Turbulent Thermal Diffusion in Particle-Laden Flows

• Turbulent thermal diffusion is a transport mechanism in temperature-stratified turbulence where particles drift toward areas of lower temperature due to correlated pressure, velocity, and temperature fluctuations.
• The theory extends to arbitrary Stokes numbers and temperature gradients, unifying weak- and strong-stratification regimes with validation from laboratory experiments and DNS studies.
• This mechanism is applied to atmospheric aerosol distributions and protoplanetary disk dust concentration, though challenges remain in high-Re, multi-phase, and complex flow conditions.

Searching arXiv for recent and foundational papers on turbulent thermal diffusion. arxiv_search(query="turbulent thermal diffusion particles stratified turbulence", max_results=10) Searching arXiv for experimental and DNS studies of turbulent thermal diffusion. arxiv_search(query="experimental study turbulent thermal diffusion inertial particles oscillating grids", max_results=10) Turbulent thermal diffusion most commonly denotes a mean-field transport effect in particle-laden, temperature-stratified turbulence whereby small particles acquire a systematic, non-diffusive drift toward regions of lower mean temperature, typically the temperature minimum. In that usage, the effect is distinct from ordinary turbulent diffusion because it generates particle flux even when the mean particle concentration is spatially uniform, and it is distinct from molecular thermal diffusion because it is produced by turbulence-scale correlations among velocity, temperature, pressure, and particle inertia. The modern formulation unifies weak- and strong-stratification regimes and extends the original small-Stokes-number theory to arbitrary temperature gradients and arbitrary Stokes numbers, with laboratory, atmospheric, and numerical support (Amir et al., 2016).

1. Definition and scope

In the particle-transport literature, turbulent thermal diffusion is the appearance of a non-diffusive turbulent particle flux aligned with the turbulent heat flux and directed opposite to the mean temperature gradient. A standard mean-field decomposition writes the turbulent particle flux as

$$\mathbf{F} \equiv \langle n \mathbf{v} \rangle = N \mathbf{V}_{\text{eff}} - \mathbf{D}_T \nabla N,$$

where $N=\langle n_p\rangle$ is the mean particle number density, $\mathbf{D}_T$ is the turbulent diffusion tensor, and $\mathbf{V}_{\text{eff}}$ is an effective velocity induced by turbulent thermal diffusion. The term $-\mathbf{D}_T\nabla N$ is the familiar diffusive contribution, while $N\mathbf{V}_{\text{eff}}$ is non-diffusive because it does not depend on $\nabla N$ (Amir et al., 2016).

For weak stratification and small Stokes number, the canonical closure is

$$\mathbf{V}_{\text{eff}} = -\alpha D_T \frac{\nabla T}{T},$$

with $\alpha=1$ for non-inertial particles and $\alpha>1$ for inertial particles. In this regime, particles drift toward the mean temperature minimum and can form large-scale inhomogeneities whose scale exceeds the turbulent integral scale (Amir et al., 2016).

A recurrent source of confusion is terminological. In other parts of fluid dynamics and astrophysics, “turbulent thermal diffusion” often refers instead to turbulence-enhanced diffusion of heat itself, quantified by a turbulent thermal diffusivity $N=\langle n_p\rangle$0 or by a turbulent Prandtl number $N=\langle n_p\rangle$1 (Käpylä et al., 2022). In accretion-disc theory it can denote radial turbulent heat transport written in terms of a turbulent potential-temperature flux and a Prandtl number (Owen et al., 2014). The particle-drift phenomenon and the eddy-diffusivity-of-heat usage are related only at the level that both involve turbulent transport in thermally stratified media; they are not the same closure.

2. Physical mechanism in stratified particle-laden turbulence

The mechanism combines temperature stratification with finite particle inertia. For small, heavy particles with $N=\langle n_p\rangle$2 and diameter much smaller than the viscous scale, the leading-order dynamics are governed by Stokes drag,

$N=\langle n_p\rangle$3

and for small Stokes number Maxey’s expansion gives

$N=\langle n_p\rangle$4

Using the fluid momentum equation, one obtains

$N=\langle n_p\rangle$5

Hence even when the carrier flow is incompressible, the particle velocity field is compressible because inertia couples it to pressure fluctuations (Amir et al., 2016).

This compressibility is the key to preferential concentration. Regions with $N=\langle n_p\rangle$6 are regions of convergent particle motion, so particles accumulate there. In isothermal turbulence those compressible features are statistically isotropic and do not produce a net large-scale drift. In temperature-stratified turbulence, however, the turbulent heat flux $N=\langle n_p\rangle$7 correlates velocity, temperature, and pressure fluctuations. Particles therefore sample compressive regions asymmetrically and acquire a mean drift in the direction of the turbulent heat flux, i.e. toward lower mean temperature (Amir et al., 2016).

