Anisotropic Navier-Stokes Equations

• Anisotropic Navier–Stokes equations are fluid models where dissipation and viscosity vary by direction, capturing differential horizontal and vertical behaviors.
• Analysis of these equations relies on specialized anisotropic Sobolev and Besov spaces to establish well-posedness, decay estimates, and singular-limit approximations.
• They underpin various regimes including hydrostatic limits, stochastic formulations, and computational methods, offering insights into both compressible and incompressible flows.

Anisotropic Navier–Stokes equations are Navier–Stokes-type systems in which dissipation, viscosity, constitutive stress, scaling, or forcing distinguishes horizontal and vertical directions, or more generally different spatial directions. In the recent literature, this designation covers incompressible systems with only horizontal Laplacian dissipation, directionally split viscous operators, variable-coefficient anisotropic stress tensors, thin-domain scalings whose limit is hydrostatic, compressible barotropic models, stochastic equations with partial viscosity, and kinetic or lattice-Boltzmann derivations of anisotropic macroscopic momentum balance. In each of these forms, anisotropy changes the coercive structure of the PDE, the natural function spaces, and the mechanisms available for compactness, decay, or singular-limit analysis (Li et al., 2021, Mikhailov, 2024, Gao et al., 2020, Kellers et al., 26 May 2026).

1. Governing equations and sources of anisotropy

A basic incompressible model with partial dissipation is the three-dimensional system

$$\partial_t u + u\cdot\nabla u - \Delta_{\rm h} u + \nabla p = 0, \qquad \nabla\cdot u = 0,$$

where $\Delta_{\rm h}=\partial_1^2+\partial_2^2$ acts only on horizontal variables. In this formulation, only the horizontal Laplacian provides dissipation; there is no $\partial_3^2$-term (Xu et al., 2021). In two dimensions, anisotropic viscosity is often written as

$$\partial_t u +(u\cdot\nabla)u-\nu_1\partial_1^2u-\nu_2\partial_2^2u+\nabla p=0, \qquad \operatorname{div}u=0,$$

with $\nu_1,\nu_2>0$, or, in the stochastic setting, with multiplicative noise added after Leray projection (Liang et al., 2018).

A broader constitutive formulation replaces scalar viscosity by an anisotropic tensor. In the periodic variable-coefficient setting one studies

$$\partial_t u_k-\partial_i(\mu_{ij}(x,t)E_{jk}(u))+\partial_k p +(u\cdot\nabla)u_k=f_k,$$

where $E_{jk}(u)=\tfrac12(\partial_j u_k+\partial_k u_j)$, the coefficient matrix $A(x,t)=(\mu_{ij}(x,t))$ is symmetric, and only a relaxed ellipticity condition is imposed on trace-free symmetric strains (Mikhailov, 2024). An even more general periodic model uses a fourth-order tensor $A(x,t)=\{a_{ki;\alpha\beta}(x,t)\}$ in the anisotropic Stokes operator

$$(\mathcal S u)_k=\partial_{x_i}\!\bigl(a_{ki;\alpha\beta}(x,t)\,E_{\alpha\beta}(u)\bigr),$$

again under a relaxed ellipticity hypothesis (Mikhailov, 2024).

In thin-domain and geophysical scalings, anisotropy is tied to aspect ratio. For $\Delta_{\rm h}=\partial_1^2+\partial_2^2$0, after the rescaling $\Delta_{\rm h}=\partial_1^2+\partial_2^2$1, one obtains

$\Delta_{\rm h}=\partial_1^2+\partial_2^2$2

together with a vertical momentum equation multiplied by $\Delta_{\rm h}=\partial_1^2+\partial_2^2$3. Here the anisotropy appears through the factor $\Delta_{\rm h}=\partial_1^2+\partial_2^2$4 in the vertical Laplacian (Li et al., 2021).

Compressible anisotropic systems introduce direction-dependent stress in the momentum balance. In one barotropic hydrostatic-scaling model on $\Delta_{\rm h}=\partial_1^2+\partial_2^2$5, the horizontal viscosity is set to $\Delta_{\rm h}=\partial_1^2+\partial_2^2$6 and the vertical one to $\Delta_{\rm h}=\partial_1^2+\partial_2^2$7, while in another compressible model on an unbounded domain the anisotropic viscous stress tensor is

$\Delta_{\rm h}=\partial_1^2+\partial_2^2$8

with the coordinate form

$\Delta_{\rm h}=\partial_1^2+\partial_2^2$9

The ellipticity window

$\partial_3^2$0

ensures that $\partial_3^2$1 is symmetric and strongly elliptic (Gao et al., 2020, Kreml et al., 23 Oct 2025).

