Papers
Topics
Authors
Recent
Search
2000 character limit reached

Anomalous Dissipation: Definition, Examples

Updated 7 November 2026
  • Anomalous dissipation refers to persistent energy, enstrophy, scalar variance, or entropy defects in the limit of vanishing molecular viscosity or diffusivity. It includes viscous and inviscid anomalous dissipation, with key distinctions in weak convergence and energy loss.
  • Typical examples include viscous anomalous dissolution where $ u$ times norm of the gradient squared remains nonzero as $ u$ tends to 0; inviscid anomalous dissolution occurs when Euler solution loses conserved quadratic quantity resulting in finite-scale distributional fluxes.
  • Construction in $2 rac{1}{2}$-dimensional limit show negligible $ u$ dissipation giving energy defect reproducing behavior similarly for nonlinear solutions in Navier-stokes.

Anomalous dissipation denotes the persistence of a nonzero energy, enstrophy, scalar-variance, or entropy defect in a limit where the explicit molecular viscosity or diffusivity vanishes, or the failure of an inviscid weak solution to satisfy the formal conservation law. The term covers related but distinct phenomena: viscous or dissipation anomaly, in which ν∫∣∇uν∣2\nu\int|\nabla u^\nu|^2 remains nonzero as ν→0\nu\to0; inviscid anomalous dissipation, in which a limiting Euler or transport solution loses the formally conserved quadratic quantity; and finite-scale or distributional energy fluxes represented by Duchon–Robert defects. The distinction is essential because weak convergence can lose energy even when viscous dissipation tends to zero, whereas positive limiting viscous dissipation necessarily produces an energy defect in the inviscid limit (Ehrich et al., 2023).

1. Definitions and mathematical distinctions

For incompressible Navier–Stokes flow,

∂tuν+(uν⋅∇)uν=−∇pν+νΔuν+fν,∇⋅uν=0,\partial_tu^\nu+(u^\nu\cdot\nabla)u^\nu=-\nabla p^\nu+\nu\Delta u^\nu+f^\nu, \qquad \nabla\cdot u^\nu=0,

the kinetic-energy identity is

∥uν(t)∥L22=∥uin∥L22−2ν∫0t∥∇uν∥L22 dτ+2∫0t(fν,uν) dτ.\|u^\nu(t)\|_{L^2}^2 = \|u_{\rm in}\|_{L^2}^2 -2\nu\int_0^t\|\nabla u^\nu\|_{L^2}^2\,d\tau +2\int_0^t(f^\nu,u^\nu)\,d\tau .

The quantity

D(t):=lim sup⁡ν→02ν∫0t∥∇uν∥L22 dτD(t):=\limsup_{\nu\to0}2\nu\int_0^t\|\nabla u^\nu\|_{L^2}^2\,d\tau

is the cumulative viscous dissipation anomaly. A limiting Euler solution has anomalous dissipation when

∥uin∥L22+W(t)−∥u(t)∥L22>0,W(t)=2∫0t(f,u) dτ.\|u_{\rm in}\|_{L^2}^2+W(t)-\|u(t)\|_{L^2}^2>0, \qquad W(t)=2\int_0^t(f,u)\,d\tau .

Weak lower semicontinuity yields

0≤D(t)≤∥uin∥L22+W(t)−∥u(t)∥L22.0\le D(t)\le \|u_{\rm in}\|_{L^2}^2+W(t)-\|u(t)\|_{L^2}^2.

Consequently, dissipation anomaly implies anomalous dissipation, but the converse fails: energy may disappear from a weak limit through oscillation or concentration without being represented by a nonzero limit of ν∥∇uν∥22\nu\|\nabla u^\nu\|_2^2 (Ehrich et al., 2023).

The same structure appears for passive scalars: ∂tθκ+u⋅∇θκ=κΔθκ,\partial_t\theta^\kappa+u\cdot\nabla\theta^\kappa =\kappa\Delta\theta^\kappa, with

12∥θκ(t)∥L22+κ∫0t∥∇θκ∥L22 ds=12∥θ0∥L22.\frac12\|\theta^\kappa(t)\|_{L^2}^2+ \kappa\int_0^t\|\nabla\theta^\kappa\|_{L^2}^2\,ds = \frac12\|\theta_0\|_{L^2}^2.

