Turbulence Explained — From Reynolds to Kolmogorov
Nobel laureate Richard Feynman called turbulence "the most important unsolved problem of classical physics." This article builds up turbulence from the Reynolds experiment (1883), through the Kolmogorov energy cascade, to the engineering turbulence models that let us simulate it without resolving every eddy.
1. What is Turbulence?
Turbulence is not random noise. It is a deterministic, chaotic state of fluid motion characterised by vortices at many scales that interact non-linearly. Three properties define it:
- Irregular: Velocity fluctuates in time and space unpredictably, even though the Navier-Stokes equations governing it are deterministic.
- Multi-scale: Eddies (vortical structures) span a continuous range of sizes, from the large energy-containing eddies (pipe diameter, cloud size) down to the tiny dissipative Kolmogorov scale (sub-millimetre).
- Dissipative: Turbulence always decays unless maintained by an energy source. Small eddies convert kinetic energy to heat by viscosity.
The same equations that describe a calm river also govern the raging rapids downstream. The difference is purely in the Reynolds number.
2. Reynolds Number and Transition
Osborne Reynolds (1883) injected dye into pipe flow and found a sharp transition. Below a critical Reynolds number, dye flowed in smooth parallel streaks (laminar). Above it, dye mixed throughout the pipe instantly (turbulent):
The physical interpretation: Re is the ratio of inertial forces to viscous forces. At low Re, viscosity damps any perturbation before it can grow. At high Re, inertia overcomes viscous damping and small disturbances amplify into turbulent fluctuations via the Kelvin-Helmholtz and other instabilities.
The transition is not instantaneous — it proceeds through stages: Tollmien-Schlichting waves → three-dimensional instabilities → turbulent spots → fully turbulent flow. In a favourable pressure gradient (accelerating flow over an aircraft wing near the leading edge), transition is delayed; adverse gradients accelerate it.
3. The Kolmogorov Energy Cascade
Andrey Kolmogorov (1941) formulated the modern understanding of turbulence structure. Energy enters the turbulent flow at the integral scale L (the largest eddies, of order the boundary/geometry size). These large eddies are unstable, break into smaller eddies, which break into smaller still, passing energy continuously downward in scale — the cascade.
The cascade continues until eddies become so small that viscosity dissipates them into heat. This occurs at the Kolmogorov microscale — the smallest scale of turbulent motion.
4. Kolmogorov Microscales
From the two parameters that govern the small scales — kinematic viscosity ν and dissipation rate ε — Kolmogorov derived the smallest turbulent length, time, and velocity scales:
This is why direct numerical simulation (DNS) — resolving all scales — is limited to Re ≲ 10⁴. Engineering flows (aircraft wings at Re ≈ 10⁷, ship hulls at Re ≈ 10⁹) require turbulence models to avoid explicitly resolving the astronomical number of cells needed.
5. Kármán Vortex Street
When flow passes a blunt body (cylinder, bridge pylon), vortices detach alternately from each side, forming a staggered double row called a Kármán vortex street. The shedding is periodic with a frequency given by the Strouhal number:
Kármán vortex streets cause structural resonance. The Tacoma Narrows Bridge (1940) collapsed when vortex shedding matched the structure's natural frequency — now called flutter. Modern bridges use aerodynamically shaped cross-sections and fairings to detune the vortex-body frequency coupling.
At low Re (40–190), the wake is laminar but oscillating (2D Kármán street). At higher Re (>190), the vortices break down into 3D turbulent structures in the wake.
6. Engineering Turbulence Models
RANS turbulence models replace the fluctuating velocity field with a turbulent viscosity ν_t that augments the molecular viscosity:
k-ε Model (Launder & Spalding 1974)
k-ε strengths: Robust, well-validated for attached flows and pipe flows. The default model in many industrial CFD packages (Fluent, OpenFOAM). Weakness: Inaccurate for separated flows (recirculation behind bodies), adverse pressure gradients, and flows with significant streamline curvature.
k-ω SST Model (Menter 1994)
Uses k-ω (better near-wall behaviour) in the boundary layer and blends to k-ε in the free stream. The Shear Stress Transport (SST) modification also limits ν_t in adverse pressure gradients, improving separated flow predictions. Most widely used "workhorse" model in aerospace CFD today.
Spalart-Allmaras (1992)
A single transport equation for ν̃ (modified turbulent viscosity). Very fast (one extra equation vs two for k-ω). Developed specifically for aerodynamic flows; less accurate for internal flows and recirculation.
7. DNS, LES, and RANS — A Practical View
| Method | Resolves | Mesh Cells (3D, Re=10⁶) | Time | Who Uses It |
|---|---|---|---|---|
| DNS | All scales | ~10¹⁴ (impossible) | Years of CPU | Fundamental turbulence research, Re < 10⁴ |
| LES | Large eddies; small modelled | ~10⁸–10¹⁰ | Days–weeks | Aeroacoustics, combustion, weather, complex flows |
| RANS (k-ε) | Mean flow only | ~10⁵–10⁷ | Hours | Industry (cars, planes, HVAC, turbines) |
| Wall-modelled LES | Large eddies; wall modelled | ~10⁷–10⁹ | Days | High-Re LES in aerodynamics |
8. The Unsolved Problem
Despite more than a century of research, turbulence remains one of the Clay Mathematics Institute's Millennium Prize Problems ($1 million prize): prove or disprove that smooth solutions to the 3D Navier-Stokes equations always exist (no finite-time blow-up).
Practically, we can simulate turbulence accurately using DNS at low Re and modelled approaches at higher Re. But we cannot predict turbulence from first principles without running the simulation. The Navier-Stokes equations are Lyapunov unstable at high Re — tiny initial perturbations grow exponentially (chaos). This is why weather is predictable only for ~10 days, and why each aircraft's wake is unique.