Hydrodynamic Attractor: Universal Dynamics

• Hydrodynamic Attractor is a universal far-from-equilibrium solution that draws varied initial conditions onto a common slow manifold governed by viscous dynamics.
• It embodies dynamical dimensionality reduction, linking early free-streaming behavior with late-time hydrodynamics without necessitating local thermal equilibrium.
• The concept underpins resummation methods for divergent gradient expansions and applies broadly—from conformal Bjorken flow in heavy-ion collisions to driven ultracold atomic gases.

A hydrodynamic attractor is a distinguished, universal far-from-equilibrium solution toward which a wide class of initial conditions rapidly converge when the dynamics is expressed in suitable scaling variables. In kinetic theory, viscous hydrodynamics, and holography, it appears as a slow invariant manifold or universal curve that organizes evolution beyond the domain of validity of a truncated gradient expansion, while nonhydrodynamic modes decay into the hydrodynamic sector. In the settings most studied in high-energy theory, especially Bjorken flow, the attractor links the early large-gradient regime to late-time viscous hydrodynamics and explains hydrodynamization without requiring local equilibrium or isotropization (Strickland et al., 2017, Jankowski et al., 2023, Heller et al., 2020, Cartwright et al., 2022).

1. Conceptual structure and physical interpretation

The attractor is defined operationally by universal collapse: trajectories generated from widely different initial data approach a unique non-perturbative solution after transients decay. In phase-space language, it is a concrete instance of dynamical dimensionality reduction: a cloud of initial conditions contracts onto a low-dimensional slow manifold and subsequently evolves along it. In this picture, the attractor need not be specified a priori through a preferred variable; it can be identified through contraction of the phase-space cloud, a hierarchy of principal directions, and alignment with a slow region where the phase-space speed is minimal (Heller et al., 2020).

A central distinction in the literature is that hydrodynamization is not the same as equilibration. The attractor describes when evolution is already controlled by hydrodynamic degrees of freedom, even though pressure anisotropy can remain sizable and local thermodynamic relations need not yet hold. In holographic Bjorken flow, reaching an attractor does not imply or require local thermal equilibrium, although it does imply that the evolution is approximately hydrodynamic in the cases studied. This distinction is also central in heavy-ion phenomenology, where hydrodynamic modeling is often initialized at times when the system is still far from isotropy (Cartwright et al., 2022).

The dominant diagnostic variables depend on symmetry and microscopic theory. In conformal Bjorken flow, common choices are the scaled time $w=\tau T(\tau)$, pressure-anisotropy observables such as $\mathcal{A}(w)$ or $P_L/P_T$, and normalized dissipative stresses such as $\bar{\pi}$ or $\Pi/\varepsilon$. In broader settings, the attractor may instead be a dispersion relation in momentum space, a manifold in a high-dimensional moment space, or a cyclic orbit under periodic driving. The unifying feature is universality after rapid loss of detailed initial-state information.

2. Bjorken-flow archetype in conformal kinetic theory and viscous hydrodynamics

The most developed formulation is the $0{+}1$d conformal Bjorken expansion. In this setting the microscopic dynamics can be taken to be the Boltzmann equation in the relaxation-time approximation,

$$p^\mu \partial_\mu f(x,p) = -\,\frac{p\cdot u(x)}{\tau_{\rm eq}(x)}\left[f(x,p) - f_{\rm eq}\!\left(\frac{p\cdot u}{T(x)}\right)\right],$$

with boost-invariant flow $u^\mu=(\cosh\varsigma,0,0,\sinh\varsigma)$ and proper time $\tau=\sqrt{t^2-z^2}$. Energy conservation reduces to

$$\tau\,\frac{d\log\varepsilon}{d\tau} = -\frac{4}{3} + \frac{\Pi}{\varepsilon},$$

