Diffusion Equation, Random Walk and Brownian Motion
CSIR NET Part C: random walk statistics, diffusion equation and its Gaussian solution, Fick's laws, Langevin equation, Einstein and Stokes–Einstein relations.
- 16 min
On this page · 12 sections
Prerequisites
- Counting, Probability and Number Properties: binomial counting, which is all a 1D random walk is.
- Partial Differential Equations: Laplace, Wave and Heat: separation of variables for the heat (diffusion) equation.
- Green’s Functions: the Gaussian spreading kernel is the Green’s function of the diffusion equation.
- Equipartition from Canonical and Grand Canonical Ensembles: \( \frac{1}{2}m\langle v_x^2\rangle = \frac{1}{2}k_BT \) fixes the noise strength below.
The Hook: Pollen That Never Settles
The Random Walk
One dimension
A walker takes \( N \) steps of length \( a \): right with probability \( p \), left with \( q = 1 - p \). If \( n_R \) steps go right, the displacement is \( x = (2n_R - N)a \) and
$$P(n_R) = \binom{N}{n_R}p^{n_R}q^{N - n_R}$$Each step contributes mean \( (p - q)a \) and variance \( a^2 - (p - q)^2a^2 = 4pqa^2 \). Steps are independent, so means and variances add:
$$\boxed{\langle x\rangle = N(p - q)a, \qquad \sigma_x^2 = \langle x^2\rangle - \langle x\rangle^2 = 4Npq\,a^2}$$For the symmetric walk (\( p = q = \frac{1}{2} \)): \( \langle x\rangle = 0 \), \( \langle x^2\rangle = Na^2 \), so the root-mean-square distance is
$$\boxed{x_{\text{rms}} = a\sqrt N}$$The spread grows as the square root of the number of steps, while a directed walk would cover \( Na \). That one fact is the whole of diffusion.
Return to the origin. After \( 2k \) steps of a symmetric walk,
$$P(x = 0) = \binom{2k}{k}\frac{1}{2^{2k}} \approx \frac{1}{\sqrt{\pi k}}$$e.g. \( \binom{4}{2}/16 = 0.375 \) after four steps. After an odd number of steps the walker cannot be at the origin.
Gaussian limit. For large \( N \) the binomial tends to a Gaussian with the same mean and variance (central limit theorem):
$$P(x) \approx \frac{1}{\sqrt{2\pi\sigma_x^2}}\exp\left[-\frac{(x - \langle x\rangle)^2}{2\sigma_x^2}\right]$$(as a density on the continuum; on the lattice only every other site is reachable, which doubles the probability per allowed site).
Higher dimensions
In \( d \) dimensions with independent steps of length \( a \) in random directions, cross terms average to zero:
$$\boxed{\langle R^2\rangle = Na^2}$$independent of \( d \). The freely jointed chain model of a polymer of \( N \) segments of length \( b \) uses exactly this: end-to-end distance \( R_{\text{rms}} = b\sqrt N \), much smaller than the contour length \( Nb \). Its entropy, \( S(R) = \text{const} - \dfrac{3k_BR^2}{2Nb^2} \), makes the chain an entropic spring with force constant \( 3k_BT/Nb^2 \): stiffer when hotter, the origin of rubber elasticity.
Recurrence (Pólya). A symmetric lattice walk in 1D or 2D returns to its starting point with probability 1; in 3D it returns with probability about 0.34 and otherwise escapes for ever.
From Random Walk to the Diffusion Equation
Let steps of length \( a \) occur every \( \tau \). For the symmetric walk,
$$P(x, t + \tau) = \tfrac{1}{2}P(x - a, t) + \tfrac{1}{2}P(x + a, t)$$Taylor-expand to first order in \( \tau \) and second order in \( a \):
$$\tau\frac{\partial P}{\partial t} = \frac{a^2}{2}\frac{\partial^2P}{\partial x^2} \ \Rightarrow\ \boxed{\frac{\partial P}{\partial t} = D\frac{\partial^2P}{\partial x^2}, \qquad D = \frac{a^2}{2\tau}}$$With bias, a drift term appears: \( \partial_tP = -v\,\partial_xP + D\,\partial_x^2P \) with \( v = (p - q)a/\tau \) and \( D = 2pq\,a^2/\tau \) (which reduces to \( a^2/2\tau \) at \( p = \frac{1}{2} \)).
