Brownian Motion Introduction

Brownian motion is the random, ceaseless jittering of small particles suspended in a fluid, caused by collisions with the fluid's molecules. Its mathematical idealization is the Wiener process: a continuous-time stochastic process whose increments are independent, Gaussian, and stationary. Key result: the mean squared displacement grows linearly in time,

⟨ x^{2} ⟩ = 2 D t

. It applies across physics, finance, biology, and probability theory. Named for botanist Robert Brown (1827); explained by Einstein (1905) and formalized by Norbert Wiener (1923).

Brownian motion refers both to the physical phenomenon of erratic particle motion in a fluid and to its rigorous mathematical model, the Wiener process. Physically, a pollen grain in water is bombarded by roughly $1 0^{21}$ water molecules per second, and the tiny imbalances in this bombardment produce a visible random walk. Mathematically, it is the canonical example of a continuous-time, continuous-path stochastic process with the Markov property and stationary independent increments — the cornerstone of stochastic calculus. It matters because it is the scaling limit of nearly every random walk (a probabilistic Central Limit Theorem in disguise), the foundation of diffusion theory, and the engine behind the Black–Scholes model of finance. The surprising fact: although a Brownian path is continuous everywhere, it is differentiable nowhere — the particle has no well-defined velocity at any instant.

The Core Insight

A microscopic particle does not move smoothly. It is pushed by an enormous number of independent molecular collisions, each tiny. By the Central Limit Theorem, the sum of these many small independent kicks over a time interval is Gaussian-distributed, with variance proportional to the length of the interval. Add the requirements of continuity and independence between disjoint intervals, and the mathematical object that emerges is uniquely determined — the Wiener process. Brownian motion is therefore what you must get whenever many small independent random influences accumulate continuously in time.

Formal Statement

The Wiener Process (Standard Brownian Motion)

A real-valued stochastic process ${W_{t}}_{t \geq 0}$ on a probability space is called standard Brownian motion (or the Wiener process) if it satisfies:

Start at origin: $W_{0} = 0$ almost surely.
Independent increments: for $0 \leq t_{0} < t_{1} < \dots < t_{n}$ , the random variables $W_{t_{1}} - W_{t_{0}}, W_{t_{2}} - W_{t_{1}}, \dots, W_{t_{n}} - W_{t_{n - 1}}$ are mutually independent.
Gaussian, stationary increments: for $0 \leq s < t$ ,

W_{t} - W_{s} \sim N (0, t - s),

⟳this is the defining distributional axiom; the variance equals the elapsed time

t - s

, which fixes the scale of "standard" Brownian motion.

Continuous paths: with probability $1$ , the map $t \mapsto W_{t}$ is continuous.

Here $N (μ, σ^{2})$ denotes the normal distribution with mean $μ$ and variance $σ^{2}$ .

The transition probability density — the chance of being near position $x$ at time $t$ given a start at the origin — is the heat kernel:

p (x, t) = \frac{1}{4 π D t} exp (- \frac{x ^{2}}{4 D t}),

⟳solve the diffusion equation

\partial_{t} p = D \partial_{x}^{2} p

with point initial condition

p (x, 0) = δ (x)

via Fourier transform; for the standard Wiener process set

2 D = 1

The mean squared displacement (Einstein's relation) follows directly:

⟨ x^{2} (t)⟩ = \int_{- \infty}^{\infty} x^{2} p (x, t) d x = 2 D t,

⟳the variance of

N (0, 2 D t)

2 D t

; equivalently use

⟨ x^{2} ⟩ = - 2 \partial_{λ} \tilde{p} ∣_{λ = 0}

on the characteristic function

\tilde{p} (λ) = e^{- D λ^{2} t}

Einstein further connected the diffusion coefficient to thermodynamic quantities, the Stokes–Einstein relation:

D = \frac{k _{B} T}{6 π η a},

⟳balance the diffusive flux against the drift induced by an external force using the fluctuation–dissipation argument; the friction coefficient

γ = 6 π η a

comes from Stokes' law for a sphere of radius

a

in fluid of viscosity

η

Symbols: $W_{t}$ position of the idealized process at time $t$ ; $D$ diffusion coefficient (units $m^{2} /s$ ); $k_{B}$ Boltzmann constant ( $1.38 \times 1 0^{- 23} J/K$ ); $T$ absolute temperature (K); $η$ dynamic viscosity (Pa·s); $a$ particle radius (m).