and a convenient dimensionless parametrization is

$\mathcal{A}(w)$0

The pressure anisotropy is then

$\mathcal{A}(w)$1

In exact RTA kinetic theory, the attractor is obtained by taking $\mathcal{A}(w)$2 and an infinitely oblate initial condition $\mathcal{A}(w)$3, which selects the universal curve $\mathcal{A}(w)$4 with $\mathcal{A}(w)$5. The early-time limit corresponds to free streaming and the late-time limit recovers Navier–Stokes. Within anisotropic hydrodynamics, the attractor equation resums an infinite series in the inverse Reynolds number through the dependence on the anisotropy parameter $\mathcal{A}(w)$6, and the resulting resummed attractor is virtually indistinguishable from the exact RTA attractor, with maximum discrepancy $\mathcal{A}(w)$7 over the displayed range. The same formulation guarantees positive longitudinal and transverse pressures, whereas MIS and DNMR can drive $\mathcal{A}(w)$8 negative at very early times. DNMR improves significantly over MIS because it uses the correct RTA value $\mathcal{A}(w)$9, but it still lacks the all-order inverse-Reynolds resummation built into aHydro. The same work also introduced an expansion solely in inverse Reynolds number, carried to third order, which reproduces the fully resummed aHydro attractor very well. Numerical evolutions with varied initial $P_L/P_T$0 converge to the aHydro and DNMR attractors by $P_L/P_T$1; for LHC-like parameters $P_L/P_T$2 MeV at $P_L/P_T$3 fm/c and $P_L/P_T$4, this corresponds to $P_L/P_T$5 fm/c near the center of the fireball, while the isotropization criterion $P_L/P_T$6 is reached only for $P_L/P_T$7 (Strickland et al., 2017).

3. Gradient expansions, transseries, Borel analysis, and convergence

A major theme in attractor theory is the status of the hydrodynamic gradient expansion. In conformal Bjorken flow, the late-time expansion takes the form

$P_L/P_T$8

with factorially growing coefficients and vanishing radius of convergence. In MIS/BRSSS the asymptotics is governed by a singulant $P_L/P_T$9, and the full solution requires a transseries with exponentially suppressed sectors,

$\bar{\pi}$0

This structure makes explicit that the attractor is not identical to any finite truncation of the gradient series, but rather to its non-perturbative completion (Jankowski et al., 2023).

In strongly coupled $\bar{\pi}$1 SYM under Bjorken flow, the first 240 coefficients of the gradient expansion were Borel summed and compared to numerical AdS/CFT simulations. The Borel-summed result provides an accurate and unambiguous approximation to the hydrodynamic attractor, reliable down to $\bar{\pi}$2, whereas first-order hydrodynamics becomes accurate only around $\bar{\pi}$3. The Borel-plane singularities lie off the positive real axis and are tied to complex quasinormal modes, so the real-axis Borel-Laplace integral is unambiguous in this case (Spaliński, 2017).

The analytic structure can change sharply with the type of expansion. For the spatial hydrodynamic attractor in a one-dimensional BGK model, the spatial Chapman–Enskog series

$\bar{\pi}$4

is factorially divergent in the non-relativistic case but strictly Borel summable, with the relevant Borel singularity on the negative real axis. When relativistic causality is imposed, so that velocities are bounded, the spatial series becomes convergent with a finite radius of convergence. This sharply contrasts with the temporal gradient expansion, which is non-Borel summable along the positive real axis and requires transseries completion (Kooshkbaghi, 23 Mar 2026).

Hubble expansions provide another controlled comparison. In both MIS theory and RTA kinetic theory with power-law Hubble expansion, the gradient and Chapman–Enskog series are factorially divergent, and median-Borel resummation yields globally attractive hydrodynamic attractors. In those systems only the first-order nonhydrodynamic mode exists, but the speed of attraction differs: in MIS, $\bar{\pi}$5 decays exponentially while $\bar{\pi}$6 decays as a power law, whereas in RTA the approach is exponential for both (Du et al., 2021).

