Frequently Asked Question

Most engineering flows are turbulent, and since directly resolving every turbulent eddy (DNS) is computationally prohibitive for practical steady-state work, turbulence models approximate turbulence's effect on the mean flow. Choosing the right one matters for both accuracy and convergence behavior.

Commonly used models for steady-state (RANS-based) simulations:

• k-ε (k-epsilon) — Robust, well-validated, computationally efficient. Performs well for free-shear flows and fully turbulent internal flows away from walls. Less accurate for flows with strong adverse pressure gradients or separation, and requires wall functions (not ideal for very fine near-wall resolution).
• k-ω (k-omega), particularly k-ω SST (Shear Stress Transport) — Generally considered more accurate near walls and for flows with separation or adverse pressure gradients. SST blends k-ω behavior near walls with k-ε behavior in the free stream, combining strengths of both. This is often the default recommendation for general-purpose external aerodynamics and internal flows with possible separation.
• Spalart-Allmaras — A single-equation model, computationally cheap, historically popular for aerospace applications (attached or mildly separated boundary layer flows). Less robust for complex separated or highly swirling flows.
• RSM (Reynolds Stress Model) — More physically complete (solves transport equations for individual Reynolds stress components rather than a simplified turbulent viscosity), better for strongly swirling or anisotropic flows, but significantly more computationally expensive and can be harder to converge.

General guidance:

• k-ω SST is a solid default for most general-purpose steady-state industrial CFD — good balance of accuracy and robustness
• Use k-ε for simpler internal flows where near-wall accuracy is less critical and robustness/speed matters
• Reserve RSM for cases with strong swirl or anisotropic turbulence where simpler models are known to underperform
• Whatever model you choose, match your mesh's near-wall resolution to that model's requirements — a mismatch (e.g., a fine near-wall mesh built for a low-Re model but paired with wall functions meant for coarser meshes) undermines the model's assumptions

Closure and scale awareness

Turbulent flow contains fluctuating velocity and pressure over a range of interacting scales. Averaging introduces additional correlations, such as turbulent momentum transport, that are not determined by the mean variables alone. A closure model is therefore an assumption about how unresolved transport relates to resolved gradients or other modeled quantities. Its suitability depends on Reynolds number, wall treatment, separation, curvature, buoyancy, compressibility, and the engineering output.

U is mean velocity and u′ is a fluctuation. The turbulent stress is not a universal constant; it depends on the flow and the adopted closure.

Worked example

For a duct with D_h = 0.10 m, U = 20 m/s, ρ = 1.2 kg/m³, and μ = 1.8×10⁻⁵ Pa·s, Re = ρUD_h/μ = 133,000. A laminar assumption then needs strong physical justification. Compare pressure loss, wall behavior, and sensitivity to near-wall resolution before accepting a model.

Limit: residual reduction alone cannot establish turbulence-model adequacy.