WorksPhysics and Astronomy
Stochastic relativistic viscous hydrodynamics from the Metropolis algorithm
Abstract
We propose an algorithm for simulating stochastic relativistic fluid dynamics based on Metropolis updates. Each step of the algorithm begins with an update based on ideal hydrodynamics. This is followed by proposing random (spatial) momentum transfers between fluid cells, keeping the total energy fixed. These proposals are then accepted or rejected using the change in entropy as a statistical weight. The algorithm reproduces relativistic viscous hydrodynamics in the ``Density Frame", which is a formulation of viscous hydrodynamics we review and clarify here. This formulation is first order in time and requires no auxiliary dynamical fields such as . The only parameters are the shear and bulk viscosities and the equation of state. By adopting the 3+1 split of general relativity, we extend the Metropolis algorithm to general space-time coordinates, such as Bjorken coordinates, which are commonly used to simulate heavy-ion collisions.
Introduction
Physical motivation
Nuclear collisions at the Relativistic Heavy Ion Collider and the Large Hadron Collider exhibit remarkable collective flows which are well described by relativistic viscous hydrodynamics without noise [1]. Current Bayesian fits to the rich phenomenology of hydrodynamic correlations have provided increasingly quantitative constraints on the shear viscosity of QCD and its equation of state [2, 3, 4, 5]. Strikingly, the shear viscosity to entropy ratio is measured to be no more than four times a quantum limit of , which was suggested by gauge-gravity duality [6].
In spite of this success, there are multiple physical motivations for developing stochastic hydrodynamics in the relativistic domain. Indeed, the strength and importance of the noise is proportional to the number of particles in the event , leading to fascinating corrections to hydrodynamics [7]. These corrections lie outside of the usual expansion in the mean-free path to system size and must be computed from hydrodynamic loops or from an appropriate set of hydro-kinetic equations [8, 9]. It is important to quantify these corrections for nucleus-nucleus collisions where the number of produced particles is limited.
In fact, one of the striking findings from the LHC and RHIC is that proton-nucleus and other small colliding systems exhibit collective flow-like correlations, although the number of produced particles in these events is not very large [10, 11]. The stochastic character of the hydrodynamic motion, which is a consequence of the finite number of particles, is paramount for these colliding systems.
There are other motivations to develop stochastic hydrodynamics. For example, there are ongoing searches for critical behavior in nucleus-nucleus collisions both at the LHC and at the fixed target program in STAR [12, 13, 14]. Close to the critical point, modeling stochastic fluctuations is essential to describing the physics. At high temperature and zero baryon density there is an increasing evidence from lattice QCD that QCD is close to an critical point, which describes chiral symmetry breaking [15]. The experimental evidence for the remnants of critical dynamics is limited, but there are suggestive hints in the production of soft pions [16]. At lower temperature and high baryon density, there is developing evidence from the functional Renormalization Group (fRG) [17] and Taylor series expansions in lattice QCD [18, 19, 20] that an Ising-like critical point should exist at a temperature and baryon chemical potential of approximately MeV and . This range of temperatures and chemical potentials can be probed by the STAR fixed target program. However, in this range of collision energies particle production is negligible and the number of particles in the event is limited by the number of nucleons in the incoming nuclei, making this a particularly challenging domain for relativistic fluid dynamics, and a domain where the evolution is decidedly stochastic.
The Metropolis algorithm for relativistic fluid dynamics
Our work is also motivated by recent conceptual and algorithmic advances in the study of dissipative relativistic fluids. Stochastic processes in thermal equilibrium are naturally modeled using a Metropolis-like algorithm which automatically respects the Fluctuation-Dissipation-Theorem (FDT) [21, 22, 23, 24, 25]. The approach was used to simulate the dissipative dynamics of the chiral critical point [26, 27]. Recently, a pioneering paper used the Metropolis updates to simulate the real time dynamics of the Ising critical point in QCD [28]. The methods developed in the current manuscript share the same principles with the algorithms used in these studies. First the system is evolved with a symplectic step of ideal hydrodynamics, and subsequently the conserved charges are randomly transferred between fluid cells. These charge transfers are accepted or rejected using the entropy as a statistical weight to complete the update. Such algorithms are fundamental in statistical mechanics and it would be surprising, and profoundly disconcerting, if they were inapplicable to highly boosted fluids.
The goal of the current work, which is a continuation of our recent study on relativistic advection diffusion equation [29], is to generalize the Metropolis update algorithm to relativistic flows and to general coordinate systems.
First order dissipative fluid dynamics in the relativistic domain as envisioned by the Eckart and Landau and Lifshitz (LL) is known to have generic instabilities [30, 31]. Ultimately these instabilities arise because the divergence of the viscous stress tensor involves second-order spatial derivatives. When the second-order derivatives are placed in a covariant formalism, the equations of motion become second order in time, which leads to runaway solutions and other pathology [30, 32]. Various approaches have been adopted to solve this problem. One approach, as initiated Mueller and Israel and Stewart (MIS) [33, 34, 35], is to add auxiliary dynamical variables to the system of evolution equations. There are many variations of this approach [34, 35, 36, 37, 38], and each variant involves some additional fields, which relax on a collisional timescale so that the system evolves according to the first-order hydrodynamics of Landau and Lifshitz at late times [39]. Indeed, in a linearized analysis the auxiliary variables lead to gapped “non-hydrodynamic” modes, i.e. modes whose frequency remains finite for . Essentially all practical simulations of dissipative relativistic fluids have been based on variants of the MIS equations.
We will adopt a different approach to hydrodynamics that is truly first order in time and has no non-hydrodynamic modes or additional variables. Recognizing that viscosity controls the diffusion of momentum, its seems physically reasonable that randomly transferring momentum between fluid cells, interspersed ideal hydrodynamic time steps, will correctly reproduce the physics of viscosity. In trying to clarify this idea and to find compatibility with Metropolis updates, we found the formulation of hydrodynamic without boosts a clarifying formalism [40, 41, 42]. In particular, we adopted the Density Frame formulation of Armas and Jain [42] (which built upon earlier works [40, 41]) and found that it neatly fits with the Metropolis updates used in statistical mechanics, even for relativistic flows in general coordinates.
