Brownian Motion

The dance no one could explain

In 1827 the botanist Robert Brown put pollen grains in water under a microscope and saw them twitch — a ceaseless, erratic jiggling that never stopped and never settled. He first guessed it was the "life force" of the pollen, then tested dust, soot, even ground-up bits of the Sphinx, and found that everything small enough did it. It was not biology. It was physics, and it would take 78 years to explain.

The puzzle: nothing visible is pushing the grain. The water looks perfectly still. Yet a particle roughly a micrometer across never stops moving, in random directions, forever. Where does the energy come from?

The answer is that the water is not still. It is a roiling crowd of molecules in violent thermal motion, and a grain suspended in it is struck from every side — about 10²¹ collisions per second for a 1 µm sphere. At any instant the kicks from one side slightly outnumber the kicks from the other, and that fleeting imbalance shoves the grain a fraction of a nanometer. Billions of times a second the imbalance reshuffles, and the accumulated nudges add up to a visible, jittering walk. The grain is a microscope for molecular chaos — too big to see molecules directly, but light enough to be jostled by them.

A random walk: why √t, not t

Strip the physics down to its skeleton and Brownian motion is a random walk. Imagine the grain takes a step of fixed length in a random direction, then another, then another, each independent of the last. After N steps, where is it?

On average, nowhere. Each step is as likely to point one way as the opposite, so the mean displacement vanishes:

⟨x⟩ = 0

But the particle clearly does move. The right quantity is the mean squared displacement, which does not cancel because squaring kills the sign. For independent steps the variances add, so:

⟨x²⟩ = N · ℓ²     (N steps of length ℓ)

Since the number of steps grows in proportion to elapsed time, ⟨x²⟩ ∝ t. The typical distance from the start is the square root of that:

typical distance ~ √⟨x²⟩ ∝ √t

This √t scaling is the fingerprint of diffusion, and it is merciless. To wander twice as far takes four times as long; ten times as far, a hundred times as long. It is why a sugar cube left undisturbed at the bottom of a coffee cup would take days to sweeten the top by diffusion alone — and why your cells use active pumps and stirring instead of waiting on it.

Einstein's 1905 result

In one of his four "miracle year" papers, Einstein derived the random walk from first principles and pinned down the proportionality constant. For motion along one axis:

⟨x²⟩ = 2 D t          (one dimension)
⟨r²⟩ = 6 D t          (three dimensions: x, y, z each add 2Dt)

Here D is the diffusion coefficient (units m²/s). Einstein's deeper move was to connect D to measurable molecular quantities by linking the random kicks to the steady drag a particle feels when it moves. Combining the diffusion law with Stokes' drag on a sphere gives the celebrated Stokes–Einstein relation:

D = kT / (6 π η r)

where k is Boltzmann's constant, T the absolute temperature, η the fluid's dynamic viscosity, and r the particle radius. Read it as thermal energy divided by drag: hotter fluid kicks harder (more D), while a bigger particle or thicker fluid drags more (less D).

This is the first written example of a fluctuation–dissipation theorem: the very same molecular collisions that randomly kick the particle are the ones that resist its motion as drag. Fluctuation and dissipation are two faces of one mechanism, locked together by temperature.

Crucially, k = R/N_A contains Avogadro's number. So measuring D, η, r, and T yields N_A — a count of how many molecules are in a mole. The dance of a pollen grain quietly encodes the number of atoms in matter.

Perrin counts the atoms

Einstein had made a falsifiable, parameter-free prediction; someone had to test it. Between 1908 and 1913 Jean Perrin did exactly that. He prepared near-perfect spheres of gamboge and mastic resin of known radius, watched them under a microscope, and recorded their positions at fixed time intervals — drawing the famous jagged trajectories by hand.

Plotting ⟨x²⟩ against t gave a straight line whose slope was 2D, exactly as Einstein predicted. Inverting the Stokes–Einstein relation, Perrin extracted Avogadro's number and got roughly 6 × 10²³ per mole. The same number fell out of the sedimentation of particles in gravity, and of several unrelated methods. Different physics, one consistent count — there was no escaping the conclusion. Atoms and molecules were real, discrete objects, not a bookkeeping convenience. Perrin received the 1926 Nobel Prize "for his work on the discontinuous structure of matter."

The numbers: how fast, how far

Plugging real values into D = kT/(6πηr) for spheres in water at 20 °C (η ≈ 1.0 × 10⁻³ Pa·s, kT ≈ 4.05 × 10⁻²¹ J) shows just how strongly size matters and how slow diffusion is.

