Reynolds averaging and closure problems

Decompose incompressible constant-density velocity as $u_i=U_i+u_i'$ with $U_i=\langle u_i\rangle$ and zero mean fluctuation. Use an ensemble average, or another averaging operation for which the required linearity and derivative-commutation properties hold. The average product becomes $\langle u_i u_j\rangle=U_iU_j+\langle u_i'u_j'\rangle$. Zero individual fluctuation means do not make the product average vanish. A streamwise fluctuation correlated with cross-stream motion transports streamwise momentum even when the mean cross-stream velocity is zero.

Define $R_{ij}=\langle u_i'u_j'\rangle$. Averaging Navier–Stokes adds the divergence of this covariance tensor to the mean momentum balance. A common stress convention writes the turbulent contribution as $\tau^R_{ij}=-\rho R_{ij}$. State the sign convention because other authors discuss the positive covariance as the Reynolds stress itself. The diagonal covariances are nonnegative, and their trace gives twice the turbulent kinetic energy per unit mass. Off-diagonal components may have either sign.

The mean equation contains additional unknown covariances. Deriving an exact transport equation for them introduces triple velocity correlations, pressure–velocity effects, and dissipation information. Deriving equations for the new quantities produces still more unknowns. This hierarchy is the closure problem. It is not solved by adding enough names to the unknown variables. A closure introduces relations or modeled transport terms, justified and assessed for the intended regime using physical constraints, resolved data, experiments, and validation cases.

The Boussinesq eddy-viscosity ansatz relates the deviatoric covariance to mean strain using a modeled turbulent viscosity $\nu_t$. Its isotropic part contains turbulent kinetic energy k. Unlike molecular viscosity, $\nu_t$ is not a material constant fixed by the fluid alone. It depends on the flow and the closure. The ansatz imposes alignment and locality assumptions that can be poor for strong anisotropy, rotation, separation, or rapid changes. More elaborate Reynolds-stress closures relax some assumptions but introduce their own modeled terms.

The turbulent kinetic-energy production per unit mass is $\mathcal P=-R_{ij}\partial_jU_i$. In a simple shear $U_x=U(y)$ this reduces to $-\langle u'v'\rangle dU/dy$. A negative streamwise–wall-normal covariance in a positive mean gradient therefore yields positive production: energy passes from the mean flow into fluctuations. This term is a transfer between two budgets, not an extra source of total kinetic energy. Molecular dissipation ultimately converts kinetic energy into internal energy.

A local production estimate does not guarantee a local production–dissipation equilibrium. Advection, pressure transport, turbulent transport, and viscous diffusion can carry energy into or out of the region. A model calibrated on one nearly equilibrated flow can fail where these terms matter strongly. Check mean forces, profiles, stress components, and energy budgets separately when possible. A closure that predicts one drag coefficient well has not thereby recovered every turbulent statistic, and a visually plausible mean profile is not a universal solution to turbulence.