Fick’s laws
For a concentration \( n(\mathbf r, t) \) of many independent walkers:
$$\boxed{\mathbf J = -D\nabla n \ \text{(Fick's first law)}}, \qquad \frac{\partial n}{\partial t} + \nabla\cdot\mathbf J = 0 \ \Rightarrow\ \boxed{\frac{\partial n}{\partial t} = D\nabla^2n \ \text{(second law)}}$$Same mathematics as heat conduction (\( \partial_tT = \kappa_{\text{th}}\nabla^2T \) with thermal diffusivity \( \kappa_{\text{th}} \)).
The point-source solution
For \( N_0 \) particles released at the origin at \( t = 0 \), in \( d \) dimensions:
$$\boxed{n(\mathbf r, t) = \frac{N_0}{(4\pi Dt)^{d/2}}\exp\left(-\frac{r^2}{4Dt}\right)}, \qquad \boxed{\langle r^2\rangle = 2dDt}$$So \( \langle x^2\rangle = 2Dt \) in 1D, \( 4Dt \) in 2D, \( 6Dt \) in 3D. The peak concentration falls as \( t^{-d/2} \). Comparing with the walk, \( \langle x^2\rangle = Na^2 = (t/\tau)a^2 = 2Dt \) confirms \( D = a^2/2\tau \).
Time to diffuse a distance \( L \): \( t \sim L^2/D \). For a small molecule in water (\( D \approx 10^{-9} \) m\( ^2 \) s\( ^{-1} \)) to reach an rms distance of 1 μm in 3D takes \( L^2/6D \approx 0.2 \) ms; to reach 1 cm takes about \( 1.7\times10^4 \) s, nearly five hours. That is why cells can rely on diffusion but organisms need circulation.
Modes in a finite region
In a slab \( 0 < x < L \) with absorbing walls (\( n = 0 \) at both ends), separation of variables gives
$$n(x, t) = \sum_{k=1}^{\infty}A_k\sin\frac{k\pi x}{L}\,e^{-t/\tau_k}, \qquad \boxed{\tau_k = \frac{L^2}{k^2\pi^2D}}$$Short-wavelength modes die fastest; at long times only \( k = 1 \) survives, and the total amount decays as \( e^{-t/\tau_1} \). Doubling \( L \) quadruples every relaxation time.
Brownian Motion and the Langevin Equation
A particle of mass \( m \) in a fluid feels a systematic drag \( -\gamma v \) and a rapidly fluctuating force \( \xi(t) \) from molecular impacts:
$$\boxed{m\frac{dv}{dt} = -\gamma v + \xi(t), \qquad \langle\xi(t)\rangle = 0, \quad \langle\xi(t)\xi(t')\rangle = 2\gamma k_BT\,\delta(t - t')}$$The noise strength \( 2\gamma k_BT \) is not arbitrary: it is the value that makes the long-time velocity distribution obey equipartition, \( \langle v^2\rangle = k_BT/m \). Friction and fluctuations come from the same molecular collisions, so their sizes are linked. This is the simplest fluctuation–dissipation relation.
Velocity relaxation
With \( \tau_v = m/\gamma \):
$$\boxed{\langle v(0)v(t)\rangle = \frac{k_BT}{m}e^{-\lvert t\rvert/\tau_v}}$$Velocities forget their initial value after a few \( \tau_v \).