2. Functional frameworks and well-posedness theory

Because full isotropic smoothing is often unavailable, the analysis is organized in anisotropic Sobolev or Besov scales. Typical examples are

$\partial_3^2$2

the two-dimensional spaces $\partial_3^2$3 and $\partial_3^2$4, and homogeneous Besov spaces such as $\partial_3^2$5 (Xu et al., 2021, Liang et al., 2018, Lemarié, 2024). This reflects a structural fact: the dissipation often acts only in horizontal directions, while vertical regularity must be recovered from incompressibility, parity, transport structure, or anisotropic interpolation.

For the deterministic two-dimensional incompressible system with anisotropic viscosity, global well-posedness holds for initial data $\partial_3^2$6, $\partial_3^2$7, with a unique global weak solution

$\partial_3^2$8

obtained through $\partial_3^2$9- and $$\partial_t u +(u\cdot\nabla)u-\nu_1\partial_1^2u-\nu_2\partial_2^2u+\nabla p=0, \qquad \operatorname{div}u=0,$$0-energy estimates, anisotropic cancellations in the convection term, Galerkin or smooth approximation, and Aubin–Lions compactness (Liang et al., 2018). In three dimensions with only horizontal dissipation, small data in the critical anisotropic space $$\partial_t u +(u\cdot\nabla)u-\nu_1\partial_1^2u-\nu_2\partial_2^2u+\nabla p=0, \qquad \operatorname{div}u=0,$$1 yield a unique global solution

$$\partial_t u +(u\cdot\nabla)u-\nu_1\partial_1^2u-\nu_2\partial_2^2u+\nabla p=0, \qquad \operatorname{div}u=0,$$2

and higher regularity propagates under an additional $$\partial_t u +(u\cdot\nabla)u-\nu_1\partial_1^2u-\nu_2\partial_2^2u+\nabla p=0, \qquad \operatorname{div}u=0,$$3 assumption with $$\partial_t u +(u\cdot\nabla)u-\nu_1\partial_1^2u-\nu_2\partial_2^2u+\nabla p=0, \qquad \operatorname{div}u=0,$$4 (Xu et al., 2021).

Several results recover global solvability by altering the constitutive structure. The modified anisotropic 3D system

$$\partial_t u +(u\cdot\nabla)u-\nu_1\partial_1^2u-\nu_2\partial_2^2u+\nabla p=0, \qquad \operatorname{div}u=0,$$5

admits a unique global solution for every divergence-free $$\partial_t u +(u\cdot\nabla)u-\nu_1\partial_1^2u-\nu_2\partial_2^2u+\nabla p=0, \qquad \operatorname{div}u=0,$$6, with

$$\partial_t u +(u\cdot\nabla)u-\nu_1\partial_1^2u-\nu_2\partial_2^2u+\nabla p=0, \qquad \operatorname{div}u=0,$$7

and no smallness assumption on the initial data; the damping term $$\partial_t u +(u\cdot\nabla)u-\nu_1\partial_1^2u-\nu_2\partial_2^2u+\nabla p=0, \qquad \operatorname{div}u=0,$$8 supplies the high-velocity control needed for global uniqueness (Bessaih et al., 2016).

In the periodic variable-coefficient setting, Mikhailov proves existence of global weak solutions under relaxed ellipticity for

$$\partial_t u +(u\cdot\nabla)u-\nu_1\partial_1^2u-\nu_2\partial_2^2u+\nabla p=0, \qquad \operatorname{div}u=0,$$9

together with a pressure $\nu_1,\nu_2>0$0 and a strong energy inequality (Mikhailov, 2024). In the companion Serrin-type theory, weak solutions satisfying

$\nu_1,\nu_2>0$1

obey the energy equality, are unique among weak solutions satisfying the same energy bound, and admit additional regularity under constant-coefficient or smoother variable-coefficient assumptions (Mikhailov, 2024).

In two dimensions with variable density, two distinct anisotropic models are analyzed. For the cross-dissipative system

$\nu_1,\nu_2>0$2

global existence and uniqueness hold for finite-energy data $\nu_1,\nu_2>0$3. For the system with only horizontal dissipation

$\nu_1,\nu_2>0$4

global well-posedness is established for sufficiently small initial velocity and sufficiently small density variation, together with exponential decay of the oscillatory part of the velocity and vorticity (Abidi et al., 17 Mar 2026).

