Novel Subgrid Characteristic Length

• The topic introduces a novel subgrid characteristic length that defines the representative scale for subgrid stresses in coarse-grained simulations.
• It integrates flow-dependent and geometry-aware formulations, such as Δₗₛq, Δ_RLS, and Liutex-based A_R, to address limitations of classical isotropic scales.
• These advances improve simulation fidelity on anisotropic, unstructured, and wall-refined meshes while ensuring computational simplicity and robustness.

A novel subgrid characteristic length is a physically or mathematically motivated definition for the representative length scale governing subgrid stress, flux, or closure terms in coarse-grained simulations such as Large-Eddy Simulations (LES), direct numerical simulation (DNS) subgrid models, and related frameworks. These length scales are vital for accurate subgrid-scale (SGS) modeling and fundamentally impact simulation fidelity, especially on anisotropic, unstructured, or wall-refined meshes. Recent years have witnessed the introduction of several new paradigms for defining subgrid length scales, addressing known failures of traditional approaches and enabling robust, grid-insensitive turbulent transport closures across diverse flow configurations.

1. Motivations and Deficiencies of Classical Approaches

Traditional subgrid characteristic lengths—most notably Deardorff's isotropic volumetric cube-root $$\Delta_{\text{vol}} = ( \Delta_x \Delta_y \Delta_z )^{1/3}$$—fail to maintain predictive accuracy on anisotropic or unstructured grids. As grid spacings become disparate, such as in "pancake" (highly refined in one direction) or "pencil" meshes, $\Delta_{\text{vol}}$ can become arbitrarily small, causing the SGS viscosity (e.g., $\nu_t \sim \Delta^2 |S|$ for Smagorinsky-type models) to collapse and SGS dissipation to vanish. This artifact results in underdamped turbulence, spectral pileup, and unphysical grid-dependence of statistics in both homogeneous and wall-bounded flows. Geometric corrections (e.g., $\Delta_{\max}$, $\Delta_{L2}$, and $\Delta_{\mathrm{lapl}}$) partially alleviate but do not fully suppress such grid artifacts (Trias et al., 27 Dec 2025, Trias et al., 2017).

A further limitation is that purely geometric scales are insensitive to local flow structure: they cannot distinguish, for example, vortical regions of high intensity or align the subgrid modeling directionally with coherent flow features. Emerging physical models require length scales reflecting both mesh and local flow anisotropy or physics.

2. Flow- and Physics-Informed Length Scale Formulations

Recent advances incorporate physical or mathematical reasoning into the SGS length scale:

• Gradient-model-based Scales (Δₗₛq): Introduced by Trias et al. (Trias et al., 2017), the $\Delta_{lsq}$ scale arises by optimally matching the leading-order gradient approximation of the SGS stress tensor for anisotropic grids,

$$\Delta_{lsq} = \sqrt{ \frac{ \mathrm{tr}( G \Delta^2 G^T ) }{ \mathrm{tr}( G G^T ) } }$$

where $G$ is the filtered velocity gradient and $\Delta$ is the local mesh spacing tensor. This ensures $\Delta_{lsq}$ is positive, bounded, frame-invariant, and sensitive to both mesh and flow orientation, reducing to geometric scaling in isotropic limits.

• Rational Length Scale (Δ_RLS): Trias et al. (Trias et al., 27 Dec 2025) propose that the implicit LES filter length in finite-volume schemes corresponds to the center-to-center distance $δ_f$ projected along the face normal,

$$\Delta_{RLS}|_f = | ( \mathbf{x}_N - \mathbf{x}_P ) \cdot \mathbf{n}_f |$$

for each face. This length is algorithmically consistent with the discretization and remains robust on arbitrary meshes. For codes with purely cell-centered SGS closures, a dissipation-matching length $\Delta_{PRLS}$, constructed from computational and physical gradient traces, efficiently substitutes $\Delta_{vol}$ without introducing mesh artifacts.

• Liutex-based Length Scale ($A_R$): For wall-bounded turbulence, Chen et al. (Chen et al., 16 Dec 2025) introduce a rotation-aligned subgrid scale based on the Liutex vector, defining

$$A_R = 2 \max_{m=1,\ldots,6} \| \mathbf{n}_R \times \mathbf{r}_{cf,m} \|$$

where $\mathbf{n}_R$ is the normalized local rotation axis and $\mathbf{r}_{cf,m}$ the vectors from cell center to face center. $A_R$ reflects the largest cross-axis cell radius, adapting to both grid geometry and local coherent rotation, thus producing a physically consistent diffusion directionality for vortical structures.

• Voinov Length $\ell_V$ in Contact-Line Hydrodynamics: In DNS of moving contact lines for boiling, the so-called Voinov length acts as an SGS cutoff bridging hydrodynamic and evaporative microphysics. Explicitly, for partial wetting with hydrodynamic slip length $l_s$ and Voinov angle $\theta_V$:

$\ell_V \simeq \frac{3}{e} \frac{l_s}{\theta_V}$

This length controls the inner boundary for Cox–Voinov-type apparent contact angle laws and enables subgrid closure without mesh refinement below $\sim 100$ nm, fully incorporating nanoscale physics only implicitly (Nikolayev et al., 2024).