Scalar anomalous dissipation means

ν→0\nu\to00

For inviscid Euler, a weak solution may instead satisfy a local energy inequality,

ν→0\nu\to01

where ν→0\nu\to02 is a nonnegative defect distribution or Radon measure. The defect is not an explicit molecular dissipation term; it is the residual of nonlinear transfer after passage to a rough limit (Dascaliuc et al., 2011).

2. The Duchon–Robert defect and Onsager regularity

For incompressible flow, spatial coarse-graining gives the Duchon–Robert expression

ν→0\nu\to03

If velocity increments obey

ν→0\nu\to04

then

ν→0\nu\to05

Thus ν→0\nu\to06 forces the defect to vanish, ν→0\nu\to07 permits a finite nonzero limit, and ν→0\nu\to08 permits anomalous dissipation. This is the local flux formulation of Onsager’s threshold.

The critical regularity is often expressed through

ν→0\nu\to09

which implies energy conservation for Euler. The condition is sharp in the sense that rougher solutions can possess nonzero nonlinear flux. However, ∂tuν+(uν⋅∇)uν=−∇pν+νΔuν+fν,∇⋅uν=0,\partial_tu^\nu+(u^\nu\cdot\nabla)u^\nu=-\nabla p^\nu+\nu\Delta u^\nu+f^\nu, \qquad \nabla\cdot u^\nu=0,0 is not a necessary condition for every individual weak solution to conserve energy: rough, energy-conserving examples below the threshold exist (Dascaliuc et al., 2011).

For compressible flow, density fluctuations introduce additional channels. With

∂tuν+(uν⋅∇)uν=−∇pν+νΔuν+fν,∇⋅uν=0,\partial_tu^\nu+(u^\nu\cdot\nabla)u^\nu=-\nabla p^\nu+\nu\Delta u^\nu+f^\nu, \qquad \nabla\cdot u^\nu=0,1

the compressible Duchon–Robert balance contains

∂tuν+(uν⋅∇)uν=−∇pν+νΔuν+fν,∇⋅uν=0,\partial_tu^\nu+(u^\nu\cdot\nabla)u^\nu=-\nabla p^\nu+\nu\Delta u^\nu+f^\nu, \qquad \nabla\cdot u^\nu=0,2

∂tuν+(uν⋅∇)uν=−∇pν+νΔuν+fν,∇⋅uν=0,\partial_tu^\nu+(u^\nu\cdot\nabla)u^\nu=-\nabla p^\nu+\nu\Delta u^\nu+f^\nu, \qquad \nabla\cdot u^\nu=0,3

and

∂tuν+(uν⋅∇)uν=−∇pν+νΔuν+fν,∇⋅uν=0,\partial_tu^\nu+(u^\nu\cdot\nabla)u^\nu=-\nabla p^\nu+\nu\Delta u^\nu+f^\nu, \qquad \nabla\cdot u^\nu=0,4

The first is the compressible inertial transfer, the second is a pressure–density or baropycnal anomaly, and the third belongs to the viscous channel. For a local Hölder profile, ∂tuν+(uν⋅∇)uν=−∇pν+νΔuν+fν,∇⋅uν=0,\partial_tu^\nu+(u^\nu\cdot\nabla)u^\nu=-\nabla p^\nu+\nu\Delta u^\nu+f^\nu, \qquad \nabla\cdot u^\nu=0,5 scales as ∂tuν+(uν⋅∇)uν=−∇pν+νΔuν+fν,∇⋅uν=0,\partial_tu^\nu+(u^\nu\cdot\nabla)u^\nu=-\nabla p^\nu+\nu\Delta u^\nu+f^\nu, \qquad \nabla\cdot u^\nu=0,6, whereas if pressure and density have equal Hölder exponent ∂tuν+(uν⋅∇)uν=−∇pν+νΔuν+fν,∇⋅uν=0,\partial_tu^\nu+(u^\nu\cdot\nabla)u^\nu=-\nabla p^\nu+\nu\Delta u^\nu+f^\nu, \qquad \nabla\cdot u^\nu=0,7, the pressure–density term scales as ∂tuν+(uν⋅∇)uν=−∇pν+νΔuν+fν,∇⋅uν=0,\partial_tu^\nu+(u^\nu\cdot\nabla)u^\nu=-\nabla p^\nu+\nu\Delta u^\nu+f^\nu, \qquad \nabla\cdot u^\nu=0,8 and has critical exponent ∂tuν+(uν⋅∇)uν=−∇pν+νΔuν+fν,∇⋅uν=0,\partial_tu^\nu+(u^\nu\cdot\nabla)u^\nu=-\nabla p^\nu+\nu\Delta u^\nu+f^\nu, \qquad \nabla\cdot u^\nu=0,9 (Zinchenko et al., 5 Aug 2025).