In theories without an underlying boost symmetry, hydrodynamics remains valid and is formulated by writing down the fluxes as spatial gradients of the conserved charge densities1. The resulting equations of motion are first order in time and second order in space. If the fluid has Lorentz symmetry, then the coefficients of the gradient expansion are related to each other, but the derivative structure of the equations of motion is unchanged. We have previously investigated the stochastic relativistic advection-diffusion equation in the Density Frame both theoretically and numerically [29]. Our goal here is to generalize this discussion to the Navier-Stokes system, and, in a companion paper, to explore the Density Frame numerically for deterministic hydrodynamic flows in d [43]. The companion paper uses some of the formalism developed here.
The quickest way to derive the equations of motion of the Density Frame is to use lowest-order equations of motion (ideal hydrodynamics) to eliminate time derivatives in the viscous strains. The resulting equations of motion are not Lorentz invariant, but are invariant under Lorentz transformations followed by a change of hydrodynamic frame. This is similar to Heavy Quark Effective Theory in high energy physics, which is only covariant to specified order in the expansion [44]. In essence, each Lorentz observer has his own hydrodynamic frame. Indeed, the Density Frame is a unique hydrodynamic frame where the energy and momentum densities measured on a single spatial slice, and , can be used to reconstruct the temperature and flow velocity of the fluid. In the Landau Frame for instance, one would also need the viscous stress on the slice, while in other approaches at least two spatial slices are needed since the equations of motion are second order in time [45, 46]. The equations of motion in the Density Frame have no additional variables or parameters beyond the shear and bulk viscosities and the equilibrium equation of state.
An outline of the paper is the following. In Sect. II we assemble the equations of motion in the Density Frame. While this has been done already in [42], we feel that most readers will benefit from the orthogonal discussion given here. In Sect. III we show how the non-covariant Density Frame arises naturally as an approximation scheme for covariant kinetic theory. Next in Sect. IV we formulate the stochastic evolution in the Density Frame as a Metropolis update. For simplicity, in Sect. IV A we will first describe the algorithm in 2 + 1 Cartesian dimensions. Briefly, the procedure is the following. One first takes a step of ideal hydrodynamics. Then a proposal is made for random spatial momentum transfers between fluid cells, which keep the energy fixed. The proposed momentum transfers are accepted or rejected using the entropy as a statistical weight. On average this procedure reproduces the mean dissipative stress of the Density Frame, and the fluctuations inherent in the procedure correctly reproduce the stochastic noise in the system. Then in Sect. IV B we show how the algorithm extends to general coordinates with a specified foliation of space-time. The only complication is that the (spatial) momentum transfers must be parallel transported from the cell interfaces to the cell centers.
In the present paper, we have not implemented the Metropolis algorithm numerically. In a companion paper [43], we conducted a first numerical study of the deterministic Density Frame hydrodynamics for 1+1 dimensional flows. These results are encouraging, and, in some respects, show that the Density Frame is numerically more robust than other variants of viscous hydrodynamics. Based on these deterministic numerical results, our previous work on the relativistic stochastic advection-diffusion equation [29], as well as the strong theoretical foundation of the Density Frame developed here and summarized in Sect. V, we hope and expect that the proposed stochastic algorithm will be robust and effective for simulating heavy-ion collisions and other physical systems.
The Density Frame
First we will write down the equations of motion of hydrodynamics in the Density Frame and in the process reproduce the form of the viscous tensor given in Eq. (90) of Armas and Jain [42]. The presentation here is markedly different from [42], and we believe that most readers will benefit from the added discussion. Further, the shear and bulk parts of the viscous tensor are cleanly separated here, and the form of the Density Frame viscous tensor has an evident mathematical structure, which was not clear (to us) in the original work.
Preliminaries
Hydrodynamics is an effective theory for the energy and momentum densities
whose time evolution is given by the conservation laws
In order to close the system of equations, the spatial stress tensor must be specified as a function of and .
This specification is usually implemented with an intermediate set of parameters, , which describe the inverse temperature and four velocity. In ideal hydrodynamics the stress tensor has the functional form2
where and are the pressure and rest frame energy density of the equilibrium equation of state. This equation for means that is determined from the energy and momentum densities,
and subsequently is used to specify the spatial stress, . In the Density Frame the algebraic relations in eq. (4) define to all orders in the derivative expansion.
However, the Density Frame receives viscous corrections of order . Anticipating the next sections, we will state the form of without justification
Here is proportional to the viscosities of the system, and is a tensor formed with , , and speed of sound . is symmetric under interchange of and and interchange of the index pairs, ; its form will ultimately given by eqs. (59) and (63) for the shear and bulk tensor respectively.
Next we will derive the form of entropy production in the Density Frame. The thermodynamics of Density Frame is the same as an ideal fluid with velocity . The entropy and four momentum in a spatial volume are3
and micro-canonical equation of state determines
The Gibbs-Duhem relation follows from extensivity of the system
The Legendre transform of the entropy, , is the logarithm of the partition function, , which is related to the pressure through the Gibbs-Duhem relation
The derivatives of the partition function determine the mean four momentum for a specified
Using the equations of motion and these thermodynamic relations it is easy to show that
Thus the positivity of entropy production forces the matrix (with the rows and columns labeled by the index pairs and ) to be positive semi-definite.
The functional form of is unfamiliar. As we will see it reflects the susceptibilities of a boosted fluid, which will be recorded here for later use. The derivatives of the ideal stress tensor form a generalized susceptibility
and also determine the ideal equations of motion
Here we defined as the spatial projector. The thermodynamic susceptibility, in a strict sense, is and describes the equilibrium fluctuations of the densities . The symmetry of follows directly from the equilibrium partition function
The inverse susceptibility determines the fluctuations of the corresponding Lagrange multipliers and reads
The factor has a simple interpretation. Indeed, the adiabat is defined by lines of constant , and the first term in (17) reflects the derivatives of pressure on the adiabat
This pressure derivative controls the speed of sound waves propagating transverse to the fluid flow.