3. Mathematical Properties and Computational Implementation

All these novel length scales share several mathematical and computational properties essential for practical deployment:

• Positivity, Locality, and Frame-invariance: Whether geometric (Δ_RLS), flow-aligned (Δₗₛq, $A_R$), or microphysics-based ($\ell_V$), all definitions remain strictly positive, are constructed from local (cell/cell-face) data, and are invariant under rigid-body transformations.
• Isotropic Consistency: Each scale recovers the correct mesh step or filter width for uniform cubic grids.
• Robustness to Mesh Anisotropy: Both gradient-aware and rational scales guarantee boundedness between minimum and maximum cell sizes, suppressing spurious grid dependencies.
• Computational Simplicity: For face-based closures (Δ_RLS), the implementation reduces to a single dot-product; for tensor-gradient versions ($\Delta_{lsq}$, $\Delta_{PRLS}$), the cost is a handful of inner-products per cell. Liutex-based models incur only modest overhead due to the triple decomposition of the velocity gradient, and Voinov-length models supply analytic closure formulas at subgrid scales.

The table summarizes key properties as described in (Trias et al., 27 Dec 2025, Trias et al., 2017, Chen et al., 16 Dec 2025, Nikolayev et al., 2024):

Scale	Grid Sensitivity	Flow Sensitivity	Mesh Type Support
Δ_vol (classic)	high	none	structured / Cartesian
Δ_RLS	low	none	any (FV/unstructured)
Δₗₛq	low	high	structured/unstructured
$A_R$	depends	high	structured
$\ell_V$	none	microphysics	not mesh-dependent

4. Physical Significance in Representative Applications

• Turbulent Channel and Isotropic Flows (LES): SGS closures with Δ_RLS and Δₗₛq eliminate artificial dissipation loss and maintain converged mean profiles and Reynolds stresses under strong grid stretching, outperforming classical approaches by fully restoring turbulence scaling in both "pancake" and "pencil" regimes (Trias et al., 27 Dec 2025, Trias et al., 2017).
• Wall-bounded Turbulence: The Liutex-based $A_R$ provides dynamic cross-axis adaptation, yielding improved resolved normal/shear stress profiles and natural near-wall damping without empirical wall corrections (Chen et al., 16 Dec 2025).
• DNS of Moving Contact Lines: The subgrid length $\ell_V$ in boiling/evaporating systems regularizes the contact-line singularity at the continuum level, enforcing realistic boundary conditions and microregion physics without explicit grid refinement (Nikolayev et al., 2024).
• Stable Planetary Boundary Layers: Physically motivated blending, as in $\lambda = [1/(\kappa z) + N/(c_n e^{1/2})]^{-1}$, corrects the spurious independence of eddy coefficients from stability for coarse grids, providing grid-invariant and stratification-aware models for atmospheric LES (Dai et al., 2020).

5. Comparative Performance and Practical Recommendations

Empirical validation across canonical LES and DNS test cases demonstrates that these novel length scales substantially mitigate grid-induced artifacts and improve agreement with direct numerical simulation or experimental benchmarks:

• For stretched and skewed meshes, Δ_RLS and Δₗₛq eliminate eddy-viscosity "pinching-off," guarantee monotonic convergence, and stabilize all lower-order and higher-order statistics even under extreme mesh distortion.
• Liutex-based models outperform grid-only and previous vorticity-based SGS closures especially in wall-near regions, with enhanced accuracy for Reynolds stress, channel velocity, and realistic switching-off at no-slip walls (Chen et al., 16 Dec 2025).
• The explicit analytic treatment of $\ell_V$ in boiling simulations enables DNS that accurately reproduce high-flux phenomena in the contact line vicinity without ad hoc parameterization or non-physical regularization (Nikolayev et al., 2024).
• Grid-size insensitivity for stable atmospheric boundary layers is achieved with harmonic-mean blending, recovering physically correct scaling from canopy through upper boundary layer and avoiding "quasi-laminarization" on coarse meshes (Dai et al., 2020).

6. Limitations and Open Challenges

Despite the robustness of these frameworks, certain limitations remain:

• Geometry-based length scales (Δ_RLS) do not capture flow anisotropy unless combined with flow gradients or invariants (e.g., QR, WALE, Liutex).
• Flow-dependent scales ($\Delta_{lsq}$, $A_R$) require accurate and stable computation of the local velocity gradient or rotation vectors, with possible numerical sensitivity near symmetric or poorly resolved regions.
• In highly inhomogeneous turbulence or flows with strong local forcing, further hybridization—potentially combining geometric, flow-aligned, and physical-microregion models—may be necessary to optimally balance computational cost and accuracy.
• For higher-order or spectral methods, the derivation of a discretization-aware implicit filter length must be revisited.
• Integration in hybrid RANS–LES methodologies, such as DDES/IDDES, is promising but requires systematic evaluation for interface and model transition artifacts (Trias et al., 27 Dec 2025).

7. Summary and Outlook

Novel subgrid characteristic lengths embody a fundamental shift in SGS closure, moving from purely geometric or ad hoc models towards mathematically principled, physically motivated, and computationally robust definitions. The recent rational, gradient-optimal, Liutex-based, and physically constrained microregion length scales each address central deficiencies in classical approaches and collectively offer a toolkit for robust, anisotropy-immune LES and DNS subgrid modeling across complex meshes and multiscale flows (Trias et al., 27 Dec 2025, Trias et al., 2017, Chen et al., 16 Dec 2025, Nikolayev et al., 2024, Dai et al., 2020). As numerical methods and computational hardware evolve, continued integration of these advances into production CFD frameworks is likely to be central to reliable and routine simulation of realistic turbulent flows.