For non-inertial particles in low-Mach-number anelastic turbulence, the mechanism can be expressed through the mean density gradient. Using $N=\langle n_p\rangle$8,

$N=\langle n_p\rangle$9

which reduces to $\mathbf{D}_T$0 when the mean pressure gradient is negligible (Elmakies et al., 2023). This form shows that the effect does not require finite particle inertia to exist, but inertia amplifies it.

The steady mean-field balance makes the clustering consequence explicit: $\mathbf{D}_T$1 When gravity is negligible and $\mathbf{D}_T$2, the equilibrium concentration increases where $\mathbf{D}_T$3 decreases, so large-scale particle maxima appear near mean temperature minima (Amir et al., 2016).

3. Generalized theory for arbitrary stratification and Stokes number

The 2016 generalization extends the original weak-stratification, small-Stokes-number theory to arbitrary temperature gradients and arbitrary Stokes numbers by modeling second-order statistics of the particle velocity field in stratified turbulence (Amir et al., 2016). In low-Mach-number anelastic flow,

$\mathbf{D}_T$4

and the theory introduces a particle-velocity correlation model with parameters $\mathbf{D}_T$5 and $\mathbf{D}_T$6 that encode inertia and Reynolds-number effects. The resulting effective velocity can be written compactly as

$\mathbf{D}_T$7

Here $\mathbf{D}_T$8 is a dimensionless function obtained by spectral integration with Kolmogorov scaling, $\mathbf{D}_T$9, and $\mathbf{V}_{\text{eff}}$0 is the dimensionless stratification parameter. In the weak-stratification limit $\mathbf{V}_{\text{eff}}$1, $\mathbf{V}_{\text{eff}}$2 and the theory reduces to

$\mathbf{V}_{\text{eff}}$3

with $\mathbf{V}_{\text{eff}}$4. In the strong-stratification limit $\mathbf{V}_{\text{eff}}$5, the effective velocity decays as $\mathbf{V}_{\text{eff}}$6, so strong stratification suppresses turbulent thermal diffusion (Amir et al., 2016).

The same framework predicts an anisotropic turbulent diffusion tensor,

$\mathbf{V}_{\text{eff}}$7

showing that stratification modifies both the drift and the diffusive part of transport. This anisotropy is not a secondary correction; it is intrinsic to strongly stratified turbulence (Amir et al., 2016).

A useful observable is the effective thermal diffusion coefficient $\mathbf{V}_{\text{eff}}$8, where $\mathbf{V}_{\text{eff}}$9. Neglecting gravity, $-\mathbf{D}_T\nabla N$0 decreases with increasing $-\mathbf{D}_T\nabla N$1, and for fixed stratification it increases with Stokes number at small $-\mathbf{D}_T\nabla N$2, reaches a maximum at intermediate $-\mathbf{D}_T\nabla N$3, then decreases at larger $-\mathbf{D}_T\nabla N$4. In laboratory conditions the maximum effective velocity occurs around $-\mathbf{D}_T\nabla N$5, while in atmospheric turbulence it shifts to $-\mathbf{D}_T\nabla N$6, reflecting the different Reynolds numbers and timescale separations (Amir et al., 2016).

Direct numerical simulations provide an independent confirmation of the same non-monotonic inertia dependence. In forced temperature-stratified turbulence, turbulent thermal diffusion causes a peak of particle number density around the minimum of the mean fluid temperature for Stokes numbers less than 1, and the effect is strongest for Stokes numbers around unity before weakening at larger values (Haugen et al., 2011).

4. Experimental and numerical evidence

Laboratory evidence now spans stably stratified, convective, and strongly inhomogeneous forced-convection configurations, while DNS resolves the same flux mechanism in controlled settings. Across these studies, the recurring empirical signature is an anti-correlation between mean particle concentration and mean temperature (Amir et al., 2016).