We have recorded the thermodynamic derivatives of a boosted fluid because they are needed to evaluate the viscous stress tensor in the Density Frame, eq. (7). In the next sections we will derive this tensor by making a change of frame from the covariant Landau Frame.
Hydrodynamic frames
Consider hydrodynamics in a general fluid frame. The full energy-momentum tensor is decomposed into an ideal stress tensor plus viscous corrections
However, the decomposition of the stress into its ideal and viscous pieces is not unique. Indeed, the ideal stress tensor is determined by four parameters , whose precise definition specifies the hydrodynamic frame. The total stress tensor is independent of these intermediate parameters. The constitutive relations provide an approximate expression for in terms of the derivatives of and starts at order .
Now consider a new frame labeled by and
and its relation to our original frame, eq. (20). If is redefined by an amount of order
the total stress tensor is unchanged. This means that and are related by
where
where we have recalled the susceptibility tensor of (14). Since is already first order in derivatives, we may at first order neglect the differences between and when evaluating , leading to
Certain combinations of are invariant under the reparametrization of . Indeed, defining the scalar and tensor projection operators
we see that
after examining the form of . Thus, we define the frame invariant bulk scalar and shear tensors:
Strictly speaking is not a projection operator. However, can play this role, leading to the algebra of projections
Landau Frame
In the Landau Frame, the parameters are chosen such that
As always, the viscous stress is expanded in strains . However, not all of the gradients in this set are independent, since the four ideal equations of motion can be used to express temporal gradients in terms of spatial ones to the same order of accuracy. With these considerations in mind, the Landau Frame stress tensor is written as [47]
where
The two terms in (28) determine the frame invariant shear tensor and bulk scalar .
Density Frame
In the Density Frame, the parameters are chosen such that
where notates the lab frame6. With this choice, the energy per volume and momentum per volume determine the temperature and flow velocity . In all other hydrodynamic frames this data is insufficient to determine . is expanded in the strains . However, using the ideal equations of motion we can rewrite the temporal derivatives appearing in the gradient expansion in terms of spatial derivatives, . Thus, the strains are written as
In the next section, we will write down the explicit form of and identify the frame invariants associated with the shear and bulk viscous tensors.
The Density Frame from the Landau Frame
In this section we will derive the Density Frame stress tensor from the Landau Frame or any other frame where the stress takes the form of (28). Our first task is to rewrite the strains in terms of the spatial strains , using the lowest order equations of motion to replace the time derivatives with the spatial derivatives. The next step is to use the frame transformation rules given in (22) and (23), to zero out the temporal components of the viscous stress, and .
Turning to our first task, the ideal equations of motion are
which can be written in terms of using the susceptibility matrix
The matrices entering in this expression are symmetric, which reflects an integrability constraint following from hydrostatic equilibrium and extensivity [48]. Indeed, symmetry of these matrices can also be seen from the explicit expression for in (14), which is totally symmetric in . Using the symmetry to exchange and in the RHS of eq. (35) and multiplying by the inverse susceptibility leads to
Now using the symmetry of the stress tensor (which is a consequence of relativistic covariance) we find finally
The transformation matrix will be evaluated in explicit form shortly. Examining the expression for the in the Landau Frame in eq. (28), we see that we have completed half of our task – we have expressed the strains in terms of the spatial data . The next half is to make a frame change, using the rules given in (22) and (23) to relate the Landau Frame to the Density Frame. This frame change zeros out the temporal components of the viscous stress in the Landau Frame yielding the Density Frame , where . From (23), we need to choose the frame shift so that
With this choice the temporal components of are zero by construction, while the spatial components are determined by the Landau Frame stress
The frame change can also be neatly rewritten using the transformation matrix given in eq. (37)
Thus we have completed our task of expressing in (33) in terms of in (28)
In the following section the transformation matrix will be evaluated in the explicit form, and the noise matrix will be evaluated.
Explicit evaluation of the viscous stress tensor in the Density Frame
The goal of this section is to determine the viscous stress tensor given in eq. (41) using the transformation matrix defined in eq. (37). Evaluating the relevant derivatives involves differentiating the ideal stress tensor
with respect to and . The algebra is straightforward. It is beneficial to introduce a number of algebraic structures associated with the comoving coordinates to make this algebra transparent.
Comoving coordinates are defined relative to Cartesian-Minkowski coordinates,
The coordinate differentials in the comoving frame are
We note that . The dot product between these differentials acts as an inverse metric in the comoving frame
Straightforward differentiation shows that these objects appear naturally in the transformation matrix
Using the susceptibilities in (17), the pressure derivatives read
Now let us use the scalar projection operator, repeated here for convenience
to decompose into its shear and bulk pieces. The part of that reflects the shear should vanish when contracted with the projector, i.e. . By analyzing which terms in (46) vanish when contracted with , we find the frame transformation matrix can be decomposed as follows
In eq. (46) we are using a compact matrix notation
and using angular brackets to denote the trace
Now when we apply the transformation matrix
we see that first term from (50) projects out the frame invariant shear tensor , while the second term projects out the frame invariant bulk scalar . Multiple identities, such as
enforce the consistency of this decomposition. Further, these relations can be “undone” by applying the projectors to the indices
Notably the shear tensors vanish when contracted with
which clearly identifies the shear component after the transformation to the Density Frame. Finally, we record the viscous stress tensor in the Density Frame
where and are given by
Using the available identities, explicit forms for these tensors can be determined and the shear tensor reads
where
Here the metric is defined in (45), the scalar projector is given in (48), and its trace is notated with angular brackets and evaluated in (52). One can easily verify that
as should be the case for the shear tensor. The bulk tensor is similar and evaluates to
We can see the relationship between the bulk tensor in the Landau and Density Frames by applying the projection operators leading to
Finally, we have checked that results for the shear and bulk tensors presented are consistent with the (rather different) tensor decomposition of Eq. (90) of [42].