Mean square displacement
Integrating twice,
$$\boxed{\langle[x(t) - x(0)]^2\rangle = \frac{2k_BT}{\gamma}\left[t - \tau_v\left(1 - e^{-t/\tau_v}\right)\right]}$$| Regime | Behaviour | Meaning |
|---|---|---|
| \( t \ll \tau_v \) | \( \langle\Delta x^2\rangle \approx \dfrac{k_BT}{m}t^2 \) | ballistic: free flight at thermal speed |
| \( t \gg \tau_v \) | \( \langle\Delta x^2\rangle \approx 2Dt \) | diffusive |
with the Einstein relation
$$\boxed{D = \frac{k_BT}{\gamma} = \mu k_BT}$$where \( \mu = 1/\gamma \) is the mobility (drift velocity per unit force). The same \( D \) follows from the Green–Kubo formula, \( D = \int_0^\infty\langle v(0)v(t)\rangle\,dt = (k_BT/m)\tau_v \).
Stokes–Einstein
For a sphere of radius \( r \) in a fluid of viscosity \( \eta \), Stokes drag gives \( \gamma = 6\pi\eta r \):
$$\boxed{D = \frac{k_BT}{6\pi\eta r}}$$and the rotational analogue \( D_r = k_BT/8\pi\eta r^3 \).
Einstein’s sedimentation argument
Particles of buoyant mass \( m^* \) in gravity drift down at \( v_s = m^*g/\gamma \), giving a downward flux \( v_sn \), while diffusion pushes them up with flux \( -D\,dn/dz \). In equilibrium the total flux vanishes:
$$-D\frac{dn}{dz} - \frac{m^*g}{\gamma}n = 0 \ \Rightarrow\ n(z) \propto \exp\left(-\frac{m^*gz}{\gamma D}\right)$$Equilibrium statistical mechanics demands the barometric form \( \exp(-m^*gz/k_BT) \). The two agree only if \( D = k_BT/\gamma \): Einstein’s relation, derived by demanding consistency between kinetics and equilibrium.
For charged particles in a field the same argument gives the Nernst–Einstein relation \( D = \mu_ek_BT/q \), with \( \mu_e \) the drift velocity per unit electric field.
graph TD
A["Molecular kicks: random force xi(t)"] --> B["Langevin equation: m dv/dt = -gamma v + xi"]
B --> C["Velocity correlation decays in tau = m/gamma"]
C --> D["t << tau: ballistic, MSD = (kT/m) t^2"]
C --> E["t >> tau: diffusive, MSD = 2Dt"]
E --> F["Einstein: D = kT/gamma"]
F --> G["Stokes: gamma = 6 pi eta r, so D = kT/(6 pi eta r)"]
F --> H["Many particles: diffusion equation dn/dt = D laplacian n"]Memory Tricks
Don’t Confuse
| This | vs | That |
|---|---|---|
| \( \langle x^2\rangle = 2Dt \) (one axis) | \( \langle r^2\rangle = 6Dt \) (3D, all axes) | |
| Ballistic: \( \langle\Delta x^2\rangle \propto t^2 \), \( t \ll m/\gamma \) | Diffusive: \( \propto t \), \( t \gg m/\gamma \) | |
| Mobility \( \mu = v/F = 1/\gamma \) | Electrical mobility \( \mu_e = v/E = q/\gamma \) | |
| Fick’s first law: flux \( \propto \) gradient | Fick’s second law: diffusion equation | |
| Velocity relaxation time \( m/\gamma \) | Diffusion time over \( L \): \( L^2/D \) | |
| Biased walk variance \( 4Npqa^2 \) | Symmetric walk \( Na^2 \) (\( p = q = \frac{1}{2} \)) | |
| 1D and 2D walks: recurrent | 3D walk: transient (returns with probability about 0.34) |
Common Mistakes
Practice
All questions below are original practice questions written for this page. They are not from any CSIR NET paper. Use \( k_B = 1.381\times10^{-23} \) J K\( ^{-1} \).
Level 1: Foundation
- What is the rms displacement after \( N \) steps of a symmetric 1D random walk of step length \( a \)?
- In three dimensions, how does \( \langle r^2\rangle \) depend on \( D \) and \( t \)?
- State the Einstein relation between diffusion coefficient and friction coefficient.
