Rayleigh-Bénard Convection

The intuition — why a heated fluid starts to churn

Put a shallow layer of fluid — oil, water, air — between a hot plate below and a cool plate above. At first nothing visibly moves. Heat crosses the layer by conduction: molecules jiggle and pass energy up through a fluid that stays still. This is the boring, stable state.

But the bottom fluid is warmer, so it has expanded and become less dense than the cool fluid sitting on top of it. That is an upside-down, top-heavy arrangement — dense fluid perched above light fluid — and gravity hates it. A blob of warm fluid that drifts up even slightly is now lighter than its surroundings, so buoyancy pushes it up further. A cool blob that drifts down is heavier than its surroundings, so it sinks further. The arrangement wants to overturn.

Two things fight back. Viscosity resists the flow — it costs energy to shear the fluid as a blob moves through it. And thermal diffusion bleeds the temperature difference away: a warm rising blob loses its heat to the cooler fluid around it, and once it matches its surroundings it has no buoyancy left. If the heating is gentle, these two dampers win and the fluid just conducts. Crank the heat up past a sharp threshold and buoyancy wins — the layer breaks into a regular array of convection cells, with warm fluid streaming up in some columns and cool fluid sinking in others. That ordered overturning is Rayleigh-Bénard convection.

The Rayleigh number — the single dial that controls everything

Whether the fluid conducts or convects is decided by one dimensionless number, the Rayleigh number:

Ra = (g · α · ΔT · d³) / (ν · κ)

where:

• g — gravitational acceleration (9.81 m/s²)
• α — thermal expansion coefficient of the fluid (1/K)
• ΔT — temperature difference between bottom and top plate (K)
• d — depth of the fluid layer (m)
• ν — kinematic viscosity (m²/s)
• κ — thermal diffusivity (m²/s)

Read it as a tug-of-war. The numerator gαΔT·d³ is the buoyant driving force — bigger temperature difference, deeper layer, or more expandable fluid all push toward overturning. The denominator νκ is the combined damping — viscosity slows the motion and thermal diffusion erases the buoyancy. When the ratio crosses a critical value, convection switches on.

The cube on the depth is the killer. Doubling the layer depth multiplies Ra by 8. This is exactly why a thick layer of the same fluid convects vigorously while a paint-thin film just conducts — and why the deep convecting layers of the Sun and Earth's mantle have astronomically large Rayleigh numbers.

The governing physics — Boussinesq equations

Rayleigh-Bénard convection is usually modeled with the Boussinesq approximation: density is treated as constant everywhere except in the buoyancy term, where it varies linearly with temperature. The fluid obeys the incompressible Navier-Stokes equations coupled to a heat equation:

∇ · u = 0                                    (incompressibility)

∂u/∂t + (u · ∇)u = −(1/ρ₀)∇p + ν∇²u + gα(T − T₀)ẑ   (momentum + buoyancy)

∂T/∂t + (u · ∇)T = κ∇²T                       (heat transport)

The buoyancy term gα(T − T₀)ẑ is the engine — it is the only place density variation enters, and it points up. The density itself follows ρ = ρ₀[1 − α(T − T₀)]: warm fluid (T > T₀) is lighter and feels an upward push.

Two dimensionless numbers fall out when you non-dimensionalize these equations. One is the Rayleigh number above. The other is the Prandtl number:

Pr = ν / κ   (momentum diffusivity / thermal diffusivity)

Pr decides which damping dominates. For air Pr ≈ 0.7, for water ≈ 7, for silicone oils 50–1000, and for the Earth's mantle it is effectively infinite (≈10²³). The onset threshold Ra_c does not depend on Pr, but the shape of the flow above onset and the route to turbulence very much do.

Onset — the critical Rayleigh number and cell size

Lord Rayleigh solved the linear stability problem in 1916: he perturbed the conducting state with a small wavy disturbance and asked when it grows instead of decays. The answer depends on the boundaries, and the threshold is remarkably precise.

Boundary condition	Critical Ra_c	Critical wavenumber k_c·d	Cell width
Two free (stress-free) surfaces	27π⁴/4 ≈ 657.5	π/√2 ≈ 2.221	≈ 2.83 d
One rigid, one free	≈ 1100.65	≈ 2.682	≈ 2.34 d
Two rigid (no-slip) plates	≈ 1707.76	≈ 3.117	≈ 2.02 d

The rigid-rigid value 1708 is the one usually quoted because most lab experiments use solid top and bottom plates. The critical wavenumber tells you the pattern's spacing: k_c ≈ 3.117/d means the wavelength λ = 2π/k_c ≈ 2.016 d, so one up-down roll pair spans about twice the layer depth. A deeper layer makes proportionally wider cells.

Just above Ra_c the growth is gentle — a "supercritical" (forward) bifurcation. The flow amplitude grows like √(Ra − Ra_c), so the rolls appear smoothly rather than jumping into existence. This is the textbook example of pattern formation through a symmetry-breaking instability.