Having analyzed the change of frames between the Landau and Density Frames, we will turn to relativistic kinetics and show how the unfamiliar and non-covariant form of the Density Frame constitutive relation follows from a covariant microscopic theory.
The Density Frame from relativistic kinetics
In this section we will show how the Density Frame arises naturally within relativistic kinetic theory. Our goal is to show how the viscous stress tensor of the non-covariant Density Frame is carried by the relativistic constituents in covariant kinetics. At a practical level of simulating heavy ion collisions, we wish to show how the first viscous correction in the Density Frame is related to the commonly used Landau-frame .
Preliminaries
Consider a bosonic system close to an equilibrium state parametrized by . The phase space distribution function is decomposed into an equilibrium distribution plus a viscous correction
where the dependence on the spacetime coordinates is through . Here the equilibrium distribution function is given by
where , and corrects this distribution order by order in gradients . As with the total stress tensor discussed in Sect. II, the total phase space distribution is independent of the intermediate parameters . In a new frame with , the phase space density is unchanged, but is re-parameterized by and
To relate the two hydrodynamic frames, we follow Sect. II and expand in , neglecting the differences between and in , which is already first order. Thus
where
The shift is adjusted to reproduce the frame conditions. It is straightforward to see that integrating over momentum to find the stress tensor reproduces the shift, in (23), where
in kinetic theory.
The Boltzmann equation determines the time evolution of and is given by
where is the non-linear collision operator [49, 50], which vanishes in equilibrium, i.e. for arbitrary . The collision operator is linearized around an equilibrium distribution to determine the first viscous correction. Defining from
the collision operator close to the equilibrium state takes the form [49, 50]
Here indicates the invariant integration over repeated indices, , which defines an inner product between two functions. is a linear positive semi-definite symmetric operator with this inner-product. We note that has a zero mode
since setting is merely a shift in equilibrium, i.e. . The conservation law for linearized fluctuations around equilibrium follows from (71) after exploiting the symmetry of and the zero mode.
To solve for in an approximate way we substitute into (71) and neglect terms second order in derivatives to arrive at an integral equation for
The solution to the integral equation is not unique: if solves (75), then so does for an arbitrary . As discussed below, the frame conditions are used to fixed this ambiguity, leading to a unique solution.
The Landau Frame
In the Landau Frame we use the ideal equations of motion
to eliminate from the strains with the approximation
Substituting into (75) we arrive at an integral equation for
Finally, the Landau Frame condition,
fixes the zero mode ambiguity so that has a unique solution.
In the Landau Frame the final solution to the integral equation takes the form
where and are scalar functions, which are ultimately responsible for the shear and bulk viscosities respectively. Neglecting the bulk viscosity for simplicity (and setting to zero), we determine the viscous tensor by integrating over momenta,
Here the tensor structure from (28) is determined by the following kinetic integrals
which relate the shear viscosity to the solution of the integral equation .
The first viscous correction can be written in terms of the frame invariant shear ,
which is the form normally adopted in practical simulations of heavy ion collisions. We will show that similar form arises in the Density Frame.
The Density Frame
In the Density Frame we use the lowest order equations of motion to eliminate from the strains,
which when substituted into eq. (75) yields the integral equation in the Density Frame
As in the previous section, we will focus on the shear viscosity and neglect the bulk viscosity for simplicity. Motivated by the Landau Frame solution to the integral equation, we write
with adjusted to reproduce the frame condition
This form clearly satisfies the integral equation, differing from (81) by a multiple of the zero mode. Since without the zero mode the first term in (82) integrates to as in (83), we choose as in (38), leading to a solution which satisfies the frame conditions
Next we will evaluate the stress,
to verify that it has the Density Frame form. Using (83) and the intermediate result
following from (70), it is straightforward to see that
as expected.
Finally, we can find a Density Frame equivalent of the Landau result (84), by rewriting (89) using the available identities in (55) as
Here we expressed the viscous correction in terms of the frame invariant shear tensor, , evaluated entirely from the Density Frame stress . We see, as expected, that the viscous correction is the Landau Frame result plus frame shift of .
To summarize, we have shown how the constitutive relation of the non-covariant Density Frame in (7) follows as an approximation to relativistic kinetics. We also have expressed the first viscous correction in the Density Frame in terms of the frame invariant shear tensor which is formed with , eq. (93). This should be compared with the Landau Frame result, eq. (84). In the next section we will study stochastic hydrodynamics in the Density Frame.
Stochastic hydrodynamics in the Density Frame and the Metropolis algorithm
In this section we will show that the stochastic hydrodynamics in the Density Frame is naturally implemented using the Metropolis algorithm. Briefly, one takes a step of ideal hydrodynamics. Then, one proposes spatial momentum transfers between fluid cells, with a variance given by the Density Frame noise kernel . These proposals are accepted or rejected using the entropy as the statistical weight. The procedure produces the dissipative dynamics of stochastic relativistic fluid dynamics.
Cartesian coordinates in 2+1D
For simplicity, we will consider Cartesian coordinates in two dimensions, , and discretize space into finite volume cells of size . The evolution variables are the lab frame energy density and momentum density , and the energy and momentum in a fluid cell
The equations of motion are the conservation laws written in (2), where stress tensor is decomposed into an ideal and viscous pieces. The viscous stress takes the form of an average stress, discussed in previous sections, plus noise
In order for the stochastic process to equilibrate to the probability distribution of the micro-canonical ensemble,
the noise should respect the Fluctuation-Dissipation Theorem (FDT)
Here is the total four momentum of the micro-canonical ensemble.
The form of the noise in the Density Frame can also be found by algebraically manipulating the Landau Frame. In the Landau Frame the viscous tensor and noise are denoted with an underline,
and variance of the noise takes the covariant form
Since the noise in the Density Frame is related to the noise in the Landau Frame by the frame change
it is a simple matter to show that
which reproduces (97) after noting (41). Generating the Landau Frame noise covariantly and applying the transformation matrix as in (100) may be the easiest way to generate the Density Frame noise .