3. Hydrostatic approximation and primitive-equation limits

One of the central roles of anisotropic Navier–Stokes equations is as parent systems for hydrostatic or primitive-equation limits in thin domains. In the incompressible setting on $\nu_1,\nu_2>0$5, with horizontal viscosity of order $\nu_1,\nu_2>0$6 and vertical viscosity of order $\nu_1,\nu_2>0$7, $\nu_1,\nu_2>0$8, the rescaled problem on $\nu_1,\nu_2>0$9 has a vertical momentum equation multiplied by $$\partial_t u_k-\partial_i(\mu_{ij}(x,t)E_{jk}(u))+\partial_k p +(u\cdot\nabla)u_k=f_k,$$0. When $$\partial_t u_k-\partial_i(\mu_{ij}(x,t)E_{jk}(u))+\partial_k p +(u\cdot\nabla)u_k=f_k,$$1, the inequality $$\partial_t u_k-\partial_i(\mu_{ij}(x,t)E_{jk}(u))+\partial_k p +(u\cdot\nabla)u_k=f_k,$$2 makes the formal limit hydrostatic: $$\partial_t u_k-\partial_i(\mu_{ij}(x,t)E_{jk}(u))+\partial_k p +(u\cdot\nabla)u_k=f_k,$$3 The limiting system is then the primitive equations with only horizontal viscosity,

$$\partial_t u_k-\partial_i(\mu_{ij}(x,t)E_{jk}(u))+\partial_k p +(u\cdot\nabla)u_k=f_k,$$4

For well-prepared initial data,

$$\partial_t u_k-\partial_i(\mu_{ij}(x,t)E_{jk}(u))+\partial_k p +(u\cdot\nabla)u_k=f_k,$$5

solutions of the scaled incompressible Navier–Stokes equations converge strongly on every finite interval $$\partial_t u_k-\partial_i(\mu_{ij}(x,t)E_{jk}(u))+\partial_k p +(u\cdot\nabla)u_k=f_k,$$6 to the anisotropic primitive equations, with rate $$\partial_t u_k-\partial_i(\mu_{ij}(x,t)E_{jk}(u))+\partial_k p +(u\cdot\nabla)u_k=f_k,$$7, $$\partial_t u_k-\partial_i(\mu_{ij}(x,t)E_{jk}(u))+\partial_k p +(u\cdot\nabla)u_k=f_k,$$8. The estimate is finite-time rather than uniform in time, and this differs from the critical case $$\partial_t u_k-\partial_i(\mu_{ij}(x,t)E_{jk}(u))+\partial_k p +(u\cdot\nabla)u_k=f_k,$$9, where the limiting primitive equations retain full dissipation and the convergence rate is $E_{jk}(u)=\tfrac12(\partial_j u_k+\partial_k u_j)$0 globally in time (Li et al., 2021).

A complementary Besov-space treatment studies the rescaled anisotropic Navier–Stokes equations

$E_{jk}(u)=\tfrac12(\partial_j u_k+\partial_k u_j)$1

on $E_{jk}(u)=\tfrac12(\partial_j u_k+\partial_k u_j)$2, with $E_{jk}(u)=\tfrac12(\partial_j u_k+\partial_k u_j)$3 or $E_{jk}(u)=\tfrac12(\partial_j u_k+\partial_k u_j)$4. For small initial data in $E_{jk}(u)=\tfrac12(\partial_j u_k+\partial_k u_j)$5, both the primitive equations and the anisotropic Navier–Stokes equations admit unique global-in-time solutions in the same solution space

$E_{jk}(u)=\tfrac12(\partial_j u_k+\partial_k u_j)$6

and if $E_{jk}(u)=\tfrac12(\partial_j u_k+\partial_k u_j)$7, then

$E_{jk}(u)=\tfrac12(\partial_j u_k+\partial_k u_j)$8

This gives a rigorous hydrostatic limit in the same critical Besov framework used for the primitive equations (Lemarié, 2024).