The claim that hydrodynamic series always diverge is therefore too strong. For an ultrarelativistic gas of hard spheres undergoing Bjorken expansion with particle-number conservation, the Knudsen number is constant, the shear-sector constitutive relation admits an exact time-independent attractor, and the corresponding gradient expansion converges within a finite radius. This was identified as the first example of a convergent hydrodynamic gradient series in a rapidly expanding relativistic system, and the hydrodynamic and Boltzmann attractors differ by at most $\bar{\pi}$7 even at large gradients (Denicol et al., 2019).

4. Extensions beyond conformal Bjorken flow

The attractor concept generalizes, but not uniformly, once conformality or simple symmetry assumptions are relaxed. In non-conformal Bjorken flow with shear and bulk viscosities, the relative decay rate of the enthalpy density, the inverse shear Reynolds number, and the inverse bulk Reynolds number all exhibit late-time universal patterns toward an ideal-fluid description. However, an early-time attractor is generically absent because the coupled shear-bulk sector contains a positive-valued eigenmode associated with free-streaming dynamics. Two remedies were identified: an approximate cancellation of the unstable mode through a shear-bulk mixture, and an exact restoration of early-time universality by enhancing the quadratic bulk–expansion coupling $\bar{\pi}$8 by at least a factor of two through the transport coefficient $\bar{\pi}$9 (Chen et al., 2021).

For Gubser flow, which combines longitudinal and transverse expansion in a conformal system mapped to $\Pi/\varepsilon$0, the attractor interpolates between an early-time free-streaming fixed point, an intermediate hydrodynamic regime near de Sitter time $\Pi/\varepsilon$1, and late-time free streaming. In the moment hierarchy of the RTA Boltzmann equation, the relevant fixed points are characterized by $\Pi/\varepsilon$2 and $\Pi/\varepsilon$3, and the attractor solution is the universal trajectory connecting them. Moderate-order truncations of the hierarchy reproduce the exact kinetic solution with high accuracy (Dash et al., 2020).

Breaking boost invariance while retaining longitudinal dynamics changes the appropriate scaling variables. In $\Pi/\varepsilon$4d viscous hydrodynamics with general rapidity distribution, universality is recovered by replacing the global Bjorken timescale by the local expansion-rate timescale. The Lorentz-invariant variables

$\Pi/\varepsilon$5

reduce to the Bjorken variables when $\Pi/\varepsilon$6, and the solutions follow the same first-order ODE as in $\Pi/\varepsilon$7d after the replacements $\Pi/\varepsilon$8 and $\Pi/\varepsilon$9. The same study found that a rapid expansion in the fluid velocity is essential for a rapid early-time attractor (Chen et al., 2024).

Additional generalizations enlarge the structure of the attractor itself. In superfluid Bjorken flow, a MIS formalism incorporating the Goldstone boson and the condensate reveals both the conventional hydrodynamic attractor on the unbroken-symmetry surface and an even number of non-dissipative symmetry-breaking fixed points. For super-critical initial temperature, the condensate becomes exponentially small very rapidly and the system is trapped by the conventional attractor for a long intermediate time before reheating and switching to one of the symmetry-breaking fixed points; those fixed points are unstable against inhomogeneous perturbations and should lead to spinodal decomposition (Mitra et al., 2020). In a hybrid viscous fluid with two interacting sectors of different viscosities, the attractor becomes a two-dimensional surface rather than a one-dimensional curve, and the total system behaves at late times like a single viscous fluid with a dynamically determined effective shear viscosity (Mitra et al., 2020). In the hyperbolic slicing of $0{+}1$0, typical solutions also approach a hydrodynamic attractor rapidly at late times despite a Knudsen number exceeding unity, while the inverse Reynolds number vanishes; this suggests that the inverse Reynolds number captures hydrodynamization more faithfully in that geometry (Soloviev, 30 Oct 2025).

