Stochastic One-Dimensional Turbulence (ODT)

• Stochastic ODT is a turbulence model that represents 3D turbulent advection along a one-dimensional line with intermittent, map-based eddy events.
• It utilizes triplet-map dynamics to model instantaneous eddy events, enabling detailed resolution of velocity, scalar, and phase fields in various flow configurations.
• The framework is adaptable as a stand-alone model or embedded in LES (ODTLES), accurately capturing transport properties, interface dynamics, and scaling behavior.

Stochastic One-Dimensional Turbulence (ODT) is a map-based, stochastic turbulence model that represents three-dimensional turbulent advection along a notional one-dimensional line of sight while advancing molecular transport deterministically. In ODT, the resolved fields on that line can include the three velocity components and, depending on the application, temperature, passive scalars, or a phase index; turbulence is represented by instantaneous eddy events, usually triplet maps, augmented by kernel terms that enforce momentum conservation and pressure-mediated energy redistribution. In stand-alone form the ODT line is typically aligned with the dominant wall-normal, vertical, or radial transport direction, whereas in ODTLES it is embedded in a coarse three-dimensional LES grid as a small-scale closure (Klein et al., 2019, Klein et al., 2019, Movaghar et al., 2024, Glawe et al., 2015).

1. Conceptual framework and scope

ODT advances a velocity vector and, when relevant, scalar or phase fields on a single spatial coordinate by combining deterministic one-dimensional diffusion with discrete stochastic eddy events that mimic turbulent advection and pressure redistribution. In turbulent channel flow the ODT domain is a wall-normal line $y\in[0,2\delta]$; in confined planar jets it is the wall-normal coordinate $y$; in Rayleigh–Bénard convection it is a vertical wall-normal line $z\in[0,L]$; in concentric annuli it is the radial coordinate $r\in[R_i,R_o]$; and in the multiphase homogeneous-isotropic setting it is a line of sight normal to the initial planar interface (Klein et al., 2021, Klein et al., 2019, Klein et al., 2019, Klein et al., 2023, Movaghar et al., 2024).

The reduction to one dimension does not remove the three velocity components. Rather, ODT retains $(u,v,w)$ or their geometry-specific analogues on the line and models the net action of turbulent stirring by intermittent mappings. In wall-bounded flows this makes the model particularly focused on wall-normal transport; in buoyant convection it isolates vertical plume-mediated transport; in annular geometry it resolves radial transport under spanwise curvature; and in multiphase flow it affords high resolution of interface creation and property gradients within each phase (Klein et al., 2021, Klein et al., 2019, Klein et al., 2023, Movaghar et al., 2024).

A recurring misconception is to treat ODT as a purely diffusive or eddy-viscosity closure. The cited studies instead formulate it as an event-driven advection model: between events, only molecular diffusion is integrated; at event times, turbulent advection is represented by instantaneous conservative remappings. This distinction is central to its ability to reproduce direct-cascade behavior, plume-like displacements, scalar microstructure, and interfacial corrugation with a one-dimensional state representation (Klein et al., 2019, Klein et al., 2019, Movaghar et al., 2024).

2. Governing equations and triplet-map dynamics

For wall-bounded channel flow, the stand-alone formulation is written as stochastic conservation equations for momentum and a passive scalar,

$$\frac{\partial \boldsymbol{u}}{\partial t} +\sum_{t_e}\,\mathcal{E}_{\boldsymbol{u}}(\boldsymbol{u})\,\tilde{\delta}(t-t_e) =\nu\,\frac{\partial^2 \boldsymbol{u}}{\partial y^2} -\frac{1}{\rho}\,\frac{d\bar{p}}{dx}\,\boldsymbol{e}_x,$$

$$\frac{\partial \theta}{\partial t} +\sum_{t_e}\,\mathcal{E}_{\theta}(\boldsymbol{u})\,\tilde{\delta}(t-t_e) =\Gamma\,\frac{\partial^2 \theta}{\partial y^2}+s_\theta,$$