In the compressible barotropic case, the anisotropic hydrostatic approximation is obtained through a relative-entropy method. On $E_{jk}(u)=\tfrac12(\partial_j u_k+\partial_k u_j)$9, with

$A(x,t)=(\mu_{ij}(x,t))$0

the scaled vertical momentum equation has the form

$A(x,t)=(\mu_{ij}(x,t))$1

so that the limit again imposes $A(x,t)=(\mu_{ij}(x,t))$2. Gao–Nečasová–Tang prove, for $A(x,t)=(\mu_{ij}(x,t))$3, that if the initial relative entropy tends to zero, then finite-energy weak solutions of the compressible anisotropic Navier–Stokes equations converge strongly in the relative-entropy sense to the strong solution of the compressible primitive equations. The paper states that this is the first work to use relative entropy inequality for proving hydrostatic approximation and derive the compressible Primitive Equations (Gao et al., 2020).

4. Decay, enhanced dissipation, and stability mechanisms

In anisotropic settings with only horizontal dissipation, large-time behavior is qualitatively different for different velocity components. Xu–Zhang study global small smooth solutions of

$A(x,t)=(\mu_{ij}(x,t))$4

and prove that the horizontal components decay like the solutions of 2D classical Navier–Stokes equations, while the third component decays as the solutions of 3D Navier–Stokes equations. Under the hypotheses $A(x,t)=(\mu_{ij}(x,t))$5, $A(x,t)=(\mu_{ij}(x,t))$6, and smallness of the norms $A(x,t)=(\mu_{ij}(x,t))$7 and $A(x,t)=(\mu_{ij}(x,t))$8, they show

$A(x,t)=(\mu_{ij}(x,t))$9

$A(x,t)=\{a_{ki;\alpha\beta}(x,t)\}$0

and hence

$A(x,t)=\{a_{ki;\alpha\beta}(x,t)\}$1

For the critical choice $A(x,t)=\{a_{ki;\alpha\beta}(x,t)\}$2, this yields $A(x,t)=\{a_{ki;\alpha\beta}(x,t)\}$3. The mechanism is the divergence constraint

$A(x,t)=\{a_{ki;\alpha\beta}(x,t)\}$4

which transfers horizontal dissipation to the vertical component (Xu et al., 2021).

A half-space analogue on $A(x,t)=\{a_{ki;\alpha\beta}(x,t)\}$5, with Dirichlet boundary condition and only horizontal dissipation, shows the same componentwise asymmetry. For small $A(x,t)=\{a_{ki;\alpha\beta}(x,t)\}$6, $A(x,t)=\{a_{ki;\alpha\beta}(x,t)\}$7, one has a unique global solution with optimal decay estimates, including

$A(x,t)=\{a_{ki;\alpha\beta}(x,t)\}$8

The proof requires an explicit Ukai-type representation of the anisotropic Stokes semigroup, boundary operators $A(x,t)=\{a_{ki;\alpha\beta}(x,t)\}$9, and horizontal Littlewood–Paley analysis in suitable Besov-type spaces because the nonlocal operators are not directly bounded on $$(\mathcal S u)_k=\partial_{x_i}\!\bigl(a_{ki;\alpha\beta}(x,t)\,E_{\alpha\beta}(u)\bigr),$$0 or $$(\mathcal S u)_k=\partial_{x_i}\!\bigl(a_{ki;\alpha\beta}(x,t)\,E_{\alpha\beta}(u)\bigr),$$1 (Fujii et al., 2024).

Another stability mechanism is frequency localization. In the infinite cylindrical domain $$(\mathcal S u)_k=\partial_{x_i}\!\bigl(a_{ki;\alpha\beta}(x,t)\,E_{\alpha\beta}(u)\bigr),$$2, large angular Fourier mode initial data of size $$(\mathcal S u)_k=\partial_{x_i}\!\bigl(a_{ki;\alpha\beta}(x,t)\,E_{\alpha\beta}(u)\bigr),$$3 produce global strong solutions of the anisotropic Navier–Stokes equations even when vertical viscosity is absent. Liu–Liu–Zhang derive mode-by-mode estimates showing, for $$(\mathcal S u)_k=\partial_{x_i}\!\bigl(a_{ki;\alpha\beta}(x,t)\,E_{\alpha\beta}(u)\bigr),$$4,

$$(\mathcal S u)_k=\partial_{x_i}\!\bigl(a_{ki;\alpha\beta}(x,t)\,E_{\alpha\beta}(u)\bigr),$$5

which yields exponential-in-$$(\mathcal S u)_k=\partial_{x_i}\!\bigl(a_{ki;\alpha\beta}(x,t)\,E_{\alpha\beta}(u)\bigr),$$6 decay for $$(\mathcal S u)_k=\partial_{x_i}\!\bigl(a_{ki;\alpha\beta}(x,t)\,E_{\alpha\beta}(u)\bigr),$$7 and polynomial decay for $$(\mathcal S u)_k=\partial_{x_i}\!\bigl(a_{ki;\alpha\beta}(x,t)\,E_{\alpha\beta}(u)\bigr),$$8 (Liu et al., 2024).