Study	Configuration	Principal finding
(Amir et al., 2016)	Oscillating-grid and multi-fan turbulence	Generalized theory agrees with measured $-\mathbf{D}_T\nabla N$7 and $-\mathbf{D}_T\nabla N$8
(Haugen et al., 2011)	DNS with and without gravity	Peak particle density forms near temperature minimum; maximum effect near $-\mathbf{D}_T\nabla N$9
(Elmakies et al., 2022)	Inhomogeneous, anisotropic stable turbulence	Particles accumulate at minimum mean temperature; measured $N\mathbf{V}_{\text{eff}}$0 and $N\mathbf{V}_{\text{eff}}$1
(Elmakies et al., 2023)	Inhomogeneous forced convective turbulence	Near-grid $N\mathbf{V}_{\text{eff}}$2, far-field $N\mathbf{V}_{\text{eff}}$3; accumulation persists in non-monotonic $N\mathbf{V}_{\text{eff}}$4 fields
(Elmakies et al., 25 Feb 2026)	Convective turbulence forced by one or two oscillating grids	Effective pumping velocity for $N\mathbf{V}_{\text{eff}}$5 particles is 2.5 times that for $N\mathbf{V}_{\text{eff}}$6 particles
(Zarbib et al., 26 Aug 2025)	Strongly inhomogeneous forced convection, $N\mathbf{V}_{\text{eff}}$7	Particle maxima coincide with local temperature minima; estimated $N\mathbf{V}_{\text{eff}}$8–$N\mathbf{V}_{\text{eff}}$9 cm/s

The 2016 experiments used a $\nabla N$0\,cm$\nabla N$1 oscillating-grid chamber with imposed vertical gradients up to $\nabla N$2 K/cm at mean $\nabla N$3 K, and a $\nabla N$4\,cm$\nabla N$5 multi-fan apparatus with gradients around $\nabla N$6 K/cm. In both devices the measured $\nabla N$7 agreed with generalized-theory curves when a single ratio $\nabla N$8 was fitted for each configuration. The same work connected the theory to tropopause aerosol layers observed in GOMOS satellite data, where aerosol enhancements correlate with temperature minima (Amir et al., 2016).

The 2022 inhomogeneous anisotropic stable-stratification experiments extended detection beyond nearly homogeneous grid turbulence. Using a single oscillating grid in a $\nabla N$9\,cm chamber, PIV and Mie scattering showed accumulation of $$\mathbf{V}_{\text{eff}} = -\alpha D_T \frac{\nabla T}{T},$$0 particles near the cold bottom wall, with effective coefficients $$\mathbf{V}_{\text{eff}} = -\alpha D_T \frac{\nabla T}{T},$$1 in the region $$\mathbf{V}_{\text{eff}} = -\alpha D_T \frac{\nabla T}{T},$$2–15 cm and $$\mathbf{V}_{\text{eff}} = -\alpha D_T \frac{\nabla T}{T},$$3 for $$\mathbf{V}_{\text{eff}} = -\alpha D_T \frac{\nabla T}{T},$$4–24 cm (Elmakies et al., 2022).

The 2023 convective experiments demonstrated that the effect survives strongly non-monotonic temperature fields. In a convective arrangement with $$\mathbf{V}_{\text{eff}} = -\alpha D_T \frac{\nabla T}{T},$$5 K, the mean temperature profile could decrease, increase, and decrease again with height depending on distance from the grid, yet particle concentration still increased wherever the mean temperature decreased. Linear fits of $$\mathbf{V}_{\text{eff}} = -\alpha D_T \frac{\nabla T}{T},$$6 versus $$\mathbf{V}_{\text{eff}} = -\alpha D_T \frac{\nabla T}{T},$$7 yielded $$\mathbf{V}_{\text{eff}} = -\alpha D_T \frac{\nabla T}{T},$$8 near the grid and $$\mathbf{V}_{\text{eff}} = -\alpha D_T \frac{\nabla T}{T},$$9 farther away (Elmakies et al., 2023).

The 2026 oscillating-grid convective study sharpened the inertia contrast: for $\alpha=1$0 inertial particles, the effective pumping velocity responsible for large-scale clustering was reported as 2.5 times larger than for $\alpha=1$1 non-inertial particles, consistent with the prediction that $\alpha=1$2 for inertial particles (Elmakies et al., 25 Feb 2026).

DNS complements the experiments by isolating the mean-field mechanism. In simulations with and without gravity, turbulent thermal diffusion caused a peak in particle number density around the mean temperature minimum for $\alpha=1$3, and the strength of the effect decreased again for larger $\alpha=1$4. Gravity altered the profile shape but did not remove the temperature-minimum clustering as long as settling did not dominate (Haugen et al., 2011).

5. Relation to adjacent transport mechanisms and to other uses of the term

Turbulent thermal diffusion in the particle sense is often conflated with several distinct mechanisms. Ordinary turbulent diffusion produces a flux $\alpha=1$5 and homogenizes concentration; it does not by itself create maxima of $\alpha=1$6 at temperature minima. Thermophoresis and molecular thermal diffusion are microscopic effects driven by molecular transport coefficients and are usually weak for micron-size particles in air compared with the turbulent mechanism. Turbophoresis is an inertial drift in inhomogeneous turbulence toward lower turbulence intensity, not a drift controlled by $\alpha=1$7. The 2023 convective experiments explicitly distinguished the observed clustering from pure gravitational settling and from turbophoresis, arguing that the dominant effect in the strong-gradient regions was turbulent thermal diffusion (Elmakies et al., 2023).

