Hybrid RANS–LES Turbulence Modeling

• HRLES is a hybrid turbulence modeling method that combines RANS near walls with LES in free-shear regions to deliver accurate, high-Reynolds number flow predictions.
• It leverages multiscale decomposition and function enrichment techniques via high-order discontinuous Galerkin methods to resolve both statistically modeled and explicitly simulated turbulence.
• HRLES achieves significant computational and grid savings while mitigating log-layer mismatch, as demonstrated in channel and periodic hill flow studies.

A hybrid Reynolds-averaged Navier–Stokes/large eddy simulation (HRLES) approach is a turbulence modeling methodology designed to combine the strengths of RANS (statistical, time-averaged modeling effective near walls or in attached boundary layers) and LES (explicitly resolved turbulence at larger scales, effective away from walls and in separated regions). The HRLES paradigm achieves this by spatially or functionally partitioning the domain or solution space, applying RANS modeling where the resolution is insufficient for turbulence resolution, and allowing LES where grid and flow allow direct eddy-resolving simulations. The principal aim is to address challenges in predictive high-Reynolds number turbulence—particularly wall-bounded or separated flows—by balancing modeling accuracy, computational efficiency, and numerical robustness (Krank et al., 2017).

1. Theoretical Foundations and Multiscale Formulation

HRLES strategies rest on multiscale decomposition of the turbulent velocity field and filtering concepts. In one prominent approach, the solution in the near-wall region is split as $u_H = \bar{u} + \tilde{u}$, where $\bar{u}$ is the LES-resolved component and $\tilde{u}$ is the RANS component. The hybrid equations are derived by applying an additive three-level filter (coefficient triplet $a_1, a_2, a_3$, usually $1,1,-1$) to the Navier–Stokes equations (Krank et al., 2017). The resulting system, in the near-wall subdomain $\widetilde{\Omega}$, is: $$\begin{aligned} &\nabla \cdot (\bar{u} + \tilde{u}) = 0, \ &\frac{\partial (\bar{u} + \tilde{u})}{\partial t} + \nabla \cdot \big(u_H \otimes u_H + p_H I - 2 \nu \varepsilon(\bar{u} + \tilde{u}) - 2\nu_t \varepsilon(\tilde{u})\big) = f, \end{aligned}$$ where $\nu_t$ is the turbulent eddy viscosity applied only to the RANS component, and $$\varepsilon(u) = \tfrac{1}{2}(\nabla u + \nabla u^T)$$ (Krank et al., 2017).

In the generic "interface-based" HRLES framework (e.g., IDDES, DDES), a single transport system with hybrid eddy viscosity is solved, where RANS and LES effective length-scales are blended via switching or shielding functions to transition the model behavior (Arolla, 2014). The canonical variation is: $$\nu_T = f_\text{RANS} \, \nu_T^\text{RANS} + (1-f_\text{RANS})\, \nu_T^\text{LES},$$ with $f_\text{RANS}$ determined by local flow diagnostics, wall distance, and/or grid resolution.

2. Discretization and Function Enrichment Techniques

The HRLES approach as implemented by Krank, Kronbichler, and Wall adopts a high-order discontinuous Galerkin (DG) finite element method enriched by specialized function subspaces (Krank et al., 2017). The trial and test spaces for velocity are decomposed into two independent spaces: $V_h^u = V_h^{\bar{u}} \oplus V_h^{\tilde{u}}.$ Here, $V_h^{\bar{u}}$ consists of piecewise polynomials of degree $k$, targeting LES-resolved dynamics, and $V_h^{\tilde{u}}$ uses a van Driest wall-law function $\psi(y^+)$ multiplied by low-degree polynomials to efficiently represent steep gradients and mean shear in the wall layer (typically only in the innermost cell layer). This specialized enrichment permits accurate near-wall modeling with a coarse mesh, obviates the need for ad-hoc log-law corrections, and supports a variational multiscale (VMS) Galerkin formulation (Krank et al., 2017).

Other HRLES formulations (e.g., hybrid filter methods) may not employ explicit function enrichment but instead manipulate the model structure or filter definitions to balance resolved and modeled scales within classical finite volume or spectral-element frameworks (Abbà et al., 2014).

3. Blending, Transition, and Shielding Mechanisms

A central challenge in HRLES is the seamless transition between RANS and LES, specifically to avoid "log-layer mismatch" and "modeled-stress depletion" at the RANS–LES interface. The enriched DG strategy achieves this naturally because the eddy viscosity operates exclusively on the RANS component in the enriched subspace, permitting under-resolved LES dynamics even adjacent to the wall. As the grid is refined, the enrichment is deactivated (i.e., $\tilde{u} \rightarrow 0$), smoothly recovering wall-resolved LES with exact boundary conditions (Krank et al., 2017).

Alternative HRLES frameworks (e.g., IDDES) utilize a local hybrid length scale: $$\ell_\text{IDDES} = \ell_\text{RANS} - f_d \max(0, \ell_\text{RANS} - C_\text{DES} \Delta),$$ where $f_d$ is a nonlinear function of the ratio of modeled eddy viscosity to wall-normal distance and local strain. $f_d \rightarrow 0$ in the wall layer (RANS), and $f_d \rightarrow 1$ in detached or free shear regions (LES) (Arolla, 2014). These blending mechanisms enable the HRLES model to respond dynamically to local flow resolution and grid anisotropy without ad-hoc zone specification or artificial forcing terms.