In two dimensions, fractional one-direction dissipation

$$(\mathcal S u)_k=\partial_{x_i}\!\bigl(a_{ki;\alpha\beta}(x,t)\,E_{\alpha\beta}(u)\bigr),$$9

has now been analyzed across the full range $\Delta_{\rm h}=\partial_1^2+\partial_2^2$00. Wang–Wu–Zhu split the problem into the regimes $\Delta_{\rm h}=\partial_1^2+\partial_2^2$01, $\Delta_{\rm h}=\partial_1^2+\partial_2^2$02, and $\Delta_{\rm h}=\partial_1^2+\partial_2^2$03, proving small-data global regularity, uniform bounds, and algebraic decay in each range; near $\Delta_{\rm h}=\partial_1^2+\partial_2^2$04, the proof introduces spatial polynomial $\Delta_{\rm h}=\partial_1^2+\partial_2^2$05 weights and uses boundedness of Riesz transforms on weighted $\Delta_{\rm h}=\partial_1^2+\partial_2^2$06-spaces (Wang et al., 22 Jan 2026).

A related but distinct line of work studies spectral anisotropy for the classical isotropic 3D Navier–Stokes equations. Chemin, Gallagher, Paicu, and collaborators show that if the Fourier support of the initial data is contained in

$\Delta_{\rm h}=\partial_1^2+\partial_2^2$07

then, for $\Delta_{\rm h}=\partial_1^2+\partial_2^2$08 sufficiently small, one can obtain global smooth solutions even when the $\Delta_{\rm h}=\partial_1^2+\partial_2^2$09 norm of the data is arbitrarily large. This is not an anisotropic viscous operator, but it shows how anisotropy in frequency space can play a role analogous to a small parameter in nonlinear control (Chemin et al., 2012).

5. Stochastic formulations and probabilistic limits

Anisotropic viscosity also has a substantial stochastic theory. In the two-dimensional stochastic equation on $\Delta_{\rm h}=\partial_1^2+\partial_2^2$10,

$\Delta_{\rm h}=\partial_1^2+\partial_2^2$11

the only viscous term is $\Delta_{\rm h}=\partial_1^2+\partial_2^2$12. The natural spaces are the anisotropic divergence-free Sobolev classes $\Delta_{\rm h}=\partial_1^2+\partial_2^2$13 and $\Delta_{\rm h}=\partial_1^2+\partial_2^2$14, and the analysis uses mixed Lebesgue norms and anisotropic trilinear bounds for $\Delta_{\rm h}=\partial_1^2+\partial_2^2$15 (Chen, 2021). Under growth and Lipschitz assumptions on $\Delta_{\rm h}=\partial_1^2+\partial_2^2$16, existence and uniqueness in the variational framework are established in

$\Delta_{\rm h}=\partial_1^2+\partial_2^2$17

and small-noise asymptotics yield both a central limit theorem and a moderate deviation principle (Chen, 2021).

The moderate deviation principle is formulated for

$\Delta_{\rm h}=\partial_1^2+\partial_2^2$18

with rate function

$\Delta_{\rm h}=\partial_1^2+\partial_2^2$19

where $\Delta_{\rm h}=\partial_1^2+\partial_2^2$20 solves the deterministic skeleton equation

$\Delta_{\rm h}=\partial_1^2+\partial_2^2$21

The proof uses the weak convergence approach of Budhiraja–Dupuis (Chen, 2021).

A complementary stochastic well-posedness theory for the anisotropic-viscosity Navier–Stokes equations on $\Delta_{\rm h}=\partial_1^2+\partial_2^2$22 proves existence of martingale solutions and pathwise uniqueness; by the Yamada–Watanabe theorem this implies existence of a probabilistically strong solution. The deterministic part uses the same anisotropic spaces $\Delta_{\rm h}=\partial_1^2+\partial_2^2$23 and $\Delta_{\rm h}=\partial_1^2+\partial_2^2$24, while the stochastic compactness step relies on Galerkin approximation, tightness in nonmetrizable weak topologies, and the Skorokhod–Jakubowski representation theorem (Liang et al., 2018).