The proposed algorithm for stochastic hydrodynamics uses operator splitting. First, the system is evolved with ideal hydrodynamics over a time ,
and then the system is evolved with a viscous step over the same interval
Next we will show how the Metropolis algorithm can be used to implement the viscous step, producing the correct mean and variance of .
To show this, we integrate the viscous equations of motion (98) over the spatial volume of a fluid cell (cell A with volume ) and over a time interval – see Figure 1(a). The change in the momentum of the cell during the viscous update is determined by the momentum transfers through the walls of the fluid cell
Figure 1. (a) Update of fluid cell A during the Metropolis step. is the area of the right wall of the fluid cell. (b) Proposed momentum transfers when updating cells ABCD due to noise living at the corner of the cell (see text). In three dimensions the reader should visualize a cube with cells A′B′C′D′ above the ones illustrated. In this case the noise is generated at the corner of the cube at the intersection of three planes, .
Here, for example,
is the three-momentum transfer across the wall with area (see figure). There is no change in the energy
The Metropolis procedure consists of picking a corner of the lattice and updating cells in a Metropolis accept-reject step (see Figure 1(b)), for every group of four cells on the lattice. Subsequently, the remaining three corners of cell can be updated in an analogous way.
In detail, we envision the noise living on the corners of the lattice site (see Figure 1(b)), as in our previous study of the advection-diffusion equation [29]. A proposal for the stress is generated with the Density Frame variance
Here and below we notate an average over the proposed noise with . Associated with the chosen corner are two walls8, and . The proposed momentum flux through the corner walls is, for instance,
and the proposed update to the three-momentum for cells and are
with analogous results for neighboring cells and - see Figure 1(b).
The change in entropy resulting from the proposed updates is
Here we expanded to first order in the updates using . The derivative in (114) is a short hand notation for the discrete difference:
In the Metropolis scheme, the proposed updates are accepted if , and are accepted with probability if . Because of this imbalance, the accepted proposals develop a mean value, determining the mean stress . The mean value of the proposals with the accept-reject step is
In passing to the second line we have expanded and used (109). Owing to the symmetry of the proposal distribution under , proposals with occur for exactly half of all the realizations of the noise. Thus the average of with the restriction is half that of the unrestricted distribution in (102), and the mean stress in the Density Frame reads
Given the mean stress, the mean momentum transfer to cell from the first corner follows from (104)
which reproduces exactly half of the update in (106). The remaining half comes from the update of the lower right corner. A complete viscous update would (i) loop through the upper right lattice corners, updating independent groups of four cells associated with this corner as just described; (ii) Then proceed to the remaining corners of cell and repeat the process one at time. To avoid potential bias the order that the corners are updated should be shuffled.
When combined with the ideal hydro step, the complete Metropolis algorithm reproduces stochastic viscous hydrodynamics in the Density Frame, with mean viscous stress and noise given in (95). We followed a similar algorithm when implementing the stochastic advection-diffusion equation numerically in [29].
General Relativity and Bjorken coordinates
The procedure outlined in the previous sections for Cartesian coordinates readily extends to general coordinates and to general relativity. Essentially the only change is that the proposed momentum transfers must be parallel transported from the cell faces to the cell centers before applying the accept-reject criterion. The definitions of energy and momentum are based on the decomposition of the stress tensor according to the fiducial observer associated with the adopted foliation of space time. Since our primary audience are heavy ion physicists, we will first present the algorithm in Bjorken coordinates [51]. Then, in a (somewhat long) appendix, we outline the steps in general coordinates, making full use of the decomposition of spacetime in general relativity.
In Bjorken coordinates, and , the metric is
In comparison to a general coordinate system (see App. A), Bjorken coordinates have a lapse of and a shift of , and the spatial metric at a fixed time-slice is flat
where , with , labels the spatial coordinates. The volume element is .
In Bjorken coordinates, the energy and momentum densities and the energy and momentum in a fluid cell are, respectively,
The conservation laws read
As in previous sections the stress tensor consists of the ideal tensor and the viscous stress, . The average viscous stress for the Density Frame in Bjorken coordinates is the same as Cartesian coordinates after replacing the usual derivatives with covariant derivatives:
Here notates the covariant derivative. The only non-zero component of the Christoffel connection in this expression is
which, in a more general context, reflects the extrinsic curvature of the foliation of space-time (see App. A).
In an operator splitting method, the system is first evolved over a time with ideal hydrodynamics, neglecting the dissipative stress and the noise in eq. (114). In a second step the system is evolved over a time with the Metropolis algorithm. The Metropolis step should reproduce the subsequent viscous dynamics
Integrating over a fluid cell and time interval , we find
where for instance,
is the momentum transfer through the wall. In comparison to the Cartesian case there are two differences. First the energy associated with the adopted foliation, , is not conserved and changes due to the viscous stress. Indeed, this is the viscous contribution to the longitudinal work, which is fundamental to heavy ion phenomenology [52]. Second, the mean stress in (125) involves the covariant derivatives . We will next show how these differences are reproduced by parallel transporting the random momentum transfers between the fluid cells.
Updates and parallel transport
The proposed noise is generated at the corner of the fluid cell A, at the intersection of the planes9 as shown Figure 1. The three-momentum transfer through the surface with normal in the (1) direction is
Here we have introduced the spatial vectors, . The area of the surface is
where the epsilon symbol appropriate to the surface is . We also denote, as the change in spatial coordinate normal to the surface, so that, for instance,
is the spatial volume of a fluid cell. The noise is drawn from the proposed distribution of the Density Frame as in (102).
