Brownian Motion

A speck of pollen, a fat droplet in milk, a colloidal sphere in water — anything small enough — is bombarded from every side by the thermal motion of the surrounding molecules. On average the impacts cancel, but in any short interval the cancellation is imperfect, and the residual net force jolts the particle along a jagged, unpredictable path. This is Brownian motion: the botanist Robert Brown watched it in 1827 under a microscope and could not explain it; Einstein in 1905, Smoluchowski in 1906, and Langevin in 1908 turned it into the founding calculation of nonequilibrium statistical mechanics and, in doing so, gave the atomic hypothesis its most direct experimental proof. The conceptual leap is that one neither tracks nor needs the $\sim10^{20}$ colliding molecules. The particle's coordinate $x(t)$ is treated as a random variable whose statistics follow from the central limit theorem: the displacement over a time long compared to a molecular collision yet short compared to observation is a sum of many independent kicks, hence Gaussian. The single sharpest consequence is that the mean-square displacement grows linearly in time, $\langle[x(t)-x(0)]^2\rangle = 2dDt$ in $d$ dimensions — the width of the packet spreads as $\sqrt{t}$ , the fingerprint of a random walk and not of ballistic motion. It was this $\sqrt{t}$ law, not the naive $t$ of steady drift, that Perrin measured to extract Avogadro's number and settle whether atoms are real. This article owns the phenomenon and the diffusion law $\langle x^2\rangle=2dDt$ , together with why it is linear in time. It stops there deliberately: the dynamical equation of motion is the Langevin equation, and the identity $D=\mu k_BT$ that fixes the diffusion constant is the Einstein relation — each is its own article. What matters here is the picture that seeds the whole tier: a mesoscopic object whose position is governed by molecular fluctuations, whose spreading is diffusive, and whose diffusion constant is set by friction and temperature.

⚡ Key Takeaways

Brownian motion is the perpetual, irregular wandering of a micron-sized particle suspended in a fluid, caused by the imbalance of incessant molecular impacts on its surface.
Einstein (1905) showed the mean-square displacement grows linearly in time, $\langle x^2\rangle = 2dDt$ , so the packet width spreads as $\sqrt{t}$ — diffusive, not ballistic.
The linearity is a central-limit-theorem statement: independent kicks add in variance, so $\langle x^2\rangle\propto N\propto t$ ; cross-correlations would break it.
The diffusion constant is fixed by friction and temperature, $D=k_BT/\gamma$ (the Einstein relation, article 1330), with the Stokes value $D=k_BT/6\pi\eta a$ for a sphere.
Perrin's measurements (1908–1913) turned the $\sqrt{t}$ law into Avogadro's number, $N_A\approx6.0\times10^{23}\,\mathrm{mol^{-1}}$ , and won the 1926 Nobel Prize — the reality of atoms, read off a dancing speck.
Key idea: a mesoscopic particle's coordinate is a random variable driven by molecular fluctuations; because successive displacements are independent, their variance adds and the mean-square displacement grows linearly in time, $\langle[x(t)-x(0)]^2\rangle=2dDt$ — the diffusive $\sqrt{t}$ spreading that killed the ballistic picture.

💡 The Core Insight

A mesoscopic particle's coordinate is a random variable driven by molecular fluctuations. Because successive displacements are statistically independent, their variances add, so the mean-square displacement grows linearly in time, $\langle[x(t)-x(0)]^2\rangle=2dDt$ in $d$ dimensions — the diffusive $\sqrt{t}$ spreading that distinguishes a random walk from ballistic motion.

