Reversible Navier–Stokes Equation (RNS)

• RNS is a time-reversible modification of the forced incompressible Navier–Stokes equations that replaces constant viscosity with a dynamically adjusted multiplier to conserve enstrophy or energy.
• It builds on Gallavotti’s construction to bridge microscopic reversibility with macroscopic dissipation, leading to ensemble equivalence in low-mode observables.
• Numerical studies reveal that the reversible viscosity self-averages to a constant value, aligning stationary statistics and energy spectra with those of irreversible flows.

The reversible Navier–Stokes equation (RNS) is a time-reversible modification of the forced incompressible Navier–Stokes equations in which the usual constant viscosity is replaced by a state-dependent scalar multiplier chosen so that a prescribed global quadratic quantity is exactly conserved. In Gallavotti’s construction, the preferred constraint is enstrophy, so the irreversible viscous term is replaced by a fluctuating “reversible viscosity” that restores invariance under velocity reversal while preserving a stationary nonequilibrium dynamics comparable, for suitable observables, to that of the standard dissipative Navier–Stokes system (Gallavotti, 2017, Gallavotti, 2019). In later work, the label RNS is also used for an energy-conserving three-dimensional thermostat model, so the term denotes a family of constrained reversible fluid equations rather than a single unique PDE (Shukla et al., 2018).

1. Conceptual origin and statistical-mechanical interpretation

RNS was introduced to address a structural mismatch between microscopic and macroscopic descriptions of fluids. At the microscopic level, the underlying particle dynamics are time reversible, whereas the incompressible Navier–Stokes equations are macroscopically irreversible because the viscous term has fixed positive sign and therefore does not transform correctly under

$t\mapsto -t,\qquad \mathbf u\mapsto -\mathbf u.$

Gallavotti’s proposal is that irreversibility at the level of the macroscopic equation may admit an alternative reversible representation, in close analogy with Gaussian thermostats in nonequilibrium statistical mechanics (Gallavotti, 2017, Gallavotti, 2019).

The central organizing idea is an equivalence of dynamical ensembles. The usual Navier–Stokes dynamics with fixed viscosity defines a family of stationary states parametrized by the Reynolds-like control parameter $R=\nu^{-1}$. The reversible constrained dynamics defines a second family parametrized instead by the fixed value of the conserved quadratic quantity, typically enstrophy. The conjecture is that, for observables depending only on finitely many low Fourier modes, the two stationary descriptions become equivalent in the ultraviolet-cutoff removal limit $N\to\infty$, in direct analogy with canonical–microcanonical equivalence in equilibrium statistical mechanics (Gallavotti, 2019, Gallavotti, 2020).

This viewpoint is not restricted to strongly turbulent flows. The strongest heuristic argument is for large $R$, where the fluctuating reversible viscosity is expected to homogenize on low-mode observables. However, the conjecture is also formulated conceptually for laminar regimes, with the caveat that multiple stationary states may coexist and correspondence may need to be defined between matching ergodic components (Gallavotti, 2017, Gallavotti, 2019).

A useful classification is the following.

Variant of RNS	Conserved quantity	Representative paper
Gallavotti-type enstrophy-conserving RNS	$$\mathcal D(\mathbf u)=\int |\nabla \mathbf u|^2\,dx$$ or its Fourier analogue	(Gallavotti, 2017)
3D enstrophy-conserving truncated RNS	$$D(u)=\sum_{\mathbf k,\beta} k^2|u_{\beta,\mathbf k}|^2$$	(Margazoglou et al., 2022)
3D energy-conserving RNS	$E=\frac12\int |\mathbf u|^2\,d\mathbf x=E_0$	(Shukla et al., 2018)

The shared mechanism is the replacement of constant viscosity by a dynamically adjusted thermostat multiplier. What changes across formulations is the constrained invariant and, in three dimensions, the explicit form of the multiplier required to compensate nonlinear vortex stretching (Shukla et al., 2018, Margazoglou et al., 2022).