with no-slip walls for velocity and either constant-scalar-value or constant-scalar-flux forcing for the scalar (Klein et al., 2021). In Rayleigh–Bénard convection the same deterministic–stochastic split is used for the three velocity components and temperature on the vertical coordinate,

$$\frac{\partial u_i}{\partial t} + \mathcal{E}_{i}(\alpha) = \nu\,\frac{\partial^2 u_i}{\partial z^2}, \qquad \frac{\partial T}{\partial t} + \mathcal{E}_T = \kappa\,\frac{\partial^2 T}{\partial z^2},$$

with buoyancy entering the eddy energetics and component coupling (Klein et al., 2019).

The elementary advective event is the triplet map. For an eddy interval $[y_0,y_0+\ell]$, the inverse map used in several formulations is

$$f(y)=y_0+\begin{cases} 3(y-y_0),&y-y_0\in[0,\ell/3],\ 2\ell-3(y-y_0),&y-y_0\in[\ell/3,2\ell/3],\ 3(y-y_0)-2\ell,&y-y_0\in[2\ell/3,\ell],\ y-y_0,&\text{otherwise.} \end{cases}$$

It preserves measure and continuity, compresses the affected interval by a factor of three, and reproduces the profile three times with the middle copy reversed, thereby steepening gradients and directing energy toward small scales (Klein et al., 2019, Klein et al., 2021).

The mapped update distinguishes scalars from velocity. In the single-phase formulations,

$y$0

with kernel $y$1. The coefficients $y$2 model rapid pressure–velocity coupling and inter-component energy redistribution while conserving total kinetic energy over the eddy interval:

$y$3

where $y$4 and $y$5 (Klein et al., 2019, Klein et al., 2021). In the multiphase formulation, the eddy update includes an additional $y$6 term with $y$7, so that the instantaneous velocity map becomes

$y$8

which is used to encode surface-tension energetics and return-to-isotropy while preserving momentum along the line (Movaghar et al., 2024).

The same structural machinery extends to cylindrical and buoyant settings. In annular flow the deterministic operators become cylindrical diffusion operators in $y$9, and the triplet map acts on radial intervals; in Rayleigh–Bénard convection the eddy coefficients are modified by buoyancy through

$z\in[0,L]$0

with $z\in[0,L]$1 distributing released potential energy among velocity components (Klein et al., 2023, Klein et al., 2019).

3. Eddy-rate models, energetics, and numerical realization

The event sequence is sampled from a Poisson process or from thinning-and-rejection constructions of a marked Poisson process. A common rate density is

$z\in[0,L]$2

where $z\in[0,L]$3 controls overall eddy activity and $z\in[0,L]$4 is a local eddy turnover time determined from the instantaneous line state (Klein et al., 2021, Movaghar et al., 2024).

In wall-bounded single-phase flows the turnover time is based on available shear energy and a viscous penalty,

$z\in[0,L]$5

with $z\in[0,L]$6 suppressing unphysically small eddies; in confined planar jets an additional large-eddy suppression condition $z\in[0,L]$7 is used to reject oversized eddies in the developing near-inlet region (Klein et al., 2021, Klein et al., 2019). In Rayleigh–Bénard convection the rate acquires a buoyancy contribution,

$z\in[0,L]$8

so that unstable stratification can enhance event occurrence while viscous damping still cuts off sub-Kolmogorov eddies (Klein et al., 2019).

The multiphase extension modifies the event energetics by treating surface tension as an energy sink. There the accepted-event criterion is based on

$z\in[0,L]$9

with $r\in[R_i,R_o]$0 determined from the interface-area increase implied by the triplet map. For a single interface inside the eddy interval, the surface-energy increase per unit mass is

$r\in[R_i,R_o]$1

If $r\in[R_i,R_o]$2, the eddy is energetically forbidden; if $r\in[R_i,R_o]$3 but the viscous term makes $r\in[R_i,R_o]$4, the eddy is suppressed (Movaghar et al., 2024).