Covariance Structure

For standard Brownian motion, the covariance is

Cov (W_{s}, W_{t}) = E [W_{s} W_{t}] = min (s, t) .

⟳for

s \leq t

write

W_{t} = W_{s} + (W_{t} - W_{s})

; the second term is independent of

W_{s}

with mean zero, so

E [W_{s} W_{t}] = E [W_{s}^{2}] = s = min (s, t)

In fact, properties (1), (4), and the Gaussian covariance $[5]$ together provide an equivalent characterization: standard Brownian motion is the unique continuous Gaussian process with mean zero and covariance $min (s, t)$ .

Why It Holds / Derivation

Brownian Motion as the Scaling Limit of a Random Walk

We show how the Wiener process emerges from a simple symmetric random walk, making axioms (2)–(3) inevitable.

Step 1 — Set up the walk. Let

ξ_{1}, ξ_{2}, \dots

be i.i.d. with

P (ξ_{i} = + 1) = P (ξ_{i} = - 1) = \frac{1}{2}

, so

E [ξ_{i}] = 0

Var (ξ_{i}) = 1

. Define the partial sums

S_{n} = \sum_{i = 1}^{n} ξ_{i}

. Step 2 — Rescale space and time. Take

N

steps per unit time, each of size

1/ N

. Define

W_{t}^{(N)} = \frac{1}{N} S_{⌊ N t ⌋} .

⟳this is diffusive scaling — to keep the variance finite as steps shrink, space must scale as

time

, not linearly.

Step 3 — Check the variance. Since the

ξ_{i}

are independent with unit variance,

Var (W_{t}^{(N)}) = \frac{1}{N} \cdot ⌊ N t ⌋ N \to \infty t .

⟳variance of a sum of independent terms adds; the factor

1/ N

outside cancels the

N

inside, leaving

t

Step 4 — Apply the Central Limit Theorem. As

N \to \infty

, the rescaled sum

\frac{1}{N} S_{⌊ N t ⌋}

converges in distribution to

N (0, t)

. More strongly, Donsker's invariance principle asserts that the entire path

W_{\cdot}^{(N)}

converges in distribution (on the space of continuous functions) to the Wiener process

W_{\cdot}

. Step 5 — Inherit the axioms. Increments of the walk over disjoint blocks of steps are independent (the

ξ_{i}

are independent), giving axiom (2). Each block's increment is a normalized sum, hence Gaussian in the limit with variance equal to the elapsed time, giving axiom (3). Continuity (axiom 4) is guaranteed by the shrinking step size and Kolmogorov's continuity criterion.

■

Albert Einstein | 1879–1955

In his annus mirabilis of 1905, Einstein published "On the Motion of Small Particles Suspended in a Stationary Liquid." Crucially, he was not trying to explain Brown's observations per se — his goal was to prove the physical reality of atoms and molecules, then still contested. He reasoned that if matter were truly molecular, the random collisions should produce measurable particle wandering, and he derived $⟨ x^{2} ⟩ = 2 D t$ with $D = k_{B} T / (6 π η a)$ . This gave an experimentally accessible route to Avogadro's number. Jean Perrin's careful measurements (1908–1909) confirmed Einstein's predictions and earned Perrin the 1926 Nobel Prize, finally settling the debate over the existence of atoms.

Building Intuition

Picture a large beach ball (the visible particle) floating among a swarm of tennis balls (water molecules) thrown at it from all directions. At any instant, slightly more tennis balls happen to strike the left side than the right, so the beach ball lurches right. A moment later the imbalance reverses. Because the molecules are independent and the imbalance is a fresh random draw each instant, the beach ball's position is a running sum of unpredictable kicks.

Two counterintuitive consequences follow. First, the particle drifts nowhere on average ( $E [W_{t}] = 0$ ), yet it travels arbitrarily far given enough time — the typical distance grows as $t$ , not $t$ . Second, the path is infinitely wiggly: between any two times, no matter how close, there are still countless collisions, so zooming in never reveals a smooth straight segment. This is why the path is continuous but nowhere differentiable — there is no instantaneous velocity.

Mean squared displacement: $\langle x^2\rangle = 2Dt$ grows linearly, while typical displacement $\sqrt{\langle x^2\rangle}\propto\sqrt t$ grows sublinearly

Worked Examples

Diffusion of a Pollen Grain in Water

Given: A spherical particle of radius

a = 0.5 μ m = 5 \times 1 0^{- 7} m

in water at

T = 293 K

, viscosity

η = 1.0 \times 1 0^{- 3} Pa \cdot s

. Find

D

, and the typical distance moved in

t = 10 s

. Step 1 — Compute $D$ from $[4]$ :

D = \frac{k _{B} T}{6 π η a} = \frac{( 1.38 \times 1 0 ^{- 23} ) ( 293 )}{6 π ( 1.0 \times 1 0 ^{- 3} ) ( 5 \times 1 0 ^{- 7} )} .