In the update of cells ABCD and A′B′C′D′ (visualized in Figure 1(b)), the proposals for the momentum transfer between cells are parallel transported from the cell face to the cell center along the appropriate links. For example, the three momentum transfer, , which is transferred to and removed from at their common cell interface, is parallel transported along the path connecting to , leading to the following increments to the four-momentum of the cells and
We have notated the four-momentum increment with a capital letter because the energy transfer during the proposal is non-zero as a result of parallel transport. Specifically, for a Bjorken expansion the energy increments are
although the momentum proposals in (102) at the interfaces are purely spatial. This is ultimately responsible for the shear stress in (127) and the non-conservation of in the viscous step. The proposed four momentum updates in each cell as a result of the three-momentum transfers at the interfaces are for example
with analogous formulas for BCD and A′B′C′D′ – see Figure 1(b).
The corresponding change in the entropy from these momentum transfers
leads to
In this expression is shorthand for the discrete approximation
Using the symmetry of the noise and noting (134), the change in entropy as a result of the momentum transfers is
At this point we may just repeat the discussion surrounding eq. (116) and compute the mean stress between fluid cells with the analogous result
Given the mean stress, the mean update of cell as a result of the Metropolis increments is
As in Cartesian coordinates, this reproduces the expected viscous Bjorken dynamics of the Density Frame given in (118), after each of the eight corners of the fluid cell is visited.
In summary, we have shown how Metropolis updates reproduce the mean viscous stress in the Density Frame, eq. (125). The stochastic nature of the algorithm automatically reproduces the noise. When these stochastic updates are complemented with the symplectic steps of ideal hydrodynamics, the stochastic fluid motion is correctly evolved.
Outlook
In the previous sections we outlined in detail how the Metropolis algorithm and the Density Frame formulation of viscous hydrodynamics can be combined to form a tool for simulating stochastic viscous fluids in heavy ion collisions and general relativity. However, we have stopped short of actually simulating the stochastic dynamics in this work. Clearly this is the next step, and it is a step that is well motivated by the current manuscript, our previous theoretical and numerical work on the stochastic advection-diffusion equation [29], and a companion paper [43], where we simulated the deterministic Density Frame dynamics in dimensions.
From a theoretical perspective the Density Frame is attractive, representing relativistic hydrodynamics in its purest form. Indeed, the Density Frame is the only formulation of relativistic viscous hydrodynamics that has no additional parameters beyond the shear and bulk viscosities and the equation of state. It combines a symplectic step of ideal hydrodynamics with a viscous step that fits nicely into the framework of dissipative stochastic processes (see for example [21, 25]). For this reason the approach is particularly useful for simulating dynamical critical phenomena. Indeed, this paper was inspired by our own simulations of the critical point in QCD for non-expanding fluids at rest [27]. Recently, in a pioneering paper, Chattopadhyay, Ott, Schaefer and Skolkov simulated the real-time dynamics of the liquid-gas critical point in QCD using an approach that is similar to the algorithm suggested here, although again the simulation was limited to non-expanding fluids at rest [28]. Given the outlines of our work and [28], it should be possible to simulate both the and liquid-gas critical points in the relativistic and expanding Bjorken geometry used in heavy ion collisions. We also hope to investigate the fascinating renormalization group properties of two dimensional stochastic fluids [7], which would provide an interesting test of the proposed algorithm.
In summary, this work and a companion paper [43] strongly motivate a Metropolis implementation of the Density Frame description of stochastic relativistic fluids. One can imagine using powerful variants of Metropolis algorithm, such as Hybrid Monte Carlo [23], to simulate the stochastic evolution of the QGP close to its critical points and in small systems.
Acknowledgments
Valuable discussions with Gökçe Başar are gratefully acknowledged. J.B. and D.T. are supported by the U.S. Department of Energy, Office of Science, Office of Nuclear Physics, grant No. DE-FG-02-88ER40388. R.S. is supported partly by a postdoctoral fellowship of West University of Timisoara, Romania.
The Metropolis algorithm in general coordinates
Preliminaries
The formalism of the Density Frame and the Metropolis updates fit naturally within the split of general relativity. The material presented here is standard and we recommend the review article [53]. The space time metric is decomposed as follows
In the foliation of space-time, each slice at fixed coordinate time, , has coordinates . The vector normal to the surface of constant coordinate time is and the vectors tangent to the surface are . In the preferred coordinate system
The dual basis is chosen such that and that
In component form
The spatial projector is
and is used below to decompose vectors and tensors into their spatial and temporal components according to the fiducial observer of the space-time foliation.
The thermal velocity is decomposed as follows
where and note . The four velocity is decomposed as
and thus the three velocity recorded by the fiducial observer is
where is the coordinate velocity.
The stress tensor is decomposed into its temporal and spatial parts
A four momentum vector with spatial and temporal components is notated with a capitol letter and is decomposed as follows
note that .
Covariant derivatives with the full 4-metric of the manifold are notated and with traditional semi-colon notation:
Covariant derivatives with respect to the three metric are written with and with a notation. Thus for three vectors
We will also need the Lie derivative for four dimensional vectors and forms
and for three dimensional vectors and forms :
The distinction between the Lie derivative in three and four dimensions will be clear from the context.
Consider a vector that is purely spatial such as the three-momentum, . When this vector is parallel transported from one point to another on the spatial slice of the foliation, it will not remain spatial. The extrinsic curvature determines the temporal component after parallel transport
The extrinsic curvature can be written as
Covariant conservation laws
We use the decomposition of space-time, and write out the covariant conservation laws , with the decomposition of in (20). The equation of energy conservation reads [53]
where and the momentum conservation equation is
The spatial stress is decomposed into its ideal and viscous pieces
In general coordinates the mean stress involves the covariant derivative
which can be written in several informative forms
The second form expresses the relevant derivatives in terms of the spatial data of the foliation and is the most important in this context. The last form shows that if is a Killing vector of the metric the viscous strains vanish [48]. In the Density Frame only the spatial components of the thermal-metric Lie derivative are used to parameterize the viscous stress.
We again use operator splitting and first evolve ideal hydrodynamics written in (A17) and (A18) for a time , approximating the stress with the ideal stress . This is followed by a stochastic viscous updates, which should evolve
over a time to incorporate the viscous correction. Noting that
we follow the general strategy of flat space and Bjorken coordinates, and integrate the equations of motion over a fluid cell, leading to the update rule for the energy conservation
and momentum conservation
Here, for example,
is a momentum increment to cell resulting from a momentum transfer across the corresponding cell wall , which was subsequently parallel transported to the center of cell . We will briefly describe how the Metropolis algorithm reproduces this dynamics in the next section.