The physical picture

Imagine a beach ball in a stadium crowd that shoves it at random. Each shove is tiny, but their net imbalance over any instant sends the ball lurching. The crowd is the fluid molecules; the ball is the colloid. Because the shoves are uncorrelated, the ball's position performs a random walk, and here is the crucial arithmetic: over $N$ independent kicks the displacement is a sum whose mean is zero, but whose variance is the sum of the individual variances. Variance grows like $N$ , so the typical distance grows like $\sqrt{N}$ . Since $N\propto t$ , the spread grows like $\sqrt{t}$ — the diffusive law. Had the kicks been correlated (a steady push), displacements would add coherently and the distance would grow like $t$ ; that is the ballistic law, and it is exactly what Brownian motion is not at long times. Two timescales organize the behavior. On scales shorter than the velocity relaxation time $\tau=m/\gamma$ , the particle coasts inertially and remembers its velocity; the MSD is momentarily ballistic, $\langle x^2\rangle\sim(k_BT/m)\,t^2$ . On longer scales the velocity has been randomized many times, momentum is forgotten, and the position diffuses, $\langle x^2\rangle\to 2Dt$ . A microscope samples the long-time regime, which is why Einstein's $\sqrt{t}$ law is what one actually sees. The deep symmetry lurking underneath is that the same molecular collisions both push the ball around (fluctuation) and slow any organized push you give it (dissipation). You cannot have one without the other at fixed temperature. That balance is the seed of the fluctuation–dissipation theorem; here it appears only as the fact that the diffusion constant is set by the friction — the full statement is developed in the Einstein-relation and fluctuation–dissipation articles.

Figure

Mean-square displacement vs time: the diffusive √t law (slope 1 on log-log) versus the ballistic parabola it replaces

Figure

A single Brownian trajectory as a chain of independent kicks, and the Gaussian packet its ensemble spreads into

The core equations

The central observable is the mean-square displacement. For times long compared to the velocity relaxation time, the Einstein diffusion law states that the variance of position grows linearly, with the diffusion constant $D$ as the single transport coefficient. This is the equation every microrheology measurement fits.

\langle [x(t)-x(0)]^2 \rangle = 2 d\, D\, t \qquad (t\gg\tau)

where: $d$ is the spatial dimensionality (so $2Dt$ in 1D, $6Dt$ in 3D); $D$ is the diffusion constant ( $\mathrm{m^2/s}$ ); $t$ is the elapsed time. Physically, the packet's standard deviation is $\sqrt{2Dt}$ — distance grows as the square root of time, the hallmark of accumulated independent randomness. The diffusion constant is not free: it is tied to the fluid friction and the temperature by the Einstein relation (derived in article 1330), and for a sphere the friction is Stokes drag, giving the Stokes–Einstein form used by Perrin.

D = \frac{k_B T}{\gamma}, \qquad \gamma = 6\pi\eta a \quad\Rightarrow\quad D = \frac{k_B T}{6\pi\eta a}

where: $\gamma$ is the friction coefficient (drag force per unit velocity); $\mu=1/\gamma$ is the mobility; $\eta$ is the fluid's dynamic viscosity; $a$ is the sphere radius. The reading is that a hotter, less viscous, or smaller particle diffuses faster, with $D\propto 1/a$ . Before the diffusive regime, on short timescales the particle coasts. The full mean-square displacement, valid at all times, interpolates between a ballistic $t^2$ and the diffusive $2Dt$ , with the crossover set by $\tau$ .

\langle x^2(t)\rangle = \frac{2k_BT}{\gamma}\Big[\,t - \tau\big(1-e^{-t/\tau}\big)\Big], \qquad \tau=\frac{m}{\gamma}

where: $\tau$ is the velocity (momentum) relaxation time; the bracket $\to t^2/2\tau$ for $t\ll\tau$ (ballistic) and $\to t$ for $t\gg\tau$ (diffusive). The velocity autocorrelation that generates it is $\langle v(0)v(t)\rangle=(k_BT/m)e^{-t/\tau}$ , whose $t=0$ value is the equipartition result $\langle v^2\rangle=k_BT/m$ .

Derivation AMean-square displacement from independent kicks

The goal is to establish the diffusion law $\langle x^2\rangle=2dDt$ from nothing but statistical independence, making explicit why it is linear in time rather than quadratic. We treat the displacement over an observation time $t$ as a sum of many small, independent, zero-mean steps taken over successive short intervals, and track how the variance accumulates. The central-limit-theorem logic is the whole content: independence converts a sum of steps into a variance that grows linearly with their number.