At finite viscosity and finite filter scale, a Duchon–Robert signal is not automatically a true anomaly. Smooth Burgers vortices have

∥uν(t)∥L22=∥uin∥L22−2ν∫0t∥∇uν∥L22 dτ+2∫0t(fν,uν) dτ.\|u^\nu(t)\|_{L^2}^2 = \|u_{\rm in}\|_{L^2}^2 -2\nu\int_0^t\|\nabla u^\nu\|_{L^2}^2\,d\tau +2\int_0^t(f^\nu,u^\nu)\,d\tau .0

as ∥uν(t)∥L22=∥uin∥L22−2ν∫0t∥∇uν∥L22 dτ+2∫0t(fν,uν) dτ.\|u^\nu(t)\|_{L^2}^2 = \|u_{\rm in}\|_{L^2}^2 -2\nu\int_0^t\|\nabla u^\nu\|_{L^2}^2\,d\tau +2\int_0^t(f^\nu,u^\nu)\,d\tau .1, although their finite-scale flux is spatially organized near vortex-core boundaries. DNS studies identify such structures as local precursors rather than proofs of a nonzero limiting defect (Zinchenko et al., 2024).

3. Cascades, mixing, and vanishing-viscosity constructions

A common mechanism is the creation of progressively finer scales while the inviscid quadratic quantity remains macroscopic. In a ∥uν(t)∥L22=∥uin∥L22−2ν∫0t∥∇uν∥L22 dτ+2∫0t(fν,uν) dτ.\|u^\nu(t)\|_{L^2}^2 = \|u_{\rm in}\|_{L^2}^2 -2\nu\int_0^t\|\nabla u^\nu\|_{L^2}^2\,d\tau +2\int_0^t(f^\nu,u^\nu)\,d\tau .2-dimensional reduction,

∥uν(t)∥L22=∥uin∥L22−2ν∫0t∥∇uν∥L22 dτ+2∫0t(fν,uν) dτ.\|u^\nu(t)\|_{L^2}^2 = \|u_{\rm in}\|_{L^2}^2 -2\nu\int_0^t\|\nabla u^\nu\|_{L^2}^2\,d\tau +2\int_0^t(f^\nu,u^\nu)\,d\tau .3

the planar divergence-free velocity transports the third component: ∥uν(t)∥L22=∥uin∥L22−2ν∫0t∥∇uν∥L22 dτ+2∫0t(fν,uν) dτ.\|u^\nu(t)\|_{L^2}^2 = \|u_{\rm in}\|_{L^2}^2 -2\nu\int_0^t\|\nabla u^\nu\|_{L^2}^2\,d\tau +2\int_0^t(f^\nu,u^\nu)\,d\tau .4 The scalar retains its ∥uν(t)∥L22=∥uin∥L22−2ν∫0t∥∇uν∥L22 dτ+2∫0t(fν,uν) dτ.\|u^\nu(t)\|_{L^2}^2 = \|u_{\rm in}\|_{L^2}^2 -2\nu\int_0^t\|\nabla u^\nu\|_{L^2}^2\,d\tau +2\int_0^t(f^\nu,u^\nu)\,d\tau .5 norm while being mixed to frequencies ∥uν(t)∥L22=∥uin∥L22−2ν∫0t∥∇uν∥L22 dτ+2∫0t(fν,uν) dτ.\|u^\nu(t)\|_{L^2}^2 = \|u_{\rm in}\|_{L^2}^2 -2\nu\int_0^t\|\nabla u^\nu\|_{L^2}^2\,d\tau +2\int_0^t(f^\nu,u^\nu)\,d\tau .6. Its viscous approximation satisfies

∥uν(t)∥L22=∥uin∥L22−2ν∫0t∥∇uν∥L22 dτ+2∫0t(fν,uν) dτ.\|u^\nu(t)\|_{L^2}^2 = \|u_{\rm in}\|_{L^2}^2 -2\nu\int_0^t\|\nabla u^\nu\|_{L^2}^2\,d\tau +2\int_0^t(f^\nu,u^\nu)\,d\tau .7

When ∥uν(t)∥L22=∥uin∥L22−2ν∫0t∥∇uν∥L22 dτ+2∫0t(fν,uν) dτ.\|u^\nu(t)\|_{L^2}^2 = \|u_{\rm in}\|_{L^2}^2 -2\nu\int_0^t\|\nabla u^\nu\|_{L^2}^2\,d\tau +2\int_0^t(f^\nu,u^\nu)\,d\tau .8 becomes order one, diffusion removes the accumulated high-frequency variance. This mechanism produces total or partial dissipation, depending on the relative scaling of viscosity and mixing frequency.