2. Governing equations and reversible viscosity

The generic thermostat construction starts from a dissipative system

$\dot x = h(x)-\nu Lx,$

with $L$ positive definite and $\nu>0$, and replaces the constant friction by

$R=\nu^{-1}$0

so that the quadratic quantity

$R=\nu^{-1}$1

is an exact constant of motion (Gallavotti, 2017). RNS is the fluid realization of this template.

In the dimensionless forced incompressible Navier–Stokes equation on a periodic box, Gallavotti writes the irreversible dynamics as

$R=\nu^{-1}$2

with $R=\nu^{-1}$3 the Reynolds number and forcing restricted to large scales (Gallavotti, 2017). The reversible counterpart replaces $R=\nu^{-1}$4 by a state-dependent multiplier:

$R=\nu^{-1}$5

This is the basic RNS equation (Gallavotti, 2017).

In the two-dimensional enstrophy-conserving formulation, the constrained quantity is

$R=\nu^{-1}$6

which is proportional, on the torus under incompressibility, to the $R=\nu^{-1}$7-norm of the scalar vorticity. In Fourier variables,

$R=\nu^{-1}$8

the reversible viscosity is

$R=\nu^{-1}$9

or equivalently, in the notation of the regularized $N\to\infty$0 system,

$N\to\infty$1

The denominator is the weighted enstrophy norm $N\to\infty$2, and the numerator is the forcing injection projected onto the enstrophy balance (Gallavotti, 2017, Gallavotti, 2019).

In three-dimensional enstrophy-conserving RNS, the nonlinear term contributes to enstrophy production, so the multiplier is more complicated. In helical Fourier variables one representative expression is

$N\to\infty$3

with

$N\to\infty$4

$N\to\infty$5

so that $N\to\infty$6 for

$N\to\infty$7

In physical space, an equivalent 3D enstrophy-conserving formula contains both forcing input and the vorticity-stretching term:

$N\to\infty$8

(Margazoglou et al., 2022, Jaccod et al., 2020).

A distinct three-dimensional variant conserves total kinetic energy rather than enstrophy. There the reversible viscosity is

$N\to\infty$9

and the constraint is

$R$0

This model is the basis of the phase-transition analysis of three-dimensional RNS steady states (Shukla et al., 2018).

3. Reversibility, constraints, and dynamical structure

The defining symmetry of RNS is the involution

$R$1

with reversibility expressed as

$R$2

For the thermostatted equations, this follows from the fact that the Euler or advective part transforms with the required oddness and, crucially, the multiplier satisfies

$R$3

Therefore, if $R$4 is a solution, then $R$5 is also a solution. Standard Navier–Stokes with fixed $R$6 lacks this property because the viscous coefficient does not change sign under reversal (Gallavotti, 2017, Gallavotti, 2019, Margazoglou et al., 2022).

The reversible multiplier is not an externally imposed parameter. It is determined instantaneously by the current state and forcing so that the selected quadratic quantity remains exactly constant. In the two-dimensional Gallavotti construction, this means exact conservation of enstrophy-like

$R$7

In the energy-conserving three-dimensional formulation, it means exact conservation of

$R$8

In both cases the thermostat acts as a Lagrange multiplier enforcing a global constraint (Gallavotti, 2017, Shukla et al., 2018).

A recurrent source of confusion is that reversibility does not imply absence of effective dissipation. The multiplier can fluctuate substantially, and at finite truncation it can in principle become negative. In the enstrophy-conserving formulation this is precisely what allows the vector field to be time reversible. At the same time, phase-space contraction remains meaningful, and the reversible flow is not described as Liouville-volume preserving (Gallavotti, 2017, Jaccod et al., 2020).