Step 1. Write the net displacement as a sum of $N=t/\Delta t$ steps, $x(t)-x(0)=\sum_{i=1}^N \delta_i$ , each with $\langle\delta_i\rangle=0$ (the fluid has no preferred direction) and a finite variance $\langle\delta_i^2\rangle=\langle\delta^2\rangle$ .

Step 2. Square and average: $\langle[x(t)-x(0)]^2\rangle=\sum_{i,j}\langle\delta_i\delta_j\rangle$ . Why the cross-terms die: molecular kicks decorrelate on times far shorter than $\Delta t$ , so $\langle\delta_i\delta_j\rangle=0$ for $i\neq j$ ; only the $N$ diagonal terms survive.

Step 3. Hence $\langle x^2\rangle=N\langle\delta^2\rangle=(t/\Delta t)\langle\delta^2\rangle$ . Why this is finite and physical: the diffusive limit keeps $\langle\delta^2\rangle\propto\Delta t$ as $\Delta t\to0$ , so the ratio $\langle\delta^2\rangle/\Delta t$ is a constant of the fluid, and we define $D\equiv\langle\delta^2\rangle/(2d\,\Delta t)$ .

Step 4. Collecting, $\langle x^2\rangle=2dDt$ . The linear-in- $t$ result is the signature of variance (not amplitude) accumulating; a coherent drift would give the sum $\propto N^2$ and the ballistic $t^2$ instead. Verified in verify_aq-sm-brownian-motion.py (checks the diffusive slope of the full MSD and $\int x^2 P\,dx=2Dt$ ).

Derivation BBallistic-to-diffusive crossover from the velocity correlation

The goal is to show that a single dynamical picture contains both the short-time ballistic $t^2$ and the long-time diffusive $2Dt$ , and that the crossover time is the momentum relaxation time $\tau=m/\gamma$ . We take the stationary velocity autocorrelation as given (it follows from the Langevin equation, article 1328, and equipartition) and integrate it twice, because position is the time integral of velocity. This is the calculation that makes the microscope's $\sqrt{t}$ law inevitable at long times.

Step 1. Position is $x(t)=\int_0^t v(t')\,dt'$ , so $\langle x^2(t)\rangle=\int_0^t\!\int_0^t\langle v(t_1)v(t_2)\rangle\,dt_1\,dt_2$ .

Step 2. Insert the stationary correlation $\langle v(t_1)v(t_2)\rangle=(k_BT/m)\,e^{-|t_1-t_2|/\tau}$ . Why this form: its $t=0$ value is $\langle v^2\rangle=k_BT/m$ (equipartition), and the exponential decay is the velocity losing memory over $\tau$ .

Step 3. Evaluate the double integral: $\langle x^2(t)\rangle=\dfrac{2k_BT}{\gamma}\big[t-\tau(1-e^{-t/\tau})\big]$ , exact for all $t$ .

Step 4. Short times, $t\ll\tau$ : expand $e^{-t/\tau}\approx1-t/\tau+t^2/2\tau^2$ to get $\langle x^2\rangle\to(k_BT/m)\,t^2$ — ballistic at the thermal speed $\sqrt{k_BT/m}$ . Why: the particle has not yet been stopped by friction.

Step 5. Long times, $t\gg\tau$ : the bracket $\to t$ , so $\langle x^2\rangle\to(2k_BT/\gamma)\,t=2Dt$ with $D=k_BT/\gamma$ — the Einstein relation read off the slope. Verified in verify_aq-sm-brownian-motion.py (short-time coefficient $k_BT/m$ , long-time slope $2k_BT/\gamma$ , Green–Kubo $D=\int_0^\infty\langle v(0)v(t)\rangle dt=k_BT/\gamma$ ).