Answers
- \( a\sqrt N \).
- \( \langle r^2\rangle = 6Dt \).
- \( D = k_BT/\gamma \), equivalently \( D = \mu k_BT \) with mobility \( \mu = 1/\gamma \).
Level 2: Part B Style
Practice Q1. A walker takes 100 unit steps on a line, each to the right with probability 0.6 and to the left with probability 0.4. The mean displacement and its standard deviation are:
(a) 20 and 9.8 (b) 20 and 10 (c) 60 and 4.9 (d) 0 and 10
Answer: (a) 20 and 9.8
Mean \( N(p - q)a = 100\times0.2 = 20 \). Variance \( 4Npq = 4\times100\times0.24 = 96 \); standard deviation \( 9.80 \). (b) uses the symmetric-walk variance \( N \); (c) gives the mean number of right steps.
Practice Q2. A symmetric 1D random walker starts at the origin. The probability that it is back at the origin after 4 steps is:
(a) 0.0625 (b) 0.25 (c) 0.375 (d) 0.5
Answer: (c) 0.375
Exactly 2 right and 2 left steps: \( \binom{4}{2}/2^4 = 6/16 = 0.375 \).
Practice Q3. In a given liquid at fixed temperature, the radius of a spherical colloidal particle is doubled. The time it takes to diffuse a fixed rms distance:
(a) halves (b) stays the same (c) doubles (d) quadruples
Answer: (c) doubles
\( D \propto 1/r \) (Stokes–Einstein), and \( t = \langle r^2\rangle/2dD \propto 1/D \propto r \). Doubling \( r \) doubles \( t \). (d) would follow if one mistakenly took \( D \propto 1/r^2 \).
Practice Q4. A small molecule has \( D = 1.0\times10^{-9} \) m\( ^2 \) s\( ^{-1} \) in water. The times for its 3D rms displacement to reach 1 μm and 1 cm are about:
(a) 0.17 ms and 4.6 h (b) 0.5 ms and 14 h (c) 0.17 ms and 1.7 s (d) 1 ms and 28 h
Answer: (a) 0.17 ms and 4.6 h
\( t = r^2/6D \). For 1 μm: \( 10^{-12}/6\times10^{-9} = 1.7\times10^{-4} \) s. For 1 cm: \( 10^{-4}/6\times10^{-9} = 1.67\times10^4 \) s \( = 4.6 \) h. (b) uses \( 2Dt \) (one axis); (c) scales linearly instead of quadratically with distance.
Level 3: Part C Style
Practice Q5. Spherical beads in water at 300 K (\( \eta = 1.0\times10^{-3} \) Pa s) are tracked in the focal plane of a microscope. The two-dimensional mean square displacement is:
| \( t \) (s) | 1 | 2 | 4 | 8 |
|---|---|---|---|---|
| \( \langle r^2\rangle \) (μm\( ^2 \)) | 1.76 | 3.52 | 7.03 | 14.06 |
Consider:
- The motion is diffusive over this range.
- The diffusion coefficient is about 0.44 μm\( ^2 \) s\( ^{-1} \).
- The bead radius is about 0.5 μm.
- If the same beads were tracked in a fluid of twice the viscosity at the same temperature, \( \langle r^2\rangle \) at 8 s would be about 28 μm\( ^2 \).
Which are correct?
(a) 1 and 2 only (b) 1, 2 and 3 only (c) 2, 3 and 4 only (d) 1, 2, 3 and 4
Answer: (b) 1, 2 and 3 only
- \( \langle r^2\rangle \) doubles when \( t \) doubles: linear in \( t \), diffusive. True.
- In 2D, \( \langle r^2\rangle = 4Dt \): \( D = 1.76/4 = 0.44 \) μm\( ^2 \) s\( ^{-1} \) \( = 4.4\times10^{-13} \) m\( ^2 \) s\( ^{-1} \). True.