The numerical realization is typically finite-volume and adaptive. Molecular diffusion is advanced explicitly between instantaneous events, and the mesh is refined dynamically to resolve thin boundary layers, small eddies, Batchelor scales, or interface gradients. In the planar-jet study $r\in[R_i,R_o]$5 is set $r\in[R_i,R_o]$6 with $r\in[R_i,R_o]$7; in channel-flow scalar transport the minimum cell size is chosen below the Kolmogorov or Batchelor scales, with reported $r\in[R_i,R_o]$8–0.4 depending on $r\in[R_i,R_o]$9 and $(u,v,w)$0 (Klein et al., 2019, Klein et al., 2021).

4. Canonical applications and quantitative behavior

In high-$(u,v,w)$1 Rayleigh–Bénard convection, ODT was used in a planar Boussinesq configuration with smooth walls and infinite aspect ratio to reach $(u,v,w)$2 for $(u,v,w)$3 and $(u,v,w)$4 for $(u,v,w)$5 on workstations. With parameters calibrated once in the classical regime and then held fixed, the model reproduced effective Nusselt scalings $(u,v,w)$6 over eight decades in $(u,v,w)$7: for $(u,v,w)$8, $(u,v,w)$9 over $$\frac{\partial \boldsymbol{u}}{\partial t} +\sum_{t_e}\,\mathcal{E}_{\boldsymbol{u}}(\boldsymbol{u})\,\tilde{\delta}(t-t_e) =\nu\,\frac{\partial^2 \boldsymbol{u}}{\partial y^2} -\frac{1}{\rho}\,\frac{d\bar{p}}{dx}\,\boldsymbol{e}_x,$$0 and $$\frac{\partial \boldsymbol{u}}{\partial t} +\sum_{t_e}\,\mathcal{E}_{\boldsymbol{u}}(\boldsymbol{u})\,\tilde{\delta}(t-t_e) =\nu\,\frac{\partial^2 \boldsymbol{u}}{\partial y^2} -\frac{1}{\rho}\,\frac{d\bar{p}}{dx}\,\boldsymbol{e}_x,$$1 over $$\frac{\partial \boldsymbol{u}}{\partial t} +\sum_{t_e}\,\mathcal{E}_{\boldsymbol{u}}(\boldsymbol{u})\,\tilde{\delta}(t-t_e) =\nu\,\frac{\partial^2 \boldsymbol{u}}{\partial y^2} -\frac{1}{\rho}\,\frac{d\bar{p}}{dx}\,\boldsymbol{e}_x,$$2; for $$\frac{\partial \boldsymbol{u}}{\partial t} +\sum_{t_e}\,\mathcal{E}_{\boldsymbol{u}}(\boldsymbol{u})\,\tilde{\delta}(t-t_e) =\nu\,\frac{\partial^2 \boldsymbol{u}}{\partial y^2} -\frac{1}{\rho}\,\frac{d\bar{p}}{dx}\,\boldsymbol{e}_x,$$3, $$\frac{\partial \boldsymbol{u}}{\partial t} +\sum_{t_e}\,\mathcal{E}_{\boldsymbol{u}}(\boldsymbol{u})\,\tilde{\delta}(t-t_e) =\nu\,\frac{\partial^2 \boldsymbol{u}}{\partial