Brué and De Lellis constructed forced three-dimensional Navier–Stokes solutions with fixed smooth initial data, smooth forces uniformly bounded in ∥uν(t)∥L22=∥uin∥L22−2ν∫0t∥∇uν∥L22 dτ+2∫0t(fν,uν) dτ.\|u^\nu(t)\|_{L^2}^2 = \|u_{\rm in}\|_{L^2}^2 -2\nu\int_0^t\|\nabla u^\nu\|_{L^2}^2\,d\tau +2\int_0^t(f^\nu,u^\nu)\,d\tau .9 for every D(t):=lim sup⁡ν→02ν∫0t∥∇uν∥L22 dτD(t):=\limsup_{\nu\to0}2\nu\int_0^t\|\nabla u^\nu\|_{L^2}^2\,d\tau0, and

D(t):=lim sup⁡ν→02ν∫0t∥∇uν∥L22 dτD(t):=\limsup_{\nu\to0}2\nu\int_0^t\|\nabla u^\nu\|_{L^2}^2\,d\tau1

The construction embeds a passive scalar into the third velocity component, reducing the problem to planar transport and diffusion (Bruè et al., 2022). Related constructions realize every dissipation level between zero and the total initial energy, partial dissipation with absolutely continuous dissipation in time, and anomalous dissipation without dissipation anomaly. They also produce multiple Euler limits from the same initial data and force (Ehrich et al., 2023).

The mechanism can be realized in unforced whole-space Navier–Stokes with viscosity-dependent initial data. For D(t):=lim sup⁡ν→02ν∫0t∥∇uν∥L22 dτD(t):=\limsup_{\nu\to0}2\nu\int_0^t\|\nabla u^\nu\|_{L^2}^2\,d\tau2, data concentrated at frequency D(t):=lim sup⁡ν→02ν∫0t∥∇uν∥L22 dτD(t):=\limsup_{\nu\to0}2\nu\int_0^t\|\nabla u^\nu\|_{L^2}^2\,d\tau3 have uniformly bounded D(t):=lim sup⁡ν→02ν∫0t∥∇uν∥L22 dτD(t):=\limsup_{\nu\to0}2\nu\int_0^t\|\nabla u^\nu\|_{L^2}^2\,d\tau4 norm. Their heat evolution dissipates a fixed fraction of its energy on a time interval of order one, while the nonlinear Navier–Stokes correction tends to zero. Hence

D(t):=lim sup⁡ν→02ν∫0t∥∇uν∥L22 dτD(t):=\limsup_{\nu\to0}2\nu\int_0^t\|\nabla u^\nu\|_{L^2}^2\,d\tau5

for smooth solutions in every dimension D(t):=lim sup⁡ν→02ν∫0t∥∇uν∥L22 dτD(t):=\limsup_{\nu\to0}2\nu\int_0^t\|\nabla u^\nu\|_{L^2}^2\,d\tau6 (Li et al., 2023).

More recent constructions establish anomalous dissipation at Onsager-critical regularity. In the D(t):=lim sup⁡ν→02ν∫0t∥∇uν∥L22 dτD(t):=\limsup_{\nu\to0}2\nu\int_0^t\|\nabla u^\nu\|_{L^2}^2\,d\tau7-dimensional setting, the velocity and scalar satisfy separate flux criteria: D(t):=lim sup⁡ν→02ν∫0t∥∇uν∥L22 dτD(t):=\limsup_{\nu\to0}2\nu\int_0^t\|\nabla u^\nu\|_{L^2}^2\,d\tau8 for the planar velocity and

D(t):=lim sup⁡ν→02ν∫0t∥∇uν∥L22 dτD(t):=\limsup_{\nu\to0}2\nu\int_0^t\|\nabla u^\nu\|_{L^2}^2\,d\tau9

for the scalar contribution. The construction violates only the mixed scalar condition, while the velocity flux vanishes. A fully three-dimensional example uses intermittent shell blocks and a slightly rough force, with nonzero nonlinear flux, zero anomalous work, and total viscous dissipation (Cheskidov et al., 31 Dec 2025).