The distinction between two and three dimensions is structural. In $R$9, the Euler nonlinearity preserves enstrophy, which yields the simple forcing-only numerator in $$\mathcal D(\mathbf u)=\int |\nabla \mathbf u|^2\,dx$$0. In $$\mathcal D(\mathbf u)=\int |\nabla \mathbf u|^2\,dx$$1, the nonlinear term produces enstrophy through vortex stretching, so the thermostat must compensate both forcing injection and nonlinear enstrophy production; this is why the 3D formulas contain the additional cubic contribution (Gallavotti, 2019, Margazoglou et al., 2022).

4. Ensemble equivalence and the ultraviolet-cutoff limit

RNS is primarily a statement about stationary ensembles rather than only about formal symmetry. For the UV-regularized system with cutoff $$\mathcal D(\mathbf u)=\int |\nabla \mathbf u|^2\,dx$$2, the irreversible dynamics generates stationary states $$\mathcal D(\mathbf u)=\int |\nabla \mathbf u|^2\,dx$$3 or, equivalently, $$\mathcal D(\mathbf u)=\int |\nabla \mathbf u|^2\,dx$$4, while the reversible constrained dynamics generates $$\mathcal D(\mathbf u)=\int |\nabla \mathbf u|^2\,dx$$5 or $$\mathcal D(\mathbf u)=\int |\nabla \mathbf u|^2\,dx$$6. The two are called corresponding when the reversible fixed enstrophy equals the mean enstrophy of the irreversible stationary state:

$$\mathcal D(\mathbf u)=\int |\nabla \mathbf u|^2\,dx$$7

This matching relation is the nonequilibrium analogue of matching inverse temperature and energy between canonical and microcanonical ensembles (Gallavotti, 2019, Margazoglou et al., 2022).

The observables relevant to the conjecture are local in Fourier space. An observable $$\mathcal D(\mathbf u)=\int |\nabla \mathbf u|^2\,dx$$8 is local if it depends only on finitely many low-wavenumber modes,

$$\mathcal D(\mathbf u)=\int |\nabla \mathbf u|^2\,dx$$9

with $$D(u)=\sum_{\mathbf k,\beta} k^2|u_{\beta,\mathbf k}|^2$$0 fixed while $$D(u)=\sum_{\mathbf k,\beta} k^2|u_{\beta,\mathbf k}|^2$$1. The equivalence statement is then

$$D(u)=\sum_{\mathbf k,\beta} k^2|u_{\beta,\mathbf k}|^2$$2

or, in the alternative notation,

$$D(u)=\sum_{\mathbf k,\beta} k^2|u_{\beta,\mathbf k}|^2$$3

for corresponding states (Gallavotti, 2019, Margazoglou et al., 2022).

The role of the ultraviolet cutoff is fundamental. The truncated dynamics retains only modes $$D(u)=\sum_{\mathbf k,\beta} k^2|u_{\beta,\mathbf k}|^2$$4, or $$D(u)=\sum_{\mathbf k,\beta} k^2|u_{\beta,\mathbf k}|^2$$5 depending on convention, thereby producing finite-dimensional ODEs on a phase space $$D(u)=\sum_{\mathbf k,\beta} k^2|u_{\beta,\mathbf k}|^2$$6. The analogue of the thermodynamic limit is not volume growth but cutoff removal:

$$D(u)=\sum_{\mathbf k,\beta} k^2|u_{\beta,\mathbf k}|^2$$7

This is the limit in which the ensemble-equivalence conjecture is formulated (Gallavotti, 2019, Gallavotti, 2020).

A central sum rule follows from the stationary energy balance. Let

$$D(u)=\sum_{\mathbf k,\beta} k^2|u_{\beta,\mathbf k}|^2$$8

be the injected power. Because forcing acts only at low modes, $$D(u)=\sum_{\mathbf k,\beta} k^2|u_{\beta,\mathbf k}|^2$$9 is local in the relevant sense. Then stationarity yields

$E=\frac12\int |\mathbf u|^2\,d\mathbf x=E_0$0

for the irreversible system, and

$E=\frac12\int |\mathbf u|^2\,d\mathbf x=E_0$1

for the reversible one. For corresponding states this implies

$E=\frac12\int |\mathbf u|^2\,d\mathbf x=E_0$2

or equivalently that the fluctuating reversible viscosity self-averages to the constant viscosity of the irreversible dynamics (Gallavotti, 2019). In the 2017 formulation this appears as the prediction

$E=\frac12\int |\mathbf u|^2\,d\mathbf x=E_0$3

for corresponding ensembles (Gallavotti, 2017).