y^2} -\frac{1}{\rho}\,\frac{d\bar{p}}{dx}\,\boldsymbol{e}_x,$$4 over $$\frac{\partial \boldsymbol{u}}{\partial t} +\sum_{t_e}\,\mathcal{E}_{\boldsymbol{u}}(\boldsymbol{u})\,\tilde{\delta}(t-t_e) =\nu\,\frac{\partial^2 \boldsymbol{u}}{\partial y^2} -\frac{1}{\rho}\,\frac{d\bar{p}}{dx}\,\boldsymbol{e}_x,$$5, $$\frac{\partial \boldsymbol{u}}{\partial t} +\sum_{t_e}\,\mathcal{E}_{\boldsymbol{u}}(\boldsymbol{u})\,\tilde{\delta}(t-t_e) =\nu\,\frac{\partial^2 \boldsymbol{u}}{\partial y^2} -\frac{1}{\rho}\,\frac{d\bar{p}}{dx}\,\boldsymbol{e}_x,$$6 at very low $$\frac{\partial \boldsymbol{u}}{\partial t} +\sum_{t_e}\,\mathcal{E}_{\boldsymbol{u}}(\boldsymbol{u})\,\tilde{\delta}(t-t_e) =\nu\,\frac{\partial^2 \boldsymbol{u}}{\partial y^2} -\frac{1}{\rho}\,\frac{d\bar{p}}{dx}\,\boldsymbol{e}_x,$$7, and $$\frac{\partial \boldsymbol{u}}{\partial t} +\sum_{t_e}\,\mathcal{E}_{\boldsymbol{u}}(\boldsymbol{u})\,\tilde{\delta}(t-t_e) =\nu\,\frac{\partial^2 \boldsymbol{u}}{\partial y^2} -\frac{1}{\rho}\,\frac{d\bar{p}}{dx}\,\boldsymbol{e}_x,$$8 over $$\frac{\partial \boldsymbol{u}}{\partial t} +\sum_{t_e}\,\mathcal{E}_{\boldsymbol{u}}(\boldsymbol{u})\,\tilde{\delta}(t-t_e) =\nu\,\frac{\partial^2 \boldsymbol{u}}{\partial y^2} -\frac{1}{\rho}\,\frac{d\bar{p}}{dx}\,\boldsymbol{e}_x,$$9. The transition thresholds were $$\frac{\partial \theta}{\partial t} +\sum_{t_e}\,\mathcal{E}_{\theta}(\boldsymbol{u})\,\tilde{\delta}(t-t_e) =\Gamma\,\frac{\partial^2 \theta}{\partial y^2}+s_\theta,$$0 for $$\frac{\partial \theta}{\partial t} +\sum_{t_e}\,\mathcal{E}_{\theta}(\boldsymbol{u})\,\tilde{\delta}(t-t_e) =\Gamma\,\frac{\partial^2 \theta}{\partial y^2}+s_\theta,$$1 and $$\frac{\partial \theta}{\partial t} +\sum_{t_e}\,\mathcal{E}_{\theta}(\boldsymbol{u})\,\tilde{\delta}(t-t_e) =\Gamma\,\frac{\partial^2 \theta}{\partial y^2}+s_\theta,$$2 for $$\frac{\partial \theta}{\partial t} +\sum_{t_e}\,\mathcal{E}_{\theta}(\boldsymbol{u})\,\tilde{\delta}(t-t_e) =\Gamma\,\frac{\partial^2 \theta}{\partial y^2}+s_\theta,$$3, and the post-transition regime was associated with a shift of the temperature–velocity cross-correlation peak from the boundary-layer edge into the bulk, consistent with Kraichnan’s picture of ultimate convection (Klein et al., 2019).

