The Energy Cascade & Kolmogorov Scaling
Intuition
Stir a cup of coffee once. You created one big swirl — and stopped adding energy. Yet moments later the coffee is full of small eddies, and moments after that it is still, slightly warmer. The energy cascaded: big whirls broke into little whirls (Richardson's rhyme), passing energy down scale after scale until eddies became so small that viscosity could finally grind them into heat.
The three ranges
- Energy injection at the integral scale \(L\) — the size of the biggest eddies, set by geometry/forcing (obstacle size, stirring). Most kinetic energy lives here.
- Inertial range (\(L \gg \ell \gg \eta\)) — energy flows down-scale at a constant rate \(\varepsilon\) (per unit mass), with negligible dissipation: pure nonlinear hand-me-down.
- Dissipation at the Kolmogorov microscale \(\eta\), where the local Reynolds number reaches order 1 and viscosity converts the flux to heat.
Kolmogorov microscales (1941)
Assume the smallest scales know only \(\nu\) [m²/s] and \(\varepsilon\) [m²/s³]. Dimensional analysis then fixes everything:
Sanity check: \(Re_\eta = u_\eta\,\eta/\nu = 1\) — the definition of "where viscosity takes over" is built in.
The −5/3 spectrum
In the inertial range, the energy spectrum \(E(k)\) (energy per unit wavenumber, \([E] = \text{m}^3/\text{s}^2\)) can depend only on \(\varepsilon\) and \(k\):
This "K41" law is among the most-verified results in fluid mechanics — tidal channels, wind tunnels, jets and atmospheric data collapse onto the same \(-5/3\) slope.
K41's assumptions: at high \(Re\), small scales are (1) statistically isotropic, (2) universal — dependent only on \(\varepsilon\) and \(\nu\), and (3) in the inertial range, independent even of \(\nu\) (self-similar energy transfer).
Scale separation in practice
An atmospheric flow with \(L \sim 1\) km and \(Re \sim 10^8\) has \(\eta\) under a millimetre — five decades of active scales. That \(Re^{3/4}\) (per direction; \(Re^{9/4}\) in 3-D) is exactly why direct simulation of high-\(Re\) turbulence is infeasible and modeling is unavoidable.
Common mistakes
- Thinking dissipation happens at large scales. Big eddies are essentially inviscid; only at \(\eta\) does friction act. (Corollary: the dissipation rate is set by the large scales — \(\varepsilon \sim U^3/L\) — since they control the supply.)
- Applying \(-5/3\) outside the inertial range or in 2-D flows (which cascade backwards in energy — a beautiful, separate story).
- Treating \(\varepsilon\) as adjustable. In steady state it is fixed by injection: what goes in must come out.
Related concepts
- Turbulence — the phenomenon
- Dimensional analysis — the method that yields everything here
- Viscosity — the terminal sink
- Reynolds averaging — statistical machinery
Knowledge graph position
Prerequisites: Turbulence, Dimensional analysis. Leads to: turbulence modeling, spectral analysis of flows.
Quiz
Q1 (computational). In a wind tunnel, \(\varepsilon \approx 1\ \text{m}^2/\text{s}^3\) and \(\nu = 1.5\times10^{-5}\ \text{m}^2/\text{s}\). Find \(\eta\) and \(\tau_\eta\).
Answer
\(\eta = (\nu^3/\varepsilon)^{1/4} = (3.4\times10^{-15})^{1/4} \approx 0.24\) mm; \(\tau_\eta = \sqrt{\nu/\varepsilon} \approx 3.9\) ms.
Q2 (conceptual). Why does the inertial-range spectrum not depend on viscosity?
Answer
In that range eddies pass energy on much faster than viscosity can act (\(\ell \gg \eta\)), so \(\nu\) is dynamically irrelevant; only the throughput \(\varepsilon\) and the scale \(k\) remain — and dimensions then force \(-5/3\).
Q3 (multiple choice). If you double the stirring power in a steady turbulent tank (doubling \(\varepsilon\)), the Kolmogorov length:
- (a) doubles (b) halves (c) shrinks by \(2^{1/4}\) (d) is unchanged
Answer
(c). \(\eta \propto \varepsilon^{-1/4}\): more power pushes dissipation to slightly smaller scales.