4. Passive scalars, stochasticity, and transport noise

For passive scalars, anomalous dissipation is closely connected to mixing, loss of renormalization, and spontaneous stochasticity. Deterministic constructions produce incompressible velocities smooth for every ∥uin∥L22+W(t)−∥u(t)∥L22>0,W(t)=2∫0t(f,u) dτ.\|u_{\rm in}\|_{L^2}^2+W(t)-\|u(t)\|_{L^2}^2>0, \qquad W(t)=2\int_0^t(f,u)\,d\tau .0 but belonging globally to

∥uin∥L22+W(t)−∥u(t)∥L22>0,W(t)=2∫0t(f,u) dτ.\|u_{\rm in}\|_{L^2}^2+W(t)-\|u(t)\|_{L^2}^2>0, \qquad W(t)=2\int_0^t(f,u)\,d\tau .1

Alternating shears create gradients growing like ∥uin∥L22+W(t)−∥u(t)∥L22>0,W(t)=2∫0t(f,u) dτ.\|u_{\rm in}\|_{L^2}^2+W(t)-\|u(t)\|_{L^2}^2>0, \qquad W(t)=2\int_0^t(f,u)\,d\tau .2 and generate a nonzero limit of

∥uin∥L22+W(t)−∥u(t)∥L22>0,W(t)=2∫0t(f,u) dτ.\|u_{\rm in}\|_{L^2}^2+W(t)-\|u(t)\|_{L^2}^2>0, \qquad W(t)=2\int_0^t(f,u)\,d\tau .3

The same examples yield nonuniqueness of weak solutions of the inviscid transport equation: vanishing-viscosity limits select a dissipative solution, whereas time reversal produces a conservative one (Drivas et al., 2019).

A fluctuation–dissipation identity provides a probabilistic formulation: ∥uin∥L22+W(t)−∥u(t)∥L22>0,W(t)=2∫0t(f,u) dτ.\|u_{\rm in}\|_{L^2}^2+W(t)-\|u(t)\|_{L^2}^2>0, \qquad W(t)=2\int_0^t(f,u)\,d\tau .4 If backward stochastic trajectories retain positive variance as ∥uin∥L22+W(t)−∥u(t)∥L22>0,W(t)=2∫0t(f,u) dτ.\|u_{\rm in}\|_{L^2}^2+W(t)-\|u(t)\|_{L^2}^2>0, \qquad W(t)=2\int_0^t(f,u)\,d\tau .5 on a set of terminal points of positive measure, scalar anomalous dissipation follows. Autonomous divergence-free ∥uin∥L22+W(t)−∥u(t)∥L22>0,W(t)=2∫0t(f,u) dτ.\|u_{\rm in}\|_{L^2}^2+W(t)-\|u(t)\|_{L^2}^2>0, \qquad W(t)=2\int_0^t(f,u)\,d\tau .6 velocities in two dimensions, for every fixed ∥uin∥L22+W(t)−∥u(t)∥L22>0,W(t)=2∫0t(f,u) dτ.\|u_{\rm in}\|_{L^2}^2+W(t)-\|u(t)\|_{L^2}^2>0, \qquad W(t)=2\int_0^t(f,u)\,d\tau .7, can exhibit this behavior (Johansson et al., 2024).

A related stochastic mechanism uses transport noise whose total ∥uin∥L22+W(t)−∥u(t)∥L22>0,W(t)=2∫0t(f,u) dτ.\|u_{\rm in}\|_{L^2}^2+W(t)-\|u(t)\|_{L^2}^2>0, \qquad W(t)=2\int_0^t(f,u)\,d\tau .8 amplitude remains bounded while its Fourier support shifts to arbitrarily high frequencies. In Itô form, the noise converges to an effective deterministic Laplacian, but the exact energy balance still contains only the physical diffusivity. Therefore, for every ∥uin∥L22+W(t)−∥u(t)∥L22>0,W(t)=2∫0t(f,u) dτ.\|u_{\rm in}\|_{L^2}^2+W(t)-\|u(t)\|_{L^2}^2>0, \qquad W(t)=2\int_0^t(f,u)\,d\tau .9, suitably chosen high-frequency transport noise can produce

0≤D(t)≤∥uin∥L22+W(t)−∥u(t)∥L22.0\le D(t)\le \|u_{\rm in}\|_{L^2}^2+W(t)-\|u(t)\|_{L^2}^2.0

The analogous result holds for two-dimensional vorticity enstrophy. The noise itself is divergence-free and energy-preserving at fixed diffusivity; anomalous dissipation arises from homogenized high-frequency transport (Agresti, 2024).