4. Implementation Details and Solver Strategies

Key implementation features of state-of-the-art HRLES methods include:

• Polynomial orders: High-order DG methods typically use degree $k \geq 4$ for the LES subspace, enrichment is constant (degree $\ell = 0$) unless further wall accuracy is needed.
• Time stepping: Semi-explicit dual-splitting or second-order schemes (e.g., Karniadakis) are used, with the CFL condition set by the smallest cell and highest polynomial degree.
• DG operators: Symmetric interior-penalty for viscous terms, Lax–Friedrichs upwinding for convection, careful harmonic averaging of viscosity on facets, and div–div stabilization for mass conservation (Krank et al., 2017).
• Quadrature: Enriched cells require high-point Gauss integration (up to $12^3$ points) due to nonpolynomial basis.
• Boundary conditions: No-slip is enforced weakly via the mirror principle for velocity, and high-order Neumann for pressure in the Poisson step.
• Stability: Coercivity-driven clipping of the turbulent viscosity sum ensures energy stability; the enriched wall-layer height is limited to $y^+ \lesssim 120$ (Krank et al., 2017).
• Blending parameters: In non-enriched IDDES-like models, shielding functions and grid-based length scales are key to interface robustness and reducing sensitivity to grid irregularity (Arolla, 2014).

5. Performance in Canonical and Complex Flows

Multiple validation studies have demonstrated the strengths of HRLES approaches:

• Channel flow (Reτ = 395–5200): Enriched HRLES achieves mean velocity, RMS fluctuations, and Reynolds-shear stress profiles in close agreement with DNS, without log-layer mismatch, and with negligible sensitivity to mesh stretching for aspect ratios up to 1.6. A speedup of up to 160× was observed over wall-resolved LES (12 × 8 × 12 mesh vs. 24^3 or 32^3) (Krank et al., 2017).
• Flow over periodic hills (Re_H = 10,595–37,000): HRLES correctly predicts separation and reattachment locations, skin-friction, and pressure distributions in agreement with DNS and experimental data. On underresolved meshes, HRLES recovers separation and large-scale vortical structures missed by no-model or equilibrium wall-model approaches. Grid refinement further improves agreement and reveals HRLES's grid-independence (Krank et al., 2017).
• Comparison with DDES/IDDES: Relative to DDES, the multiscale HRLES approach resolves near-wall turbulence without introducing log-layer mismatch or excessive pressure drag, and converges more rapidly with mesh refinement (Krank et al., 2017).
• Speed and grid savings: HRLES demonstrates computational savings of $O(10^2)$ times over wall-modeled LES at engineering Reynolds numbers, due to coarser mesh requirements and larger stable time-steps.

6. Strengths, Limitations, and Extensions

Strengths:

• Fully consistent coupling of RANS and LES in a unified variational (VMS) framework with no ad-hoc blending or zone specification.
• Transition from wall-modeled RANS to wall-resolved LES occurs automatically and continuously with mesh refinement.
• Complete elimination of the log-layer mismatch that afflicts additive-filter and length-scale switching approaches.
• Significant speed-up and grid independence, without loss of fidelity in attached, separated, or massive unsteady turbulent flows.
• Ease of integration into high-order DG solvers via function enrichment (Krank et al., 2017).

Limitations:

• Stability constraints require limiting the height of the wall-enriched layer to $y^+ \lesssim 120$.
• Computational cost per element in the enriched layer is elevated due to quadrature, though total cost is mitigated by allowing coarser meshes.
• The formulation as presented uses standard algebraic eddy viscosity for the RANS component; more advanced closures could improve performance in massively separated or three-dimensional flows (Krank et al., 2017).

Potential Extensions:

• Direct inclusion of two-equation or Reynolds-stress transport models within the RANS subspace.
• Extension to compressible flows by enrichment of the energy equation.
• Adaptive determination of the enrichment region, based on local wall distance ($y^+$) or turbulence intensity.
• Coupling to dynamic subgrid-scale models in the LES subspace for improved anisotropy modeling.
• Application to real-world industrial or exascale DG environments (Krank et al., 2017).

7. Comparative Summary of HRLES Approaches

Approach	Blending/Transition Mechanism	Discretization	Log-Layer Strategy	Robustness/Advantages
Enriched DG (Krank et al.) (Krank et al., 2017)	Multiscale variational; function enrichment	High-order DG	Exact (via subspace)	Full RANS–LES coupling; automatic mesh-driven transition; grid independence
IDDES (Shur et al., Employed in (Arolla, 2014))	Length-scale blending via shielding f_d	FV, mostly low-order	Switch + Shielding	Widely used; simple in FV solvers; needs careful grid and tuning
Hybrid Filter (Germano, Abbà et al. (Abbà et al., 2014))	Additive operator, constant blend	High-order DG	Not explicit	No explicit RANS solve; low overhead on DG truncation

The HRLES field encompasses a range of compatible frameworks, but unified multiscale formulations (as in (Krank et al., 2017)) provide a rigorous path to grid-independent, high-fidelity turbulence prediction across diverse boundary-layer regimes.