Limiting cases

Case	Result	Meaning
Short-time ballistic $t\ll\tau=m/\gamma$	$\langle x^2\rangle\simeq (k_BT/m)\,t^2$	The particle coasts at its thermal velocity $\sqrt{k_BT/m}$ before friction acts; displacement $\propto t$ , MSD $\propto t^2$ .
Long-time diffusive $t\gg\tau$	$\langle x^2\rangle\simeq 2dDt$	Velocity memory is lost; a pure random walk with $D=k_BT/\gamma$ ; the observed $\sqrt{t}$ spreading.
Overdamped (Smoluchowski) $m\to0$	$\gamma\dot x=-V'(x)+\xi(t)$	Inertia negligible, velocity equilibrates instantly; only position dynamics survive (the overdamped Langevin / Smoluchowski limit).
Confining harmonic trap $\tfrac12 m\omega_0^2 x^2$	$\langle x^2\rangle\to k_BT/(m\omega_0^2)$	The MSD saturates at the equipartition value instead of growing without bound — a trapped bead, not a free walker.

⚠️ Diffusive is not ballistic — and the noise is singular

The typical Brownian displacement scales as $\sqrt{t}$ , not $t$ : the long-time trajectory is diffusive, and confusing it with the short-time ballistic regime ( $\langle x^2\rangle\sim t^2$ ) is the classic error. Equally, do not treat the random force as a smooth function of time. In the idealized white-noise limit it is singular, $\langle\xi(t)\xi(t')\rangle\propto\delta(t-t')$ , with no ordinary time derivative; the velocity is continuous but the position path is nowhere differentiable. Handling that singular object correctly is precisely why stochastic calculus (Itô vs Stratonovich, article 1332) and the Wiener process (article 1331) exist.

Worked examples

Example 1 — Stokes–Einstein diffusion constant of a colloidal bead. Take a sphere of radius $a=1\,\mu\mathrm{m}$ in water at $T=300\,\mathrm{K}$ , viscosity $\eta=10^{-3}\,\mathrm{Pa\cdot s}$ .

Step 1. Friction: $\gamma=6\pi\eta a = 6\pi(10^{-3})(10^{-6})=1.885\times10^{-8}\,\mathrm{kg/s}$ .

Step 2. Diffusion constant: $D=k_BT/\gamma=(1.381\times10^{-23})(300)/(1.885\times10^{-8})$ . Working it out, $D=\mathbf{2.20\times10^{-13}\ \mathrm{m^2/s}}$ .

Step 3. Root-mean-square displacement in one second (1D): $\sqrt{\langle x^2\rangle}=\sqrt{2Dt}=\sqrt{2(2.20\times10^{-13})(1)}=\mathbf{6.63\times10^{-7}\ \mathrm{m}\approx0.66\ \mu m}$ . A real, microscope-visible distance — the bead wanders about two-thirds of its own radius per second.

Example 2 — Perrin's Avogadro number, backwards. Suppose the same bead is observed to have a 1D mean-square displacement $\langle x^2\rangle$ after time $t$ . Perrin's method inverts the Stokes–Einstein law to get $N_A$ .

Step 1. From $\langle x^2\rangle=2Dt=k_BTt/(3\pi\eta a)$ and $k_B=R/N_A$ , solve: $N_A=\dfrac{RTt}{3\pi\eta a\,\langle x^2\rangle}$ .

Step 2. For our bead $\langle x^2\rangle=2Dt=4.39\times10^{-13}\,\mathrm{m^2}$ at $t=1\,\mathrm{s}$ ; plugging back, $N_A=\dfrac{(8.314)(300)(1)}{3\pi(10^{-3})(10^{-6})(4.39\times10^{-13})}=\mathbf{6.02\times10^{23}\ \mathrm{mol^{-1}}}$ .

Step 3. This matches the CODATA value $6.022\times10^{23}$ to the precision of the inputs — the number Perrin actually read off dancing specks, and the closing of the debate over whether atoms are real.