For isotropic Gaussian white-in-time velocities with spatial Hölder exponent 0≤D(t)≤∥uin∥L22+W(t)−∥u(t)∥L22.0\le D(t)\le \|u_{\rm in}\|_{L^2}^2+W(t)-\|u(t)\|_{L^2}^2.1, the dissipative regime is associated with spontaneous stochasticity and kernel-valued rather than map-valued flows. Every nonzero 0≤D(t)≤∥uin∥L22+W(t)−∥u(t)∥L22.0\le D(t)\le \|u_{\rm in}\|_{L^2}^2+W(t)-\|u(t)\|_{L^2}^2.2 datum then loses mean energy continuously: 0≤D(t)≤∥uin∥L22+W(t)−∥u(t)∥L22.0\le D(t)\le \|u_{\rm in}\|_{L^2}^2+W(t)-\|u(t)\|_{L^2}^2.3 The limiting dissipation is a nonnegative random Radon measure. In the incompressible case, its expected value satisfies a Duchon–Robert-type formula involving second-order scalar increments normalized by 0≤D(t)≤∥uin∥L22+W(t)−∥u(t)∥L22.0\le D(t)\le \|u_{\rm in}\|_{L^2}^2+W(t)-\|u(t)\|_{L^2}^2.4, while the scalar gains regularity up to, but not including, 0≤D(t)≤∥uin∥L22+W(t)−∥u(t)∥L22.0\le D(t)\le \|u_{\rm in}\|_{L^2}^2+W(t)-\|u(t)\|_{L^2}^2.5 (Drivas et al., 12 Sep 2025).

5. Geometry of dissipation and mechanisms preventing it

Anomalous dissipation is constrained geometrically. For bounded Euler solutions, any nonzero anomalous-dissipation measure must be concentrated on a space-time set of ordinary Hausdorff dimension at least the spatial dimension 0≤D(t)≤∥uin∥L22+W(t)−∥u(t)∥L22.0\le D(t)\le \|u_{\rm in}\|_{L^2}^2+W(t)-\|u(t)\|_{L^2}^2.6. This lower bound is sharp for entropy-producing shocks, passive scalars, and dissipative incompressible Euler constructions. For Navier–Stokes, the corresponding estimates are naturally parabolic; in three dimensions, the Prodi–Serrin regime is consistent with concentration on sets controlled by parabolic Hausdorff measure of dimension one (Rosa et al., 2023).

Boundaries provide a distinct mechanism. In bounded three-dimensional domains with Navier-slip conditions, uniform fractional regularity

0≤D(t)≤∥uin∥L22+W(t)−∥u(t)∥L22.0\le D(t)\le \|u_{\rm in}\|_{L^2}^2+W(t)-\|u(t)\|_{L^2}^2.7

implies

0≤D(t)≤∥uin∥L22+W(t)−∥u(t)∥L22.0\le D(t)\le \|u_{\rm in}\|_{L^2}^2+W(t)-\|u(t)\|_{L^2}^2.8

The threshold 0≤D(t)≤∥uin∥L22+W(t)−∥u(t)∥L22.0\le D(t)\le \|u_{\rm in}\|_{L^2}^2+W(t)-\|u(t)\|_{L^2}^2.9 supplies an ν∥∇uν∥22\nu\|\nabla u^\nu\|_2^20 boundary trace and controls high-frequency concentration. Conversely, a boundary-singular tangential forcing in a ball produces boundary layers whose trace grows sufficiently rapidly to force a positive dissipation limit. The construction is fully three-dimensional but relies on boundary concentration rather than an interior Onsager singularity (Dudley, 1 Jan 2025).