This equivalence is not a claim that the full stationary measures are identical at finite truncation. The papers stress that the two distributions can be globally very different while becoming indistinguishable on low-mode observables as $E=\frac12\int |\mathbf u|^2\,d\mathbf x=E_0$4 (Gallavotti, 2019).

5. Numerical evidence, regimes, and phase-transition phenomenology

The empirical literature separates naturally into three lines: preliminary two-dimensional tests of self-averaging, three-dimensional studies of the enstrophy-conserving model, and three-dimensional studies of the energy-conserving model.

In Gallavotti’s preliminary two-dimensional computations, $E=\frac12\int |\mathbf u|^2\,d\mathbf x=E_0$5 was evaluated along trajectories of the irreversible Navier–Stokes flow using the reversible formula. For cutoff $E=\frac12\int |\mathbf u|^2\,d\mathbf x=E_0$6 and $E=\frac12\int |\mathbf u|^2\,d\mathbf x=E_0$7, the instantaneous quantity $E=\frac12\int |\mathbf u|^2\,d\mathbf x=E_0$8 fluctuated strongly while its running average stayed close to $E=\frac12\int |\mathbf u|^2\,d\mathbf x=E_0$9. In a larger run described as “960 modes” with $\dot x = h(x)-\nu Lx,$0, the same behavior was observed, with the running average close to $\dot x = h(x)-\nu Lx,$1 within a few percent. These calculations were presented as evidence for self-averaging of the reversible viscosity rather than as a full side-by-side comparison of reversible and irreversible stationary measures (Gallavotti, 2017).

A more direct three-dimensional comparison was carried out in fully developed turbulence using an enstrophy-conserving RNS integrated from Taylor–Green initial data and forced by Taylor–Green forcing. For $\dot x = h(x)-\nu Lx,$2, with resolutions $\dot x = h(x)-\nu Lx,$3 and $\dot x = h(x)-\nu Lx,$4, the reversible enstrophy was set equal to the mean Navier–Stokes enstrophy,

$\dot x = h(x)-\nu Lx,$5

The predicted reciprocity

$\dot x = h(x)-\nu Lx,$6

was reported to hold within numerical errors of about $\dot x = h(x)-\nu Lx,$7 at all Reynolds numbers. The same study reported strong agreement for second and fourth velocity moments, one-point PDFs, energy spectra, temporal autocorrelations, and coarse-grained scale-by-scale energy fluxes, with only minor discrepancies at the highest Reynolds number attributed to statistical convergence (Jaccod et al., 2020).

A systematic pseudospectral investigation of the three-dimensional enstrophy-conserving truncated model refined the locality aspect of the conjecture. In the hydrodynamic regime, where the cutoff exceeds the dissipation scale by enough margin, the means of single-mode energies $\dot x = h(x)-\nu Lx,$8, their standard deviations, and mean shell energies $\dot x = h(x)-\nu Lx,$9 were found to remain very close between irreversible and reversible ensembles up to

$L$0

where

$L$1

For the standard deviation of shell energies,

$L$2

the locality range was more restrictive, with empirical

$L$3

The same study emphasized that negative values of the fluctuating viscosity become extremely rare in the fully resolved hydrodynamic regime and are instead associated mainly with under-resolved or quasithermalized cases (Margazoglou et al., 2022).