Particle	Radius r	D in water (20 °C)	Time to drift ~1 µm (3D)
Water molecule	~0.14 nm	~1.5 × 10⁻⁹ m²/s	~110 µs
Protein (lysozyme)	~2 nm	~1.1 × 10⁻¹⁰ m²/s	~1.5 ms
Virus capsid	~50 nm	~4 × 10⁻¹² m²/s	~40 ms
Pollen-sized grain	~0.5 µm	~4 × 10⁻¹³ m²/s	~0.4 s
Visible dust mote	~5 µm	~4 × 10⁻¹⁴ m²/s	~4 s
Sand grain	~0.5 mm	~4 × 10⁻¹⁶ m²/s	~7 min (but it sinks long before)

D scales as 1/r, so a particle ten times bigger diffuses ten times slower — and because distance grows only as √(Dt), the larger particle barely seems to move at all on the timescale of a video. Above ~10 µm, gravity-driven sedimentation overwhelms the random jiggle entirely, which is why you see Brownian motion in micron-scale grains but never in a fleck of sand.

Two ways to write the same physics

Brownian motion can be described from either end of the microscope, and the two pictures must agree.

	Single-particle view	Population view
What it tracks	One tagged particle's path	Concentration of many particles
Governing equation	Langevin: m·a = −γv + ξ(t)	Diffusion eq.: ∂c/∂t = D∇²c
Output	A jagged random trajectory	A smooth spreading Gaussian cloud
Key prediction	⟨x²⟩ = 2Dt	Width of cloud grows as √(2Dt)
The bridge	Both share the same D. Probability of one particle = density of many.

The Langevin equation splits the molecular bombardment into a smooth drag term −γv (the average resistance, with γ = 6πηr for a sphere) plus a rapidly fluctuating random force ξ(t) (the moment-to-moment imbalance of kicks). The Fokker–Planck / diffusion equation drops the individual trajectory and tracks how the probability cloud spreads. Both contain the same diffusion coefficient because they describe the same underlying collisions — Einstein's relation γD = kT is what stitches them together.

Where Brownian motion shows up

• Cell biology. Proteins, ions, and signaling molecules reach their targets largely by Brownian diffusion. Enzymes meet substrates, oxygen crosses membranes, and neurotransmitters cross synapses — all on diffusion timescales that the √t law makes very short over nanometer gaps and hopelessly slow over centimeters (hence circulatory systems).
• Single-molecule microscopy. Particle-tracking measures ⟨x²⟩(t) of fluorescent beads or labeled molecules to read out local viscosity, membrane fluidity, and whether a molecule is freely diffusing, confined, or being actively transported (microrheology).
• Quantitative finance. The Black–Scholes option-pricing model treats stock prices as geometric Brownian motion. The Wiener process — the mathematical idealization of Brownian motion — underlies most of stochastic calculus.
• Optical tweezers. A focused laser traps a micron bead; its residual Brownian jiggle in the trap calibrates the trap stiffness and lets the system measure piconewton forces from single motor proteins.
• Nanofabrication and colloids. Paint, ink, milk, and blood are colloidal suspensions kept uniformly mixed by Brownian motion. It is also why nanoparticles must be stabilized against clumping.
• Statistical foundations. Brownian motion is the textbook entry point to the fluctuation–dissipation theorem, the central limit theorem in action, and the modern theory of stochastic processes.

Misconceptions and edge cases

• "The molecules push it in one direction." No — the kicks are isotropic on average. The motion comes from the fluctuation in the imbalance, not from any net flow. Remove the imbalance (a true continuum fluid) and the jitter vanishes.
• "It violates the second law — perpetual motion!" It does not. The energy is borrowed from the fluid's thermal energy and continuously returned; there is no way to extract net work from equilibrium fluctuations. A ratchet that tried to rectify the kicks would itself be jiggled into uselessness (Feynman's Brownian ratchet).
• "The particle drifts steadily away." The average displacement is exactly zero. It only spreads in the squared sense — the probability cloud widens around the start, it does not march off.
• "Bigger objects do it too, just slower." Technically yes, but for anything you can see with the naked eye D is so tiny — and gravity so dominant — that Brownian motion is utterly negligible. It is fundamentally a microscale phenomenon.
• "Velocity is well defined." The mathematical Brownian path is continuous but nowhere differentiable — zoom in and it stays jagged at every scale (it is fractal). A measured "velocity" depends entirely on your sampling interval; the physically meaningful quantity is ⟨x²⟩/t, not dx/dt.
• "It's the same as turbulence or convection." Those are bulk, deterministic fluid flows. Brownian motion is a single particle's response to molecular thermal noise in an otherwise quiescent fluid — a different beast entirely.