In turbulent channel flow with passive scalar transport, ODT was calibrated once to the velocity boundary layer and then used across broad $$\frac{\partial \theta}{\partial t} +\sum_{t_e}\,\mathcal{E}_{\theta}(\boldsymbol{u})\,\tilde{\delta}(t-t_e) =\Gamma\,\frac{\partial^2 \theta}{\partial y^2}+s_\theta,$$4, $$\frac{\partial \theta}{\partial t} +\sum_{t_e}\,\mathcal{E}_{\theta}(\boldsymbol{u})\,\tilde{\delta}(t-t_e) =\Gamma\,\frac{\partial^2 \theta}{\partial y^2}+s_\theta,$$5, and $$\frac{\partial \theta}{\partial t} +\sum_{t_e}\,\mathcal{E}_{\theta}(\boldsymbol{u})\,\tilde{\delta}(t-t_e) =\Gamma\,\frac{\partial^2 \theta}{\partial y^2}+s_\theta,$$6 ranges. The model reproduced state-space statistics of the surface scalar-flux fluctuations and mean scalar transfer quantified by the Sherwood number. For high asymptotic $$\frac{\partial \theta}{\partial t} +\sum_{t_e}\,\mathcal{E}_{\theta}(\boldsymbol{u})\,\tilde{\delta}(t-t_e) =\Gamma\,\frac{\partial^2 \theta}{\partial y^2}+s_\theta,$$7 and $$\frac{\partial \theta}{\partial t} +\sum_{t_e}\,\mathcal{E}_{\theta}(\boldsymbol{u})\,\tilde{\delta}(t-t_e) =\Gamma\,\frac{\partial^2 \theta}{\partial y^2}+s_\theta,$$8, the predicted $$\frac{\partial \theta}{\partial t} +\sum_{t_e}\,\mathcal{E}_{\theta}(\boldsymbol{u})\,\tilde{\delta}(t-t_e) =\Gamma\,\frac{\partial^2 \theta}{\partial y^2}+s_\theta,$$9 lay between the Dittus–Boelter scaling $$\frac{\partial u_i}{\partial t} + \mathcal{E}_{i}(\alpha) = \nu\,\frac{\partial^2 u_i}{\partial z^2}, \qquad \frac{\partial T}{\partial t} + \mathcal{E}_T = \kappa\,\frac{\partial^2 T}{\partial z^2},$$0 and the Colburn scaling $$\frac{\partial u_i}{\partial t} + \mathcal{E}_{i}(\alpha) = \nu\,\frac{\partial^2 u_i}{\partial z^2}, \qquad \frac{\partial T}{\partial t} + \mathcal{E}_T = \kappa\,\frac{\partial^2 T}{\partial z^2},$$1, but closer to the former; for finite $$\frac{\partial u_i}{\partial t} + \mathcal{E}_{i}(\alpha) = \nu\,\frac{\partial^2 u_i}{\partial z^2}, \qquad \frac{\partial T}{\partial t} + \mathcal{E}_T = \kappa\,\frac{\partial^2 T}{\partial z^2},$$2 and $$\frac{\partial u_i}{\partial t} + \mathcal{E}_{i}(\alpha) = \nu\,\frac{\partial^2 u_i}{\partial z^2}, \qquad \frac{\partial T}{\partial t} + \mathcal{E}_T = \kappa\,\frac{\partial^2 T}{\partial z^2},$$3, the predictions reproduced the Schwertfirm–Manhart relation; and in the diffusive limit the model extrapolated to $$\frac{\partial u_i}{\partial t} + \mathcal{E}_{i}(\alpha) = \nu\,\frac{\partial^2 u_i}{\partial z^2}, \qquad \frac{\partial T}{\partial t} + \mathcal{E}_T = \kappa\,\frac{\partial^2 T}{\partial z^2},$$4 as $$\frac{\partial u_i}{\partial t} + \mathcal{E}_{i}(\alpha) = \nu\,\frac{\partial^2 u_i}{\partial z^2}, \qquad \frac{\partial T}{\partial t} + \mathcal{E}_T = \kappa\,\frac{\partial^2 T}{\partial z^2},$$5 for constant-scalar-value forcing, with the reported low-$$\frac{\partial u_i}{\partial t} + \mathcal{E}_{i}(\alpha) = \nu\,\frac{\partial^2 u_i}{\partial z^2}, \qquad \frac{\partial T}{\partial t} + \mathcal{E}_T = \kappa\,\frac{\partial^2 T}{\partial z^2},$$6 fit $$\frac{\partial u_i}{\partial t} + \mathcal{E}_{i}(\alpha) = \nu\,\frac{\partial^2 u_i}{\partial z^2}, \qquad \frac{\partial T}{\partial t} + \mathcal{E}_T = \kappa\,\frac{\partial^2 T}{\partial z^2},$$7 (Klein et al., 2021). A closely related study of scalar transfer to a wall reported that, after calibration at $$\frac{\partial u_i}{\partial t} + \mathcal{E}_{i}(\alpha) = \nu\,\frac{\partial^2 u_i}{\partial z^2}, \qquad \frac{\partial T}{\partial t} + \mathcal{E}_T = \kappa\,\frac{\partial^2 T}{\partial z^2},$$8 with $$\frac{\partial u_i}{\partial t} + \mathcal{E}_{i}(\alpha) = \nu\,\frac{\partial^2 u_i}{\partial z^2}, \qquad \frac{\partial T}{\partial t} + \mathcal{E}_T = \kappa\,\frac{\partial^2 T}{\partial z^2},$$9, $[y_0,y_0+\ell]$0, and $[y_0,y_0+\ell]$1, the model captured the exact low-$[y_0,y_0+\ell]$2 diffusive limit $[y_0,y_0+\ell]$3, the intermediate-$[y_0,y_0+\ell]$4 Schwertfirm–Manhart behavior with fitted $[y_0,y_0+\ell]$5, $[y_0,y_0+\ell]$6, $[y_0,y_0+\ell]$7, and a high-$[y_0,y_0+\ell]$8 asymptote $[y_0,y_0+\ell]$9 corresponding to a near-wall eddy-diffusivity exponent $$f(y)=y_0+\begin{cases} 3(y-y_0),&y-y_0\in[0,\ell/3],\ 2\ell-3(y-y_0),&y-y_0\in[\ell/3,2\ell/3],\ 3(y-y_0)-2\ell,&y-y_0\in[2\ell/3,\ell],\ y-y_0,&\text{otherwise.} \end{cases}$$0 (Klein et al., 2020).

