Reynolds-Averaged Navier-Stokes

RANS time-averages the Navier-Stokes equations and models all turbulent scales using a closure model, making it computationally cheap but less accurate for unsteady phenomena. LES resolves large eddies explicitly and models only small scales, requiring finer grids but capturing more turbulent structure. Choose RANS for steady engineering problems; choose LES for time-dependent flows or where eddy dynamics matter.

y+ is a dimensionless wall distance that measures the height of the first grid point above a wall in wall units. For wall-resolved RANS, y+ should be <1 to resolve the viscous sublayer. For wall-modeled RANS with wall functions, y+ should be 30-300. Using the wrong y+ range leads to poor boundary layer prediction and incorrect skin friction.

k-epsilon is robust for fully turbulent attached flows and industrial applications; k-omega excels in separated and adverse pressure gradient flows but is sensitive to freestream turbulence; Spalart-Allmaras is economical (one equation) and works well for aerodynamics applications. Start with k-omega if separation is expected, otherwise k-epsilon for general engineering.

Yes, RANS can predict separation and recirculation, but accuracy depends strongly on mesh quality and turbulence model choice. Separated flows are more challenging than attached flows; k-omega models generally perform better. Wall-resolved RANS with fine meshes near boundaries gives better separation predictions than wall-modeled approaches.

A 2D RANS case may use 50,000-500,000 grid points and solve in minutes to hours on a workstation. A 3D case typically requires 1-50 million points and several hours to days on multi-core computers or clusters. LES and DNS require orders of magnitude more cells, making RANS the choice when computational budget is limited.