In two-dimensional vortex-sheet flows, the opposite result holds under a distinguished-sign condition. If the initial vorticity is a nonnegative bounded measure plus an ν∥∇uν∥22\nu\|\nabla u^\nu\|_2^21 component, ν∥∇uν∥22\nu\|\nabla u^\nu\|_2^22, and the vorticity and forcing satisfy uniform-integrability conditions, then

ν∥∇uν∥22\nu\|\nabla u^\nu\|_2^23

The proof uses a concentration-sensitive refinement of Nash’s inequality. Positivity of the measure component yields logarithmic control of mass in small disks, preventing the concentration required for anomalous viscous dissipation. This does not imply strong ν∥∇uν∥22\nu\|\nabla u^\nu\|_2^24 convergence or exact energy equality for the limiting Euler solution (Elgindi et al., 25 Apr 2025).

Dissipation can also be structurally stable. Perturbations of the central curves in quasi-self-similar mixing constructions, subject to area-preservation and equal-length constraints, preserve ν∥∇uν∥22\nu\|\nabla u^\nu\|_2^25 stability of coordinate maps and ν∥∇uν∥22\nu\|\nabla u^\nu\|_2^26 stability of local fields. The resulting forced Navier–Stokes solutions retain a positive dissipation lower bound independent of the perturbation amplitude. The result is nevertheless restricted to a specific family of smooth normal perturbations and does not concern arbitrary turbulent flows (Li, 18 May 2026).

6. Turbulence, long-time averages, and scope

For statistically stationary turbulence, anomalous dissipation is commonly expressed through the dimensionless mean rate

ν∥∇uν∥22\nu\|\nabla u^\nu\|_2^27

The Doering–Foias estimate has the form

ν∥∇uν∥22\nu\|\nabla u^\nu\|_2^28

Periodic extensions of finite-time dissipation-anomaly constructions produce smooth time-periodic Navier–Stokes solutions for which

ν∥∇uν∥22\nu\|\nabla u^\nu\|_2^29

for any prescribed ∂tθκ+u⋅∇θκ=κΔθκ,\partial_t\theta^\kappa+u\cdot\nabla\theta^\kappa =\kappa\Delta\theta^\kappa,0. Thus the ∂tθκ+u⋅∇θκ=κΔθκ,\partial_t\theta^\kappa+u\cdot\nabla\theta^\kappa =\kappa\Delta\theta^\kappa,1 upper bound is sharp: the normalized mean dissipation need not vanish at high Reynolds number (Ehrich et al., 2023).

The Kolmogorov and Onsager formulations describe different levels of the same transfer process. Kolmogorov’s dissipation anomaly concerns nonvanishing viscous dissipation in the Navier–Stokes family and an order-one normalized long-time rate. Onsager’s anomalous dissipation concerns the energy defect of the limiting Euler solution. Duchon–Robert theory identifies the defect as a nonlinear flux of velocity increments. Passive-scalar results provide analogous scalar-variance fluxes, while compressible extensions add pressure–density and density–velocity channels.

Several qualifications are necessary. A finite-filter Duchon–Robert signal in a smooth vortex is not a proof of a limiting anomaly. Positive dissipation in a specially designed forcing sequence does not imply the same behavior for arbitrary forces or generic Leray–Hopf solutions. Anomalous dissipation may occur without a viscous dissipation anomaly, and absence of viscous anomaly does not guarantee strong convergence or exact limiting energy balance. Existing constructions often use viscosity-dependent data, forces, or transport coefficients; others employ a ∂tθκ+u⋅∇θκ=κΔθκ,\partial_t\theta^\kappa+u\cdot\nabla\theta^\kappa =\kappa\Delta\theta^\kappa,2-dimensional reduction, boundary singularities, stochastic homogenization, or carefully designed intermittent shells.

The common mathematical structure is the transfer of a conserved quadratic quantity to scales, frequencies, trajectories, or boundary layers at which the formal conservation argument ceases to be valid: ∂tθκ+u⋅∇θκ=κΔθκ,\partial_t\theta^\kappa+u\cdot\nabla\theta^\kappa =\kappa\Delta\theta^\kappa,3 Depending on the model, the defect is represented by a viscous limit, a Duchon–Robert distribution, a positive Radon measure, a stochastic trajectory variance, an effective diffusivity, or a boundary-layer contribution.

Topic to Video (Beta)

No one has generated a video about this topic yet.

Whiteboard

No one has generated a whiteboard explanation for this topic yet.

Follow Topic

Get notified by email when new papers are published related to Anomalous Dissipation.