In confined planar jets, stand-alone ODT reproduced mean momentum transport and Reynolds stress with fixed parameters at $$f(y)=y_0+\begin{cases} 3(y-y_0),&y-y_0\in[0,\ell/3],\ 2\ell-3(y-y_0),&y-y_0\in[\ell/3,2\ell/3],\ 3(y-y_0)-2\ell,&y-y_0\in[2\ell/3,\ell],\ y-y_0,&\text{otherwise.} \end{cases}$$1 and $$f(y)=y_0+\begin{cases} 3(y-y_0),&y-y_0\in[0,\ell/3],\ 2\ell-3(y-y_0),&y-y_0\in[\ell/3,2\ell/3],\ 3(y-y_0)-2\ell,&y-y_0\in[2\ell/3,\ell],\ y-y_0,&\text{otherwise.} \end{cases}$$2, although the streamwise and wall-normal rms velocity fluctuations were systematically underestimated by about $$f(y)=y_0+\begin{cases} 3(y-y_0),&y-y_0\in[0,\ell/3],\ 2\ell-3(y-y_0),&y-y_0\in[\ell/3,2\ell/3],\ 3(y-y_0)-2\ell,&y-y_0\in[2\ell/3,\ell],\ y-y_0,&\text{otherwise.} \end{cases}$$3–$$f(y)=y_0+\begin{cases} 3(y-y_0),&y-y_0\in[0,\ell/3],\ 2\ell-3(y-y_0),&y-y_0\in[\ell/3,2\ell/3],\ 3(y-y_0)-2\ell,&y-y_0\in[2\ell/3,\ell],\ y-y_0,&\text{otherwise.} \end{cases}$$4, approximately a factor $$f(y)=y_0+\begin{cases} 3(y-y_0),&y-y_0\in[0,\ell/3],\ 2\ell-3(y-y_0),&y-y_0\in[\ell/3,2\ell/3],\ 3(y-y_0)-2\ell,&y-y_0\in[2\ell/3,\ell],\ y-y_0,&\text{otherwise.} \end{cases}$$5. For passive-scalar transport at $$f(y)=y_0+\begin{cases} 3(y-y_0),&y-y_0\in[0,\ell/3],\ 2\ell-3(y-y_0),&y-y_0\in[\ell/3,2\ell/3],\ 3(y-y_0)-2\ell,&y-y_0\in[2\ell/3,\ell],\ y-y_0,&\text{otherwise.} \end{cases}$$6 and $$f(y)=y_0+\begin{cases} 3(y-y_0),&y-y_0\in[0,\ell/3],\ 2\ell-3(y-y_0),&y-y_0\in[\ell/3,2\ell/3],\ 3(y-y_0)-2\ell,&y-y_0\in[2\ell/3,\ell],\ y-y_0,&\text{otherwise.} \end{cases}$$7, resolving the Batchelor scale led to scalar fluctuation variances up to ten times larger than in the under-resolved references near the splitter wakes, and the scalar spectrum at $$f(y)=y_0+\begin{cases} 3(y-y_0),&y-y_0\in[0,\ell/3],\ 2\ell-3(y-y_0),&y-y_0\in[\ell/3,2\ell/3],\ 3(y-y_0)-2\ell,&y-y_0\in[2\ell/3,\ell],\ y-y_0,&\text{otherwise.} \end{cases}$$8 exhibited a Batchelor-like regime that was absent at $$f(y)=y_0+\begin{cases} 3(y-y_0),&y-y_0\in[0,\ell/3],\ 2\ell-3(y-y_0),&y-y_0\in[\ell/3,2\ell/3],\ 3(y-y_0)-2\ell,&y-y_0\in[2\ell/3,\ell],\ y-y_0,&\text{otherwise.} \end{cases}$$9. The study interpreted the reference-data agreement at nominally high $y$00 as an effect of implicit filtering that acts similarly to a reduced Schmidt number (Klein et al., 2019).

