Rayleigh-Bénard Convection: How Heat Organizes Itself

Conduction fights buoyancy

Take a thin layer of fluid trapped between two horizontal plates, heat the bottom plate and cool the top one. As long as the temperature difference is small, heat simply diffuses upward by conduction and the fluid stays perfectly still — any tiny parcel of fluid that starts to rise, buoyed by being warmer (and so lighter) than its surroundings, is damped back down by viscosity before it gets anywhere, and its excess heat leaks away by thermal diffusion before it can matter. Buoyancy wants to overturn the layer; viscosity and thermal diffusion want to keep it still. Which one wins is set by a single dimensionless number.

Ra = (g · α · ΔT · d³) / (ν · κ)

g  = gravitational acceleration
α  = thermal expansion coefficient of the fluid
ΔT = temperature difference, bottom minus top
d  = depth of the fluid layer
ν  = kinematic viscosity
κ  = thermal diffusivity

The critical Rayleigh number, 1708

Henri Bénard first photographed the striking honeycomb pattern in a heated fluid layer in 1900, but it was Lord Rayleigh who, in 1916, worked out the theory of when and why it should appear. Below a critical value of Ra, any small perturbation to the still, purely conductive state decays away. Above it, perturbations of the right wavelength grow instead, and the fluid organizes itself into steady convection.

rigid-rigid boundaries (both plates no-slip)   Ra_c ≈ 1707.76
rigid-free boundaries (one plate a free surface) Ra_c ≈ 1100.65
free-free boundaries (both surfaces free)        Ra_c ≈ 657.51

The exact number depends on the boundary conditions, but Ra_c ≈ 1708 for the standard case of two rigid, no-slip plates is one of the best-known results in hydrodynamic stability theory — a threshold you can reproduce in a kitchen pan of oil on a hot stove.

From rolls to hexagons to chaos

Just above onset, the linear theory predicts straight, parallel convection rolls with a wavelength close to twice the layer depth. Real experiments, though, very often show hexagonal cells instead — fluid rising in the center of each hexagon and sinking at its shared edges, or the reverse. Rolls are the preferred pattern only when the system is perfectly symmetric top-to-bottom; a free upper surface or fluid properties that change with temperature (most fluids get less viscous when hot) break that symmetry and favor hexagons. Push the driving further — Ra of order 10⁴ to 10⁵ — and the rolls start to wobble, then oscillate in time, and eventually the flow becomes fully turbulent. This orderly cascade from stillness to pattern to chaos made Rayleigh-Bénard convection one of the classic laboratory systems for studying the routes to turbulence, and it is where Edward Lorenz found his famous attractor: his 1963 three-variable model is a drastic truncation of the full convection equations down to just three Fourier modes.

The Nusselt number: how much better is convection?

The payoff for all this organization is heat transport. The Nusselt number Nu compares the actual rate of heat transfer to what conduction alone would achieve; at and below onset Nu = 1, and it climbs steadily as Ra increases and convection takes over the job. In the turbulent regime, decades of experiments and theory have argued over the exact scaling — a classical boundary-layer argument gives roughly Nu ∝ Ra^(1/3), while Robert Kraichnan predicted in 1962 that an ultimate regime at extremely high Ra should scale closer to Ra^(1/2) — and settling the question experimentally, in facilities that can reach astrophysically relevant Rayleigh numbers, remains an active area of fluid dynamics research.

Where it shows up

This is not just a tabletop curiosity. Cloud streets in the atmosphere are large-scale Bénard cells; the slow churning of the Earth's mantle that drives plate tectonics is buoyancy-driven convection on a geological timescale; the mottled, ever-changing texture of the Sun's visible surface, called granulation, is convection in the outer layers of a star; and industrial crystal-growth processes (like the Czochralski method used to make silicon wafers) have to be engineered around exactly this instability, because unwanted convection cells in the melt distort the crystal that grows from it.

Frequently asked questions

Try it live

Everything above runs in your browser — open Rayleigh-Bénard Convection and change the parameters while it is running. Nothing is installed, nothing is uploaded, the whole model lives in one tab.