At the level of singular limits, a recent result concerns the anisotropic vanishing-viscosity limit for the 3D stochastic Navier–Stokes equations posed between two plates,

$\Delta_{\rm h}=\partial_1^2+\partial_2^2$25

with Dirichlet no-slip boundary condition, horizontal viscosity $\Delta_{\rm h}=\partial_1^2+\partial_2^2$26, vertical viscosity $\Delta_{\rm h}=\partial_1^2+\partial_2^2$27, and anisotropic transport–stretching noise. If $\Delta_{\rm h}=\partial_1^2+\partial_2^2$28 and $\Delta_{\rm h}=\partial_1^2+\partial_2^2$29, then there exists a sequence of martingale weak solutions convergent to the strong solution of the deterministic Euler equation on its lifetime of existence. A specific difficulty is that the anisotropic scaling destroys the divergence-free property for the effective spatial correlation functions of the noise, so the proof must control commutator terms generated in the Itô–Stratonovich correction (Goodair, 13 Mar 2026).

6. Compressible, generalized, and computational extensions

In the compressible theory, anisotropy has recently been combined with generalized solution concepts adapted to low regularity and unbounded domains. Kreml–Nečasová–Tang prove global existence of dissipative turbulent solutions for the compressible anisotropic Navier–Stokes equations on a large class of unbounded domains of invading Lipschitz type, together with weak–strong uniqueness. Their framework introduces a Reynolds defect $\Delta_{\rm h}=\partial_1^2+\partial_2^2$30 and an energy defect $\Delta_{\rm h}=\partial_1^2+\partial_2^2$31, linked by

$\Delta_{\rm h}=\partial_1^2+\partial_2^2$32

and covers all $\Delta_{\rm h}=\partial_1^2+\partial_2^2$33 under the ellipticity conditions on $\Delta_{\rm h}=\partial_1^2+\partial_2^2$34. In contrast to Bresch–Jabin on $\Delta_{\rm h}=\partial_1^2+\partial_2^2$35, no smallness of $\Delta_{\rm h}=\partial_1^2+\partial_2^2$36 is assumed, and the domain is unbounded, which the paper identifies as more conform to geophysical context (Kreml et al., 23 Oct 2025).

Variable-coefficient anisotropy shows that even the notion of “viscosity” can depart substantially from scalar diffusion. In Mikhailov’s periodic theory, relaxed ellipticity is imposed only on trace-free symmetric strains, not on all matrices. This extends the weak formulation and Galerkin method to anisotropic fluids whose constitutive tensors are merely bounded and measurable in space-time. In the Serrin-type sequel, the same framework supports uniqueness and higher regularity in the borderline space $\Delta_{\rm h}=\partial_1^2+\partial_2^2$37, and in two dimensions every weak solution is automatically Serrin-type, so global regularity and uniqueness follow without smallness assumptions on the data (Mikhailov, 2024, Mikhailov, 2024).

Anisotropy can also be introduced at the mesoscopic level. In a single-relaxation-time lattice Boltzmann derivation with anisotropic Maxwell–Boltzmann equilibrium

$\Delta_{\rm h}=\partial_1^2+\partial_2^2$38

the macroscopic momentum balance becomes

$\Delta_{\rm h}=\partial_1^2+\partial_2^2$39

with anisotropic pressure tensor

$\Delta_{\rm h}=\partial_1^2+\partial_2^2$40

and anisotropic shear-viscosity tensor

$\Delta_{\rm h}=\partial_1^2+\partial_2^2$41

The derivation shows that a properly discretized anisotropic equilibrium macroscopically leads to an anisotropic variation of the Navier–Stokes equations while retaining locality of the collision operator, isotropic discrete position and velocity space, and mass and momentum conservation (Kellers et al., 26 May 2026).

Taken together, these results suggest a consistent structural picture. Loss of isotropy does not eliminate rigorous analysis, but it alters the mechanisms by which control is obtained. Depending on the regime, the decisive substitute for full isotropic smoothing may be hydrostatic balance, divergence-induced transfer of dissipation, anisotropic Poincaré inequalities, relative entropy, weighted estimates, mode localization, or generalized solution concepts with weak–strong uniqueness. A plausible implication is that “anisotropic Navier–Stokes equations” should be understood less as a single PDE and more as a family of directionally structured fluid models whose mathematical behavior is governed by how anisotropy is introduced into the stress, the geometry, or the scaling.