In concentric annuli, cylindrical ODT has been used both for passive heat transfer and for momentum transfer under spanwise curvature. For heated concentric coaxial pipe flow at $y$01, the model captured spanwise curvature and finite Reynolds number effects with fixed adjustable ODT parameters, reproduced the inner–outer asymmetry of the mean perturbation temperature, and resolved the geometry-dependent structure of the turbulent radial heat flux. For $y$02, the high-resolution LES case with $y$03 cells required $y$04 h user time for $y$05 advective time units on $y$06 CPU cores, whereas the standalone adaptive ODT simulation required $y$07 h on a single core (Klein et al., 2023). In later annular momentum simulations, standalone ODT was calibrated at $y$08 for $y$09 and $y$10, then used up to $y$11; it showed that spanwise wall-curvature effects remain sensible in the momentum boundary layer, especially near the convex inner wall, and yielded curvature-aware corrections to the law of the wall in both viscous- and Reynolds-stress-dominated regions (Tsai et al., 13 Aug 2025).

In decaying turbulent interfacial flow, the multiphase ODT formulation was validated against DNS using interface-number density and same-phase probability statistics. With $y$12 and $y$13 tuned to the $y$14 homogeneous-isotropic-turbulence baseline, the model reproduced the trends and parameter dependencies of the Kolmogorov critical scale beyond the DNS-accessible regime. After shifting to the median interface location, the ODT and DNS interface-density profiles aligned much better in shape, and the normalized PDF $y$15 of local critical-scale fluctuations collapsed to a universal curve across both inertial and dissipative cascade sub-ranges, with tails showing an apparent power-law decay $y$16 for sufficiently large $y$17 (Movaghar et al., 2024).