The energy-conserving three-dimensional RNS exhibits a different phenomenology. At finite Galerkin truncation, it develops two statistically distinct stationary regimes: a high-enstrophy “warm” phase at small

$L$4

with partial ultraviolet thermalization, and a low-enstrophy hydrodynamic phase at large $L$5. The crossover occurs in a narrow interval roughly

$L$6

with critical estimate

$L$7

Near this region the system becomes strongly bursty, and the transition is interpreted as the finite-size remnant of a continuous phase transition, with normalized enstrophy as order parameter and $L$8 as the symmetry-breaking field (Shukla et al., 2018).

This phase-transition picture is then used to reinterpret Gallavotti’s equivalence conjecture. The relevant asymptotic route is not merely vanishing viscosity at fixed truncation, which leads toward truncated-Euler warm states, but a joint approach to criticality from the hydrodynamic side together with cutoff removal. In the direct comparison performed in that study, an RNS run slightly above criticality was matched to a Navier–Stokes run with viscosity chosen as

$L$9

The Navier–Stokes energy satisfied approximate reflexivity,

$\nu>0$0

and the two systems showed very good agreement in spectra and fluxes over roughly one decade of wavenumbers $\nu>0$1 (Shukla et al., 2018).

6. Phase-space contraction, limitations, and related directions

Because the thermostat coefficient depends on the state, RNS connects naturally to nonequilibrium fluctuation theory. For the truncated reversible system, the phase-space divergence $\nu>0$2 can be computed explicitly. In the two-dimensional formulation,

$\nu>0$3

with

$\nu>0$4

Under the chaotic hypothesis, a Gallavotti–Cohen type fluctuation relation is then proposed for the normalized time-averaged contraction (Gallavotti, 2017). Closely related formulas and the possibility of transferring fluctuation-theorem predictions from reversible to irreversible turbulence through ensemble equivalence are discussed in the 2D truncation framework of (Gallavotti, 2020).

The literature is explicit that the status of these claims is conjectural or heuristic rather than theorem-level for the Navier–Stokes PDE. Several limitations recur across the papers. RNS well-posedness is not analyzed mathematically in the foundational 2017 work. Most direct equivalence tests are numerical and finite-dimensional. The fluctuation-relation discussion relies on spectral truncation because otherwise some phase-space sums diverge. In low-Reynolds-number regimes, multiple attractors, fixed points, or periodic orbits may coexist, so correspondence between irreversible and reversible ensembles must be interpreted componentwise (Gallavotti, 2017, Gallavotti, 2019).

Open problems are stated concretely. They include systematic comparison of many $\nu>0$5-local observables in corresponding stationary states; the dependence of equivalence on cutoff $\nu>0$6 and Reynolds number $\nu>0$7; the comparison of Lyapunov spectra or $\nu>0$8-local Lyapunov exponents; the behavior of the negative tail of the fluctuating viscosity distribution; and the three-dimensional regularity question for enstrophy-conserving RNS, especially whether positivity bounds on $\nu>0$9 can hold with probability one in stationary regimes (Gallavotti, 2017, Margazoglou et al., 2022).

A final boundary of the subject is with stochastic thermodynamic completions of Navier–Stokes. A paper derives an exact reversible/irreversible decomposition in which nonlinear convection is the reversible Hamiltonian sector and viscosity is paired with fluctuation-dissipation noise, but it explicitly does not construct a deterministic Gallavotti-style reversible viscosity or an exactly time-reversal-invariant deterministic thermostat. This suggests a close conceptual adjacency to RNS while remaining a distinct framework rather than an RNS model in the strict sense (Braunstein, 20 May 2026).

In its most compact technical form, RNS is therefore best understood as a thermostatted Navier–Stokes dynamics in which

$R=\nu^{-1}$00

with $R=\nu^{-1}$01 chosen so that a global quadratic invariant is exactly conserved, and with the central conjecture that corresponding irreversible and reversible stationary states yield the same statistics for low-mode observables in the appropriate cutoff-removal limit (Gallavotti, 2017, Gallavotti, 2019).