Metropolis dynamics in general coordinates
This section parallels Sect. IV B closely and thus we will be quite brief. The proposed noise follows eq. (132) and eq. (102) with the following replacements and identifications:
The momenta in the fluid cells are updated by parallel transporting the momentum transfers from the cell interfaces back to the cell centers. For example, the three momentum given to and removed from at the cell interface is then parallel transported along the link connecting to as in (135). However, in general coordinates the decomposition into energy and three momentum increments must be generalized:
This yields the increments
which reflect the parallel transport rules given in (A15) and (A16). The complete increment for cell is
Using the thermodynamic derivative
the change in entropy from the updates for cells ABCDA′B′C′D′ is analogous to (139) and yields
From here it is straightforward to apply the discussion in (116) to show that the mean stress has the required form
Given the mean stress, the mean update of cell as a result of the Metropolis increments is
Finally, after updating all cells in the lattice with the first corner, and then repeating the process for the additional seven corners of the fluid cells, the Markov chain reproduces the expected viscous dynamics of the Density Frame given in (A25). When the viscous step is combined with the ideal step outlined in (A17) and (A18), the stochastic viscous fluid is correctly evolved.
References
References
- [1]U. Heinz and R. Snellings, Collective flow and viscosity in relativistic heavy-ion collisions, Ann. Rev. Nucl. Part. Sci. 63, 123 (2013), arXiv:1301.2826 [nucl-th].DOI
- [2]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, 1113 (2019).DOI
- [3]G. Nijs, W. van der Schee, U. Gürsoy, and R. Snellings, Transverse momentum differential global analysis of heavy-ion collisions, Phys. Rev. Lett. 126, 202301 (2021).DOI
- [4]D. Everett et al. (JETSCAPE), Multisystem bayesian constraints on the transport coefficients of qcd matter, Phys. Rev. C 103, 054904 (2021).DOI
- [5]M. Heffernan, Quantification of the Quark–Gluon Plasma with statistical learning, Ph.D. thesis, McGill U. (2023).DOI
- [6]P. Kovtun, D. T. Son, and A. O. Starinets, Viscosity in strongly interacting quantum field theories from black hole physics, Phys. Rev. Lett. 94, 111601 (2005), arXiv:hep-th/0405231.pubmed.ncbi.nlm.nih.gov/15903845
- [7]P. Kovtun, Lectures on hydrodynamic fluctuations in relativistic theories, J. Phys. A 45, 473001 (2012), arXiv:1205.5040 [hep-th].DOI
- [8]Y. Akamatsu, A. Mazeliauskas, and D. Teaney, A kinetic regime of hydrodynamic fluctuations and long time tails for a Bjorken expansion, Phys. Rev. C 95, 014909 (2017), arXiv:1606.07742 [nucl-th].DOI
- [9]X. An, G. Basar, M. Stephanov, and H.-U. Yee, Relativistic Hydrodynamic Fluctuations, Phys. Rev. C 100, 024910 (2019), arXiv:1902.09517 [hep-th].DOI
- [10]A. Adare et al. (PHENIX), Measurements of elliptic and triangular flow in high-multiplicity ³He+Au collisions at √sNN = 200 GeV, Phys. Rev. Lett. 115, 142301 (2015), arXiv:1507.06273 [nucl-ex].DOI
- [11]V. Khachatryan et al. (CMS), Evidence for Collective Multiparticle Correlations in p-Pb Collisions, Phys. Rev. Lett. 115, 012301 (2015), arXiv:1502.05382 [nucl-ex].
- [12]ALICE (ALICE), Letter of intent for ALICE 3: A next generation heavy-ion experiment at the LHC, Tech. Rep. (CERN, Geneva, 2022) 202 pages, 103 captioned figures, 19 tables, arXiv:2211.02491.arxiv.org/abs/2211.02491
- [13]Expression of Interest for an ALICE ITS Upgrade in LS3 (2018).
- [14]L. Du, A. Sorensen, and M. Stephanov, The QCD phase diagram and Beam Energy Scan physics: a theory overview, Int. J. Mod. Phys. E 33, 2430008 (2024), arXiv:2402.10183 [nucl-th].DOI
- [15]O. Kaczmarek, F. Karsch, A. Lahiri, L. Mazur, and C. Schmidt, QCD phase transition in the chiral limit (2020) arXiv:2003.07920 [hep-lat].arxiv.org/abs/2003.07920
- [16]E. Grossi, A. Soloviev, D. Teaney, and F. Yan, Soft pions and transport near the chiral critical point, Phys. Rev. D 104, 034025 (2021), arXiv:2101.10847 [nucl-th].DOI
- [17]W.-j. Fu, J. M. Pawlowski, and F. Rennecke, QCD phase structure at finite temperature and density, Phys. Rev. D 101, 054032 (2020), arXiv:1909.02991 [hep-ph].DOI
- [18]S. Borsanyi, Z. Fodor, J. N. Guenther, R. Kara, S. D. Katz, P. Parotto, A. Pasztor, C. Ratti, and K. K. Szabo, QCD Crossover at Finite Chemical Potential from Lattice Simulations, Phys. Rev. Lett. 125, 052001 (2020), arXiv:2002.02821 [hep-lat].
- [19]A. Bazavov et al. (HotQCD), Chiral crossover in QCD at zero and non-zero chemical potentials, Phys. Lett. B 795, 15 (2019), arXiv:1812.08235 [hep-lat].DOI
- [20]D. A. Clarke, P. Dimopoulos, F. Di Renzo, J. Goswami, C. Schmidt, S. Singh, and K. Zambello, Searching for the QCD critical point using Lee-Yang edge singularities, PoS LATTICE2023, 168 (2024), arXiv:2401.08820 [hep-lat].