Robert Brown

1773–1858 · Scottish botanist, keeper of the Banksian collection at the British Museum

Robert Brown was a meticulous botanist, not a physicist. In 1827, examining pollen grains of Clarkia pulchella suspended in water, he saw the tiny particles ejected from the grains jitter ceaselessly. His decisive control was to check inorganic dust and even a fragment of the Sphinx — the motion persisted, so it was not "life." He described it faithfully and left the cause open. It took three-quarters of a century and Einstein's statistical argument to explain the "Brownian" motion that bears his name; Brown's contribution was the rigor to establish that the phenomenon was real, universal, and physical rather than biological — a reminder that a well-posed observation can outlast every wrong explanation offered for it.

⏳ History

1827

— Robert Brown observes the ceaseless jitter of pollen-derived particles under the microscope

1828

— Brown publishes "A Brief Account of Microscopical Observations," ruling out a biological cause

1905

— Einstein derives ⟨x²⟩=2Dt and the Stokes–Einstein relation, predicting the effect from atomism

1905

— William Sutherland independently obtains D=k_BT/6πηa (the Sutherland–Einstein relation)

1906

— Marian von Smoluchowski gives an independent kinetic derivation

1908

— Paul Langevin writes the stochastic equation of motion, an "infinitely simpler" route

1909

— Jean Perrin measures Brownian displacements and extracts Avogadro's number

1926

— Perrin awarded the Nobel Prize in Physics for confirming the atomic nature of matter

Where this lives today

Brownian motion is the everyday workhorse of soft-matter and biophysics measurement. In single-particle tracking and microrheology, fitting $\langle x^2(t)\rangle$ of an embedded probe bead reads out the local viscosity or the full frequency-dependent viscoelastic modulus of a cell, gel, or polymer melt; a sub-linear exponent ( $\langle x^2\rangle\sim t^\alpha$ , $\alpha<1$ ) flags crowding or transient binding. Optical and magnetic tweezers are calibrated through the Lorentzian position power spectrum of a trapped bead and the equipartition $\langle x^2\rangle=k_BT/\kappa$ , setting the piconewton force scale used to pull on single DNA and motor proteins. The mathematical skeleton — the Wiener process and the diffusion equation — reaches far beyond physics. Johnson–Nyquist thermal noise in a resistor is the electrical Brownian motion; the Black–Scholes model prices options with geometric Brownian motion; and diffusion-limited kinetics governs reaction rates in crowded cellular interiors. Every one of these inherits the single lesson of this article: independent randomness accumulates in variance, so signals spread as $\sqrt{t}$ .

Going further

References. Pathria & Beale, Statistical Mechanics, 3rd ed., Ch. 13 (The Theory of the Brownian Motion) · Kardar, Statistical Physics of Particles, Ch. 9 · Reif, Fundamentals of Statistical and Thermal Physics, Ch. 15 · Sethna, Statistical Mechanics: Entropy, Order Parameters, and Complexity, Ch. 2 · Einstein, Ann. Phys. 17, 549 (1905) · Langevin, C. R. Acad. Sci. 146, 530 (1908).

Provenance. The diffusion law $\langle x^2\rangle=2dDt$ , the Stokes–Einstein $D=k_BT/6\pi\eta a$ , the velocity autocorrelation, and the full MSD are verified against Pathria & Beale Ch. 13, Kardar Ch. 9, and Reif Ch. 15. verify_aq-sm-brownian-motion.py recomputes the double-integral MSD and its short/long-time limits, confirms the Gaussian propagator solves the diffusion equation with $\int x^2 P=2Dt$ , checks the Stokes–Einstein number ( $D=2.20\times10^{-13}\,\mathrm{m^2/s}$ ), and closes the Perrin formula onto the CODATA $N_A=6.022\times10^{23}$ .

Builds on: Equipartition Theorem · Central Limit Theorem · Kinetic Theory of Gases. Leads to: The Langevin Equation · The Diffusion Equation · The Einstein Relation · Wiener Process and White Noise · The Fokker–Planck Equation · The Ornstein–Uhlenbeck Process.

