Turbulent energy cascade and Kolmogorov scaling

Let $\varepsilon$ be mean kinetic-energy dissipation per unit mass, with dimensions length squared divided by time cubed. In statistically steady turbulence, mean injection balances mean dissipation after accounting for boundary transport. An inertial interval is a set of scales where direct forcing and molecular dissipation are both small in the scale-by-scale budget, so the mean transfer rate is approximately constant. A spectrum alone shows energy distribution; establishing a cascade also requires information about transfer direction and the budget.

The forward energy cascade is a characteristic of many three-dimensional turbulent regimes, not a theorem that every instantaneous transfer must be downscale. Backscatter and spatially intermittent transfer can occur while the mean flux remains forward. Rotation, stratification, magnetic fields, walls, and restricted dimensionality may alter the dynamics and statistical assumptions. A picture with several small swirls cannot by itself establish an inertial range. Scale separation and a resolved statistical measurement are needed before fitting a universal-looking slope.

For homogeneous isotropic turbulence, define the one-dimensional shell-integrated spectrum E(k) so that its integral over positive wavenumber equals mean kinetic energy per unit mass. It therefore has dimensions length cubed divided by time squared. If the inertial-range spectrum depends only on $\varepsilon$ and k, dimensional consistency gives the familiar minus-five-thirds exponent. A dimensionless coefficient remains undetermined by this argument. It depends on the precise spectrum convention and is not obtained by dimensional analysis.

This reasoning is a statistical similarity hypothesis. It does not make every instantaneous velocity increment equal to its estimated characteristic value, nor does it prove that all higher-order statistics share simple self-similarity. The exact longitudinal third-order law, under its homogeneous, isotropic, incompressible high-Reynolds-number inertial-range conditions, provides a stronger flux-related result than the dimensional spectrum argument. Keep these different logical statuses visible: an exact limiting relation and a phenomenological dimensional estimate are not interchangeable evidence.

At sufficiently small scale, the local Reynolds number $\delta u_\ell\ell/\nu$ becomes order one. Combining this criterion with the inertial estimate gives the Kolmogorov length $\eta=(\nu^3/\varepsilon)^{1/4}$, time $\tau_\eta=(\nu/\varepsilon)^{1/2}$, and velocity $u_\eta=(\nu\varepsilon)^{1/4}$. These are characteristic scales constructed from mean dissipation. Strong local fluctuations can require finer resolution for particular observables, especially high derivatives or intense events. Resolving an averaged length estimate is not an automatic accuracy certificate.

A direct numerical simulation must resolve the energy-containing region as well as the dissipative scales, and must run long enough for its desired statistics. Increasing Reynolds number widens the separation and rapidly raises spatial and temporal costs. A large-eddy simulation instead resolves a range of larger scales and models the influence of smaller ones. Neither a fitted spectral line nor a visually complicated animation settles whether a numerical calculation has the correct energy budget, adequate resolution, or statistically converged rare-event measurements.