5. ODT as a multiscale closure: ODTLES and XLES

ODT can be embedded in a coarse three-dimensional solver through ODTLES. In the XLES formulation, three mutually overlapping two-dimensionally filtered grids are used, each highly resolved in one Cartesian direction and carrying ODT lines aligned with that direction. ODT then supplies the subgrid turbulent advection and linewise molecular diffusion, while the coarse three-dimensional fields are linked to the line-resolved fields through upscaling and deconvolution operators. In this construction, ODTLES does not close subgrid stresses with eddy viscosity; instead, the resolved fine-scale advection and diffusion on the ODT lines provide the modeled fluxes (Glawe et al., 2015).

The 2015 XLES-to-ODTLES formulation studied turbulent channel and duct flows up to $y$18. It set the maximum eddy size on each ODT line to the coarse LES cell size, $y$19, so that the ODT/LES separation is tied directly to the three-dimensional grid scale. Reported channel-flow simulations used $y$20, $y$21, and recovered the law of the wall, laminar sublayer, rms profiles, and turbulent-kinetic-energy budgets at coarse resolutions for which unclosed XLES did not capture the near-wall structure (Glawe et al., 2015).

A later development introduced an IMEX time-advancement scheme to remove the fine-grid CFL restriction that had degraded the multiscale advantage. The stiff advection in the finely resolved line direction is treated implicitly, while ODT eddies and the remaining non-stiff terms are advanced explicitly. The resulting time-step constraint is LES-based,

$y$22

with $y$23 used in the channel-flow tests, instead of the earlier fine-grid restriction based on $y$24. At $y$25, the IMEX-ODTLES run required $y$26 time steps and $y$27 minutes, compared with $y$28 steps and $y$29 minutes for the CN-RK3 reference at coarse-grid $y$30; despite a per-step cost roughly $y$31 larger, the IMEX scheme achieved about a tenfold speedup overall and was reported stable and accurate up to $y$32 on a single Banana Pi M64 (Glawe et al., 2018).

6. Assumptions, artifacts, and interpretive boundaries

Stand-alone ODT resolves turbulence only along a single line, so mean and large-scale circulation are not explicitly captured. In Rayleigh–Bénard convection this means that the large-scale mean circulation is absent and the mean velocity is zero by construction; the same study assumes Boussinesq conditions, smooth planar walls, and $y$33. In multiphase decaying homogeneous-isotropic turbulence, ODT has no mechanism for lateral non-vortical displacements and no mechanism to decrease interface area, so comparisons to DNS were restricted to early-time growth and improved by shifting statistics to the median interface location. The Rayleigh–Bénard study also documents a weak undulating feature due to the triplet map in $y$34 and $y$35 at intermediate wall distances (Klein et al., 2019, Movaghar et al., 2024).

Additional limitations are application-dependent. In confined planar jets, $y$36 and $y$37 are systematically underestimated by about $y$38–$y$39; in turbulent channel flow, near-wall scalar-fluctuation peaks are underpredicted at very high $y$40 and the turbulent Schmidt number remains closer to unity than in DNS; and in heated annuli the near-wall scalar fluctuation peak $y$41 at $y$42 is underestimated, while very small radius ratios show bulk deviations attributed to unresolved three-dimensional structures around thin inner cylinders (Klein et al., 2019, Klein et al., 2021, Klein et al., 2023).

These constraints delimit direct comparability rather than invalidating the model. The same literature shows that fixed-parameter ODT can nevertheless reproduce low-order transport statistics, scaling regimes, and near-wall structure across substantial ranges of $y$43, $y$44, $y$45, $y$46, $y$47, $y$48, and geometric curvature. This suggests that ODT is most reliable when the dominant unresolved physics is wall-normal, vertical, or radial transport with strong intermittency and broad scale separation, and when missing three-dimensional coherent structures are either secondary or supplied by an embedding framework such as ODTLES (Klein et al., 2020, Glawe et al., 2015).