Numerator: $1.38 \times 1 0^{- 23} \times 293 = 4.04 \times 1 0^{- 21} J$ . Denominator: $6 π \times 1 0^{- 3} \times 5 \times 1 0^{- 7} = 9.42 \times 1 0^{- 9} Pa \cdot s \cdot m$ .

D = \frac{4.04 \times 1 0 ^{- 21}}{9.42 \times 1 0 ^{- 9}} = 4.3 \times 1 0^{- 13} m^{2} /s .

Step 2 — Mean squared displacement from $[3]$ (using the 1-D form):

⟨ x^{2} ⟩ = 2 D t = 2 (4.3 \times 1 0^{- 13}) (10) = 8.6 \times 1 0^{- 12} m^{2} .

Step 3 — Typical (root-mean-square) displacement:

⟨ x^{2} ⟩ = 8.6 \times 1 0^{- 12} = 2.9 \times 1 0^{- 6} m .

D \approx 4.3 \times 1 0^{- 13} m^{2} /s, x_{rms} \approx 2.9 μ m

The grain wanders only a few micrometres in 10 seconds — visible under a microscope but agonizingly slow.

Probability Computation for the Wiener Process

Given: Standard Brownian motion

W_{t}

. (a) Find

P (W_{4} > 2)

. (b) Find

P (W_{1} > 0, W_{2} > 0)

using covariance. (a) By axiom

[1]

W_{4} \sim N (0, 4)

, so its standard deviation is

σ = 2

. Standardize:

P (W_{4} > 2) = P (Z > \frac{2 - 0}{2}) = P (Z > 1) = 1 - Φ (1) .

With

Φ (1) = 0.8413

P (W_{4} > 2) = 0.1587.

(b)

(W_{1}, W_{2})

is jointly Gaussian, mean zero, with

Var (W_{1}) = 1

Var (W_{2}) = 2

, and from

[5]

Cov (W_{1}, W_{2}) = min (1, 2) = 1

. The correlation is

ρ = \frac{1}{1 2} = \frac{1}{2} \approx 0.707.

For a bivariate normal both-positive orthant probability:

P (W_{1} > 0, W_{2} > 0) = \frac{1}{4} + \frac{arcsin ρ}{2 π} = \frac{1}{4} + \frac{arcsin ( 0.707 )}{2 π} = \frac{1}{4} + \frac{0.785}{6.283} .

P (W_{4} > 2) \approx 0.159, P (W_{1} > 0, W_{2} > 0) \approx 0.375

Common Misconceptions

"The Brownian particle has a velocity, it just changes rapidly." False. The Wiener path is nowhere differentiable, so the limit

lim_{h \to 0} (W_{t + h} - W_{t}) / h

does not exist at any

t

. Over an interval of length

h

the displacement is of order

h

, so the difference quotient behaves like

1/ h \to \infty

. A physical particle does have a true (Maxwell–Boltzmann) velocity, but the idealized mathematical model deliberately discards inertia, which is why it has no velocity. The fuller physical picture is the Ornstein–Uhlenbeck / Langevin model, in which momentum relaxes over a tiny but finite time. "Brownian motion is caused by convection currents, evaporation, or stirring." No. Brown himself ruled these out by observing the motion in sealed drops and even in fluid inclusions trapped in quartz for millions of years. The motion persists in perfectly still, isothermal fluid — it is intrinsic thermal molecular agitation, not bulk fluid flow.

A continuous random walk is NOT automatically Brownian motion

Consider a process built from a single coin flip $ξ = \pm 1$ : $X_{t} = ξ \cdot t$ . Its paths are continuous and it is random. But it fails to be Brownian motion: its increments are not independent ( $X_{2} - X_{1} = ξ = X_{1} - X_{0}$ , perfectly correlated), and $Var (X_{t}) = t^{2}$ , not $t$ . Randomness plus continuity are far from sufficient — the independence and Gaussian variance-linear-in-time structure is essential. This shows why all four axioms are needed to pin down the Wiener process.