5. Quantifying universality, decay rates, and theory-to-theory matching

A quantitative description of attractor formation has been developed in phase space. For non-autonomous Bjorken systems one can embed the dynamics into state vectors such as $0{+}1$1, $0{+}1$2, or high-dimensional moment spaces, construct the covariance matrix on fixed-time slices, and analyze the explained variance ratios and participation ratio. In MIS/BRSSS, the phase-space cloud collapses rapidly from a two-dimensional region onto an effectively one-dimensional curve; in HJSW the collapse proceeds through an intermediate oscillatory stage because of oscillatory nonhydrodynamic modes; and in a 16-dimensional truncation of RTA kinetic theory the subdominant principal components begin to decay exponentially at late times, with decay rates consistent with twice the nonhydrodynamic decay rate because PCA measures variances rather than amplitudes (Heller et al., 2020).

The same logic was extended to Yang–Mills effective kinetic theory with Color Glass Condensate initial conditions. There, the late-time attractor is characterized by a single principal component determining the overall energy scale, while a single subleading component governs deviations of the pressure anisotropy, the screening mass, and the scattering rate. The variance of the pressure anisotropy decays exponentially in the scaled time $0{+}1$3 with a rate close to the conformal RTA prediction $0{+}1$4, and the coupling dependence maps onto a simple dependence on $0{+}1$5 (Du et al., 2022).

Another line of work compares attractors across hydrodynamic theories of different order. In conformal Bjorken flow, second-order theories (MIS, DNMR), third-order theories (such as PJP), and transient higher-order theories with coupled rank-four moments (BD, YJG) all possess attractor solutions. A slow-roll expansion shows that the dominant correction beyond Navier–Stokes is the $0{+}1$6 term, which depends only on second-order transport coefficients. Numerically, the DNMR, PJP, BD, and YJG attractors coincide by $0{+}1$7, MIS joins by $0{+}1$8, and all theories approach the Navier–Stokes limit by $0{+}1$9. This identifies an intermediate regime in which higher-order attractors converge to the same solution as an appropriately parameterized second-order theory (Chen et al., 10 Sep 2025).

Holography provides further diagnostics beyond the usual pressure anisotropy. In $$p^\mu \partial_\mu f(x,p) = -\,\frac{p\cdot u(x)}{\tau_{\rm eq}(x)}\left[f(x,p) - f_{\rm eq}\!\left(\frac{p\cdot u}{T(x)}\right)\right],$$0 SYM undergoing Bjorken expansion, one can define far-from-equilibrium longitudinal and transverse speeds of sound through derivatives at fixed proper time,

$$p^\mu \partial_\mu f(x,p) = -\,\frac{p\cdot u(x)}{\tau_{\rm eq}(x)}\left[f(x,p) - f_{\rm eq}\!\left(\frac{p\cdot u}{T(x)}\right)\right],$$1

Borel resummation reveals two distinct attractors, one longitudinal and one transverse, and the corresponding attractor times satisfy a $$p^\mu \partial_\mu f(x,p) = -\,\frac{p\cdot u(x)}{\tau_{\rm eq}(x)}\left[f(x,p) - f_{\rm eq}\!\left(\frac{p\cdot u}{T(x)}\right)\right],$$2 criterion at approximately $$p^\mu \partial_\mu f(x,p) = -\,\frac{p\cdot u(x)}{\tau_{\rm eq}(x)}\left[f(x,p) - f_{\rm eq}\!\left(\frac{p\cdot u}{T(x)}\right)\right],$$3 for the longitudinal sector and $$p^\mu \partial_\mu f(x,p) = -\,\frac{p\cdot u(x)}{\tau_{\rm eq}(x)}\left[f(x,p) - f_{\rm eq}\!\left(\frac{p\cdot u}{T(x)}\right)\right],$$4 for the transverse sector. The same study emphasized that reaching these attractors does not imply local thermal equilibrium (Cartwright et al., 2022).