Regimes — from rolls to turbulence as you turn up the heat

The Rayleigh number doesn't just decide on/off — as you push it higher, the flow climbs a ladder of increasingly complex states:

Rayleigh number	State	What you see
Ra < 1708	Conduction	No motion; heat crosses by molecular diffusion only
Ra ≈ 1708 – 10⁴	Steady rolls / cells	Stationary, regular convection rolls or hexagons
Ra ≈ 10⁴ – 10⁵	Oscillatory / wavy rolls	Rolls develop time-dependent wobbles (oscillatory instability)
Ra ≈ 10⁵ – 10⁷	Weak turbulence (soft)	Chaotic plumes; thermal boundary layers form at the plates
Ra > 10⁸	Hard turbulence	Large-scale circulation ("wind") plus violent intermittent plumes

The headline result of the turbulent regime is the Nusselt number — the ratio of total heat transport to pure-conduction transport. Empirically Nu scales roughly as Nu ∝ Ra^(1/3) over a wide range, meaning convection at high Ra moves heat far more effectively than conduction. At Ra ≈ 10⁹ a convecting layer can carry tens of times more heat than the same layer would by conduction alone.

Real-world Rayleigh numbers — kitchen to cosmos

Because Ra spans the conduction/convection boundary at ~1708, you can predict whether any given layer will convect just by plugging in numbers.

System	Depth d	Approx. Rayleigh number	State
1 mm oil film on a warm plate	~1 mm	~10²–10³	Often just below onset — conducts or barely convects
Cup of hot coffee / miso soup	~5 cm	~10⁶	Visible surface cells
Room air heated by a radiator	~2 m	~10⁹	Turbulent convection plumes
Earth's atmosphere (boundary layer)	~1 km	~10¹⁷	Turbulent; forms cloud cells
Earth's mantle	~2,900 km	~10⁷–10⁸	Slow turbulent convection (drives plate tectonics)
Sun's convective zone	~200,000 km	~10²⁰–10²⁴	Vigorous turbulence; granulation

The mantle is the surprise: despite its enormous depth, its colossal viscosity (≈10²¹ Pa·s) keeps Ra "only" in the 10⁷–10⁸ range — high enough to convect, but with overturning timescales of ~100 million years. That slow churn is what carries continents.

Where it shows up

• Solar physics. The Sun's surface "granulation" — bright cells ~1,000 km across, each lasting ~8 minutes — is the top of giant convection cells in the solar convective zone. Larger "supergranules" span ~30,000 km.
• Plate tectonics. Mantle convection cells drag the lithosphere, opening ocean ridges where fluid upwells and driving subduction where it sinks. Convection is the engine of continental drift.
• Weather and climate. Atmospheric convection forms thunderstorms, the polygonal "open" and "closed" cloud cells over oceans, and the Hadley/Ferrel/polar circulation cells. Ocean convection at the poles drives the global thermohaline conveyor.
• Engineering and manufacturing. Convection limits how thin you can pour paint or coatings before Bénard cells leave a visible "orange-peel" texture; it governs crystal growth, casting, and the cooling of electronics and nuclear reactors.
• Chaos theory. Edward Lorenz's 1963 model — the origin of the butterfly effect and the strange attractor — is a three-mode truncation of a 2D Rayleigh-Bénard layer.
• Everyday kitchen. The cellular skin on hot soup, the rolling currents in a pan of heating oil, and the patterns in a mug of coffee with a little cream are all Rayleigh-Bénard (often with a Marangoni assist at the free surface).

Common misconceptions and edge cases

• "Bénard's original hexagons prove buoyancy makes hexagons." Not quite. Henri Bénard's celebrated 1900 hexagons in thin spermaceti films were mostly driven by surface tension (the Marangoni effect), not buoyancy. Pure buoyant Rayleigh-Bénard convection near onset prefers parallel rolls; it makes hexagons only when up-down symmetry is broken.
• "Heating from the side gives the same thing." No. A side temperature gradient drives convection at any Ra, with no threshold — there's no top-heavy unstable layer to overcome. The sharp onset is unique to heating from below (against gravity).
• "Bigger temperature difference always means a bigger Rayleigh number proportionally." ΔT enters linearly, but depth enters as d³ — geometry usually dominates. Also, the fluid's α, ν, and κ change with temperature, so very large ΔT can shift the threshold itself.
• "Convection cells are turbulent." Just above onset they are perfectly steady and laminar — clean, stationary rolls. Turbulence only appears thousands of times above critical.
• "Heating from above can convect too." No — heating from above puts dense cool fluid on the bottom, a stable arrangement. It's gravitationally locked and conducts only. (This is why lakes stratify in summer.)
• "It's the same as Rayleigh-Taylor instability." Related but distinct. Rayleigh-Taylor is a heavy fluid sitting on a light one (a density inversion held by an interface). Rayleigh-Bénard has a continuous temperature-driven density gradient and a competing thermal diffusion that sets a threshold.