Connections

Brownian motion sits at the crossroads of several fields. The transition density $[2]$ is precisely the Green's function of the Heat Equation, so diffusion of probability and diffusion of heat are the same mathematics. As the scaling limit of the Random Walk via Donsker's theorem, it is a path-space generalization of the Central Limit Theorem. It is the driving noise of Stochastic Differential Equations and the integrator in Itô Calculus, where the quadratic variation identity $(d W)^{2} = d t$ replaces ordinary calculus rules. Its connection to potential theory and harmonic functions links it to the Laplace Equation: the probability of a Brownian particle hitting a region is harmonic. Finally, the Markov Property of $W_{t}$ — the future depends on the present alone — places it among continuous-time Markov Processes.

Real-World Applications

Financial mathematics (Black–Scholes, 1973): Stock prices are modeled as geometric Brownian motion $d S = μ S d t + σ S d W_{t}$ , the foundation of option pricing and modern quantitative finance.
Statistical physics — measuring Avogadro's number: Perrin's tracking of resin particles, combined with Einstein's $[3]$ – $[4]$ , yielded $N_{A} \approx 6 \times 1 0^{23}$ and proved atoms exist.
Biology — single-particle tracking: Diffusion coefficients of proteins and lipids in cell membranes are extracted by fitting observed trajectories to $⟨ r^{2} ⟩ = 4 D t$ (2-D), revealing crowding and anomalous subdiffusion.
Chemical kinetics: Diffusion-limited reaction rates (Smoluchowski theory) use Brownian motion to compute how fast reactants encounter one another.
Signal processing & control: Thermal (Johnson–Nyquist) noise in resistors and the random forcing in Kalman filters are modeled as Brownian/white-noise processes.
Climate and ecology: Random dispersal of pollutants, seeds, and animal foraging paths are modeled with diffusion and Lévy/Brownian motion.
Computer science — randomized algorithms: Stochastic gradient descent dynamics near minima are analyzed using Brownian approximations and the associated Fokker–Planck equation.

Historical Development

1827 — Robert Brown, examining pollen of Clarkia pulchella in water, observes the persistent jittering of suspended particles; rules out that it is biological "life force" by seeing it in inorganic dust.
1880 — Thorvald Thiele gives an early mathematical treatment of least squares involving what amounts to Brownian increments.
1900 — Louis Bachelier, in his thesis Théorie de la spéculation, models stock prices with Brownian motion — the first mathematical use, five years before Einstein, but largely overlooked.
1905 — Albert Einstein derives $⟨ x^{2} ⟩ = 2 D t$ and the Stokes–Einstein relation, linking the motion to molecular reality and Avogadro's number.
1906 — Marian Smoluchowski independently develops a kinetic theory of Brownian motion.
1908 — Paul Langevin introduces the stochastic differential equation $m \overset{x}{¨} = - γ \overset{x}{˙} + F (t)$ , giving the modern force-based formulation.
1908–1909 — Jean Perrin experimentally confirms Einstein's predictions, settling the atomic debate (Nobel Prize 1926).
1923 — Norbert Wiener constructs the process rigorously as a measure on continuous functions, proving its existence; the Wiener process is born.
1944–1951 — Kiyoshi Itô develops stochastic calculus and the Itô integral, providing the analytic machinery to do calculus with Brownian paths.

Subtleties & Limits

Several deep and active questions surround Brownian motion. (1) Path properties: the exact modulus of continuity is governed by Lévy's theorem ( $∣ W_{t + h} - W_{t} ∣ \sim 2 h lo g (1/ h)$ ), and the Hausdorff dimension of the graph is $3/2$ — fine fractal structure with still-debated multifractal refinements. (2) Anomalous diffusion: many real systems (crowded cells, porous media, financial markets) show $⟨ x^{2} ⟩ \sim t^{α}$ with $α \neq = 1$ ; characterizing when and why ordinary Brownian scaling breaks down (fractional Brownian motion, continuous-time random walks) remains an active research frontier. (3) Higher-dimensional intersections: whether independent Brownian paths intersect depends subtly on dimension (they do in $d \leq 3$ , generically not in $d \geq 4$ ), and the geometry of intersection-local-time and the Brownian frontier connects to Schramm–Loewner Evolution (SLE) and 2-D conformal field theory — a Fields-Medal-winning area. (4) Quantum analogues: the precise mathematical status of "quantum Brownian motion" and the Euclidean (imaginary-time) path-integral relation $e^{- t H}$ to the heat semigroup remains a rich, only partially rigorous program.