6. Cold-atom realizations, cyclic attractors, and fluctuation effects

Hydrodynamic attractors are not restricted to relativistic plasmas. In a uniform two-component Fermi gas with contact interactions, a time-dependent $$p^\mu \partial_\mu f(x,p) = -\,\frac{p\cdot u(x)}{\tau_{\rm eq}(x)}\left[f(x,p) - f_{\rm eq}\!\left(\frac{p\cdot u}{T(x)}\right)\right],$$5-wave scattering length produces the same bulk hydrodynamic forcing as an isotropic expansion. The effective bulk strain rate is

$$p^\mu \partial_\mu f(x,p) = -\,\frac{p\cdot u(x)}{\tau_{\rm eq}(x)}\left[f(x,p) - f_{\rm eq}\!\left(\frac{p\cdot u}{T(x)}\right)\right],$$6

and the bulk sector obeys a nonrelativistic MIS-type relaxation equation,

$$p^\mu \partial_\mu f(x,p) = -\,\frac{p\cdot u(x)}{\tau_{\rm eq}(x)}\left[f(x,p) - f_{\rm eq}\!\left(\frac{p\cdot u}{T(x)}\right)\right],$$7

For a power-law ramp of the inverse scattering length toward unitarity, the dynamics can be solved analytically and the attractor is

$$p^\mu \partial_\mu f(x,p) = -\,\frac{p\cdot u(x)}{\tau_{\rm eq}(x)}\left[f(x,p) - f_{\rm eq}\!\left(\frac{p\cdot u}{T(x)}\right)\right],$$8

The corresponding gradient expansion is divergent, but its Borel sum equals the attractor. This furnishes a complete analytic realization of universal equilibration in an ultracold-atom platform (Fujii et al., 2024).

A related nearly-unitary Fermi-gas analysis isolates two stages that parallel heavy-ion attractor phenomenology. Stage E is an expansion-driven early-time attractor generated by the externally imposed drive $$p^\mu \partial_\mu f(x,p) = -\,\frac{p\cdot u(x)}{\tau_{\rm eq}(x)}\left[f(x,p) - f_{\rm eq}\!\left(\frac{p\cdot u}{T(x)}\right)\right],$$9; Stage T is the exponential decay of nonhydrodynamic transients with relaxation time $u^\mu=(\cosh\varsigma,0,0,\sinh\varsigma)$0; and Stage H is the late-time viscous hydrodynamic regime. The bulk pressure is extracted through Tan’s relation,

$u^\mu=(\cosh\varsigma,0,0,\sinh\varsigma)$1

and the analysis shows phase-space flattening toward a quasi-one-dimensional manifold in $u^\mu=(\cosh\varsigma,0,0,\sinh\varsigma)$2 (Heller et al., 3 Jul 2025).

Periodic driving opens a qualitatively different possibility. In ultracold quantum gases with oscillating isotropic expansion, a novel cyclic attractor behavior was proposed: instead of a monotonic interpolation between early and late asymptotics, the system exhibits an attractor associated with repeated cycles under periodic forcing. The same proposal identifies ultracold gases with externally modulated scattering length as an experimental avenue for observing such behavior (Mazeliauskas et al., 31 Jan 2025).

Stochastic fluctuations modify, but do not remove, universality. In a noisy plasma undergoing Bjorken expansion, the leading-order evolution is still captured by the classical hydrodynamic attractor, while quadratic couplings of fluctuations are described by a generalized hydrodynamic kinetic equation. The backreaction of fluctuations renormalizes transport coefficients, generates long-time tails of high orders, and leads to a renormalized hydrodynamic attractor for which the evolution toward equilibrium becomes non-monotonic (Chen et al., 2022).

Across these realizations, the hydrodynamic attractor remains a statement about universality after transient information loss, but the underlying analytic structures vary substantially: temporal attractors can require transseries completion, spatial attractors can be strictly Borel summable or convergent, non-conformal systems can lack an early-time attractor in standard observables, and driven quantum gases can display both monotonic and cyclic forms of attractor behavior.