A second misconception is terminological. In homogeneous-turbulence studies of heat transport, the same phrase can mean the enhancement of thermal transport by eddies, expressed through $\alpha=1$8 and $\alpha=1$9. In isotropically forced homogeneous turbulence, $\alpha>1$0 approaches $\alpha>1$1 at sufficiently large Péclet number and the turbulent Prandtl number approaches $\alpha>1$2, largely independent of microscopic $\alpha>1$3 over $\alpha>1$4 (Käpylä et al., 2022). That literature concerns turbulent diffusion of heat, not non-diffusive particle pumping.

Related astrophysical literatures use the term even more broadly. In protoplanetary discs it can denote radial turbulent heat transport parameterized through a potential-temperature flux and a Prandtl number, with consequences for the gravo-magnetic limit cycle (Owen et al., 2014). In strongly stratified galaxy-cluster turbulence, turbulent heat diffusion is suppressed by stratification, scaling as $\alpha>1$5, while dissipation scales as $\alpha>1$6 (Wang et al., 2022). In diffusive stratified shear flows relevant to stellar radiation zones, thermal diffusion weakens buoyancy, enabling secular shear instability and yielding effective compositional mixing $\alpha>1$7 in an intermediate regime (Garaud et al., 2015, Garaud et al., 2016). These are important neighboring topics, but they address eddy diffusivities of heat or composition, not the non-diffusive particle drift toward temperature minima.

6. Applications, limitations, and open problems

The most developed application is atmospheric aerosol transport. The generalized theory has been used to interpret aerosol layers near the tropopause, where observed concentration maxima correlate with temperature minima. Under typical atmospheric conditions, $\alpha>1$8 is of order unity for common aerosol sizes $\alpha>1$9–$N=\langle n_p\rangle$00 and turbulent diffusion $N=\langle n_p\rangle$01 cm$N=\langle n_p\rangle$02/s, sufficient to shape large-scale vertical aerosol distributions even though $N=\langle n_p\rangle$03 (Amir et al., 2016).

Cloud physics and environmental flows are natural extensions. In stratified cloud-topped boundary layers, heat exchangers, combustion systems, and pollutant plumes, the mechanism provides a directed transport channel beyond down-gradient diffusion. A plausible implication is that neglecting it biases predictions of deposition, residence time, and concentration maxima whenever turbulence and persistent temperature gradients coexist, although the detailed magnitude remains flow-specific (Amir et al., 2016).

Astrophysical applications include dust concentration in protoplanetary discs. Recasting the theory in disc language, one study derived a mean drift velocity

$N=\langle n_p\rangle$04

and argued that in local, quasi-isobaric cold annuli the steady-state dust distribution $N=\langle n_p\rangle$05 can yield concentration factors from $N=\langle n_p\rangle$06 up to $N=\langle n_p\rangle$07, potentially sufficient to trigger streaming instability and planetesimal formation (Hubbard, 2015). That application is explicitly local: the same work concluded that smooth global disc temperature gradients are generally too shallow relative to pressure gradients for turbulent thermal diffusion to dominate radial drift.

The principal limitations are also clear. Existing laboratory and DNS studies occupy moderate Reynolds numbers and, in the DNS, Mach numbers around $N=\langle n_p\rangle$08, far from the very low-Mach, very high-Reynolds-number regimes of the atmosphere. Most analyses assume dilute suspensions with one-way coupling, neglecting feedback of particles on turbulence, particle collisions, shape effects, and detailed non-Stokes drag. The generalized theory for strong stratification introduces empirical parameters through $N=\langle n_p\rangle$09 and $N=\langle n_p\rangle$10, and the ratio $N=\langle n_p\rangle$11 is fitted in each flow configuration rather than predicted ab initio (Amir et al., 2016).

Open problems therefore center on closure and regime extension: systematic mapping of $N=\langle n_p\rangle$12 across Stokes, Reynolds, and Péclet numbers; separation of turbulent thermal diffusion from turbophoresis in strongly inhomogeneous turbulence; extension to rotating, sheared, or magnetized flows; and quantitative coupling to condensation, evaporation, chemistry, or dust back-reaction. What the existing literature establishes, however, is already precise: turbulent thermal diffusion is a statistically robust, non-diffusive particle-transport mechanism generated by the interplay of stratification, turbulence, and particle inertia, and it produces particle accumulation at mean temperature minima across DNS, laboratory flows, and atmospheric-scale interpretations (Amir et al., 2016).