Is Brownian motion the same as diffusion?

They are two views of one phenomenon. "Brownian motion" names the microscopic random trajectory of a single particle; "diffusion" names the macroscopic spreading of the probability density (or concentration) of many such particles. Einstein's insight was that the two are quantitatively identical: the single-particle mean-square displacement $\langle x^2\rangle=2dDt$ uses exactly the diffusion constant $D$ that appears in the diffusion equation $\partial_t P=D\nabla^2 P$ (article 1329). One is the sample path; the other is the ensemble density that path is drawn from.

Why doesn't a Brownian particle ever come to rest?

Because it is in thermal contact with the fluid at temperature $T>0$ . Equipartition guarantees a mean kinetic energy $\tfrac12 m\langle v^2\rangle=\tfrac{d}{2}k_BT$ that friction cannot remove — the same collisions that dissipate organized motion continually re-inject random energy. "Coming to rest" would violate the second law by letting the particle sit colder than its surroundings. The particle relaxes only its directed velocity (on the timescale $\tau=m/\gamma$ ), never its thermal agitation.

If the average displacement is zero, what is actually "moving"?

The mean displacement $\langle x(t)-x(0)\rangle$ is indeed zero — the fluid has no preferred direction. What grows is the variance $\langle[x(t)-x(0)]^2\rangle=2dDt$ , i.e. the spread of possible positions. A single particle does travel a net distance of order $\sqrt{2dDt}$ , but its sign is random, so averaging over many particles (or many runs) cancels. Brownian "motion" is the growth of the distribution's width, not of its center.

Why is the mean-square displacement linear in time and not quadratic?

Quadratic growth ( $\langle x^2\rangle\propto t^2$ ) is what you get from coherent motion — a steady drift, where displacements add up in phase. Brownian steps are instead independent and zero-mean, so they add in variance, not amplitude: over $N\propto t$ steps the variance grows like $N$ , giving $\langle x^2\rangle\propto t$ . The only place you see the quadratic law is the very short-time ballistic window $t\ll\tau$ , before friction has randomized the velocity even once.

What exactly does "mesoscopic" mean here, and why does it matter?

Mesoscopic means intermediate: the particle is large enough that the fluid can be treated as a continuum with a well-defined viscosity and temperature (so Stokes drag and equipartition apply), yet small enough that the residual fluctuations of the molecular bombardment are visible as motion. A grain of sand is too big — fluctuations are utterly negligible. A single molecule is too small — there is no separation of scales. The colloidal micron scale is exactly where statistical mechanics bridges the molecular and the macroscopic, which is why it was the decisive test of atomism.

How did Perrin get Avogadro's number from a jiggling speck?

By turning the Stokes–Einstein relation into an equation with only one unknown. He measured the mean-square displacement $\langle x^2\rangle$ of beads of known radius $a$ in a fluid of known viscosity $\eta$ at known $T$ , over a measured time $t$ . Since $\langle x^2\rangle=2Dt=k_BTt/(3\pi\eta a)$ and $k_B=R/N_A$ with $R$ the (macroscopic) gas constant, everything but $N_A$ is measurable, giving $N_A=RTt/(3\pi\eta a\langle x^2\rangle)\approx6\times10^{23}$ . Its agreement with values from utterly different methods is what convinced the last skeptics that atoms are real.

Does inertia ever matter for a real colloidal bead?

Only on timescales far shorter than an experiment usually resolves. The momentum relaxation time is $\tau=m/\gamma=m/6\pi\eta a$ ; for a micron water-borne bead it is of order $10^{-7}\,\mathrm{s}$ , below optical tracking rates. So the ballistic $t^2$ regime is real but nearly always unresolved, and the overdamped (Smoluchowski) description — drop the inertial term entirely — is an excellent approximation. The short-time ballistic regime was finally measured directly in optically trapped beads around 2010, confirming the crossover the full MSD formula predicts.