- [21]R. Fox and G. Uhlenbeck, Contributions to non-equilibrium thermodynamics. i. theory of hydrodynamical fluctuations, Physics of Fluids 13 (1970).
- [22]P. J. Rossky, J. D. Doll, and H. L. Friedman, Brownian dynamics as smart Monte Carlo simulation, The Journal of Chemical Physics 69, 4628 (1978).DOI
- [23]S. Duane, A. Kennedy, B. J. Pendleton, and D. Roweth, Hybrid monte carlo, Physics Letters B 195, 216 (1987).DOI
- [24]G. D. Moore, The Sphaleron rate: Bodeker’s leading log, Nucl. Phys. B 568, 367 (2000), arXiv:hep-ph/9810313.DOI
- [25]M. Grmela and H. C. Öttinger, Dynamics and thermodynamics of complex fluids. i. development of a general formalism, Phys. Rev. E 56, 6620 (1997).
- [26]A. Florio, E. Grossi, A. Soloviev, and D. Teaney, Dynamics of the O(4) critical point in QCD, Phys. Rev. D 105, 054512 (2022), arXiv:2111.03640 [hep-lat].DOI
- [27]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, 054037 (2024), arXiv:2306.06887 [hep-lat].
- [28]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, 032301 (2024), arXiv:2403.10608 [nucl-th].
- [29]G. Başar, J. Bhambure, R. Singh, and D. Teaney, Stochastic relativistic advection diffusion equation from the Metropolis algorithm, Phys. Rev. C 110, 044903 (2024), arXiv:2403.04185 [nucl-th].DOI
- [30]W. A. Hiscock and L. Lindblom, Generic instabilities in first-order dissipative relativistic fluid theories, Phys. Rev. D 31, 725 (1985).pubmed.ncbi.nlm.nih.gov/9955752
- [31]L. Gavassino, Bounds on transport from hydrodynamic stability, Physics Letters B 840, 137854 (2023).DOI
- [32]L. Gavassino, Can We Make Sense of Dissipation without Causality?, Phys. Rev. X 12, 041001 (2022), arXiv:2111.05254 [gr-qc].DOI
- [33]I. Müller, Zum paradoxon der wärmeleitungstheorie, Zeitschrift für Physik 198, 329 (1967).
- [34]W. Israel, Nonstationary irreversible thermodynamics: A Causal relativistic theory, Annals Phys. 100, 310 (1976).DOI
- [35]W. Israel and J. M. Stewart, Transient relativistic thermodynamics and kinetic theory, Annals Phys. 118, 341 (1979).DOI
- [36]R. P. Geroch and L. Lindblom, Dissipative relativistic fluid theories of divergence type, Phys. Rev. D 41, 1855 (1990).pubmed.ncbi.nlm.nih.gov/10012554
- [37]H. C. Öttinger, Relativistic and nonrelativistic description of fluids with anisotropic heat conduction, Physica A: Statistical Mechanics and its Applications 254, 433 (1998).DOI
- [38]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, 114047 (2012), [Erratum: Phys.Rev.D 91, 039902 (2015)], arXiv:1202.4551 [nucl-th].DOI
- [39]L. Lindblom, The Relaxation effect in dissipative relativistic fluid theories, Annals Phys. 247, 1 (1996), arXiv:gr-qc/9508058.DOI
- [40]I. Novak, J. Sonner, and B. Withers, Hydrodynamics without boosts, JHEP 07, 165, arXiv:1911.02578 [hep-th].DOI
- [41]J. de Boer, J. Hartong, E. Have, N. A. Obers, and W. Sybesma, Non-Boost Invariant Fluid Dynamics, SciPost Phys. 9, 018 (2020), arXiv:2004.10759 [hep-th].DOI
- [42]J. Armas and A. Jain, Effective field theory for hydrodynamics without boosts, SciPost Phys. 11, 054 (2021), arXiv:2010.15782 [hep-th].DOI
- [43]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, (2024), arXiv:2412.10303 [nucl-th].arxiv.org/abs/2412.10303
- [44]A. Manohar and M. Wise, Heavy Quark Physics, Cambridge Monographs on Particle Physics, Nuclear Physics and Cosmology (Cambridge University Press, 2000).DOI
- [45]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, 104064 (2018), arXiv:1708.06255 [gr-qc].DOI
- [46]P. Kovtun, First-order relativistic hydrodynamics is stable, JHEP 10, 034, arXiv:1907.08191 [hep-th].DOI
- [47]L. D. Landau and E. M. Lifshitz, Fluid Mechanics: Course of Theoretical Physics, Vol. 6 (Elsevier, 2013).
- [48]K. Jensen, M. Kaminski, P. Kovtun, R. Meyer, A. Ritz, and A. Yarom, Towards hydrodynamics without an entropy current, Phys. Rev. Lett. 109, 101601 (2012), arXiv:1203.3556 [hep-th].DOI
- [49]R. F. Fox and G. E. Uhlenbeck, Contributions to nonequilibrium thermodynamics. ii. fluctuation theory for the boltzmann equation, The Physics of Fluids 13, 2881 (1970), https://pubs.aip.org/aip/pfl/article-pdf/13/12/2881/12261085/2881_1_online.pdf.DOI
- [50]P. B. Arnold, G. D. Moore, and L. G. Yaffe, Transport coefficients in high temperature gauge theories. 1. Leading log results, JHEP 11, 001, arXiv:hep-ph/0010177.
- [51]J. D. Bjorken, Highly Relativistic Nucleus-Nucleus Collisions: The Central Rapidity Region, Phys. Rev. D 27, 140 (1983).DOI
- [52]P. Danielewicz and M. Gyulassy, Dissipative Phenomena in Quark Gluon Plasmas, Phys. Rev. D 31, 53 (1985).pubmed.ncbi.nlm.nih.gov/9955508
- [53]E. Gourgoulhon, 3+1 Formalism in General Relativity, Lecture Notes in Physics (Springer, 2012) arXiv:gr-qc/0703035 [gr-qc].