perturbative expansions in Gubser flow,” Phys. Rev. D 99 no. 11, (2019) 116004, arXiv:1804.04771 [nucl-th].DOI
- [65]F. Gelis, E. Iancu, J. Jalilian-Marian, and R. Venugopalan, “The Color Glass Condensate,” Ann. Rev. Nucl. Part. Sci. 60 (2010) 463–489, arXiv:1002.0333 [hep-ph].DOI
- [66]D. Teaney and L. Yan, “Second order viscous corrections to the harmonic spectrum in heavy ion collisions,” Phys. Rev. C 89 no. 1, (2014) 014901, arXiv:1304.3753 [nucl-th].DOI
- [67]A. Kurkela, A. Mazeliauskas, J.-F. Paquet, S. Schlichting, and D. Teaney, “Effective kinetic description of event-by-event pre-equilibrium dynamics in high-energy heavy-ion collisions,” Phys. Rev. C 99 no. 3, (2019) 034910, arXiv:1805.00961.DOI
- [68]H. Marrochio, J. Noronha, G. S. Denicol, M. Luzum, S. Jeon, and C. Gale, “Solutions of Conformal Israel-Stewart Relativistic Viscous Fluid Dynamics,” Phys. Rev. C 91 no. 1, (2015) 014903, arXiv:1307.6130 [nucl-th].DOI
- [69]P. Danielewicz and M. Gyulassy, “Dissipative Phenomena in Quark Gluon Plasmas,” Phys. Rev. D 31 (1985) 53–62.pubmed.ncbi.nlm.nih.gov/9955508
- [70]B. Schenke, P. Tribedy, and R. Venugopalan, “Event-by-event gluon multiplicity, energy density, and eccentricities in ultrarelativistic heavy-ion collisions,” Phys. Rev. C 86 (2012) 034908, arXiv:1206.6805 [hep-ph].
- [71]HotQCD Collaboration, A. Bazavov et al., “Equation of state in ( 2+1 )-flavor QCD,” Phys. Rev. D 90 (2014) 094503, arXiv:1407.6387 [hep-lat].
- [72]T. Nunes da Silva, D. Chinnellato, M. Hippert, W. Serone, J. Takahashi, G. S. Denicol, M. Luzum, and J. Noronha, “Pre-hydrodynamic evolution and its signatures in final-state heavy-ion observables,” Phys. Rev. C 103 (2021) 054906, arXiv:2006.02324 [nucl-th].DOI
- [73]B. Schenke, C. Shen, and P. Tribedy, “Running the gamut of high energy nuclear collisions,” Phys. Rev. C 102 no. 4, (2020) 044905, arXiv:2005.14682 [nucl-th].DOI
- [74]G. S. Denicol, S. Jeon, and C. Gale, “Transport Coefficients of Bulk Viscous Pressure in the 14-moment approximation,” Phys. Rev. C 90 no. 2, (2014) 024912, arXiv:1403.0962 [nucl-th].DOI
- [75]G. Nijs, W. van der Schee, U. Gürsoy, and R. Snellings, “Bayesian analysis of heavy ion collisions with the heavy ion computational framework Trajectum,” Phys. Rev. C 103 no. 5, (2021) 054909, arXiv:2010.15134 [nucl-th].DOI
- [76]S. Ryu, J. F. Paquet, C. Shen, G. S. Denicol, B. Schenke, S. Jeon, and C. Gale, “Importance of the Bulk Viscosity of QCD in Ultrarelativistic Heavy-Ion Collisions,” Phys. Rev. Lett. 115 no. 13, (2015) 132301, arXiv:1502.01675 [nucl-th].DOI
- [77]J.-F. Paquet, M. Singh, and D. Teaney, “MUSIC with Density Frame Engine.” https://github.com/SinghMayank23/MUSIC_Density_Frame, 2026. GitHub repository.
- [78]M. P. Heller, A. Serantes, M. Spaliński, and B. Withers, “Rigorous Bounds on Transport from Causality,” Phys. Rev. Lett. 130 no. 26, (2023) 261601, arXiv:2212.07434 [hep-th].
- [79]L. Gavassino, M. M. Disconzi, and J. Noronha, “Dispersion Relations Alone Cannot Guarantee Causality,” Phys. Rev. Lett. 132 no. 16, (2024) 162301, arXiv:2307.05987 [hep-th].DOI
- [80]R. E. Hoult and P. Kovtun, “Causality and classical dispersion relations,” Phys. Rev. D 109 no. 4, (2024) 046018, arXiv:2309.11703 [hep-th].DOI
- [81]Y. Hatta, J. Noronha, and B.-W. Xiao, “Exact analytical solutions of second-order conformal hydrodynamics,” Phys. Rev. D 89 no. 5, (2014) 051702, arXiv:1401.6248 [hep-th].DOI
- [82]Y. Hatta, J. Noronha, and B.-W. Xiao, “A systematic study of exact solutions in second-order conformal hydrodynamics,” Phys. Rev. D 89 no. 11, (2014) 114011, arXiv:1403.7693 [hep-th].DOI
- [83]A. Behtash, C. N. Cruz-Camacho, and M. Martinez, “Far-from-equilibrium attractors and nonlinear dynamical systems approach to the Gubser flow,” Phys. Rev. D 97 no. 4, (2018) 044041, arXiv:1711.01745 [hep-th].DOI