- \( r = \dfrac{k_BT}{6\pi\eta D} = \dfrac{1.381\times10^{-23}\times300}{6\pi\times10^{-3}\times4.4\times10^{-13}} = 5.0\times10^{-7} \) m. True.
- Doubling \( \eta \) halves \( D \): about 7.0 μm\( ^2 \) at 8 s, not 28. False.
Practice Q6. A particle of mass \( m \) obeys \( m\dot v = -\gamma v + \xi(t) \) with \( \langle\xi(t)\xi(t')\rangle = 2\gamma k_BT\delta(t - t') \). Consider:
- \( \langle v(0)v(t)\rangle = (k_BT/m)e^{-\gamma t/m} \).
- For \( t \ll m/\gamma \), \( \langle\Delta x^2\rangle \approx (k_BT/m)t^2 \).
- For \( t \gg m/\gamma \), \( \langle\Delta x^2\rangle \approx 2k_BTt/\gamma \).
- Doubling the mass at fixed \( \gamma \) halves the long-time diffusion coefficient.
Which are correct?
(a) 1 and 3 only (b) 1, 2 and 3 only (c) 2 and 4 only (d) 1, 2, 3 and 4
Answer: (b) 1, 2 and 3 only
- The velocity relaxes with time constant \( m/\gamma \), and \( \langle v^2\rangle = k_BT/m \). True.
- Expanding \( 2D[t - \tau(1 - e^{-t/\tau})] \approx Dt^2/\tau = (k_BT/m)t^2 \). True.
- \( 2Dt \) with \( D = k_BT/\gamma \). True.
- \( D = k_BT/\gamma \) has no mass; mass changes only the crossover time. False.
Practice Q7. A solute diffuses (\( D = 1.0\times10^{-9} \) m\( ^2 \) s\( ^{-1} \)) out of a slab of thickness 1.0 mm whose faces are held at zero concentration. Consider:
- At long times the total amount in the slab decays exponentially with time constant about 100 s.
- For a slab 2.0 mm thick, that time constant becomes about 400 s.
- The concentration mode \( \sin(2\pi x/L) \) decays four times faster than the slowest mode.
- Doubling \( D \) doubles the time constant.
Which are correct?
(a) 1 and 2 only (b) 1, 2 and 3 only (c) 3 and 4 only (d) 1, 2, 3 and 4
Answer: (b) 1, 2 and 3 only
- \( \tau_1 = L^2/\pi^2D = 10^{-6}/(9.87\times10^{-9}) = 101 \) s. True.
- \( \tau \propto L^2 \): \( 4\times101 \approx 405 \) s. True.
- \( \tau_k \propto 1/k^2 \): \( k = 2 \) decays 4 times faster. True.
- \( \tau \propto 1/D \): it halves. False.
Quick Revision Table
| Idea | Formula or fact |
|---|---|
| Biased walk | \( \langle x\rangle = N(p - q)a \), \( \sigma^2 = 4Npqa^2 \) |
| Symmetric walk | \( x_{\text{rms}} = a\sqrt N \); \( P_0(2k) = \binom{2k}{k}/4^k \) |
| Any dimension | \( \langle R^2\rangle = Na^2 \); polymer \( R = b\sqrt N \) |
| Recurrence | 1D, 2D: certain return; 3D: about 0.34 |
| Continuum limit | \( D = a^2/2\tau \) |
| Fick | \( \mathbf J = -D\nabla n \); \( \partial_tn = D\nabla^2n \) |
| Point source | \( (4\pi Dt)^{-d/2}e^{-r^2/4Dt} \); \( \langle r^2\rangle = 2dDt \) |
| Slab modes | \( \tau_k = L^2/k^2\pi^2D \) |
| Langevin noise | \( \langle\xi\xi\rangle = 2\gamma k_BT\delta(t - t') \) |
| Velocity correlation | \( (k_BT/m)e^{-t\gamma/m} \) |
| MSD | ballistic \( (k_BT/m)t^2 \); diffusive \( 2Dt \) |
| Einstein | \( D = k_BT/\gamma = \mu k_BT \); ions \( D = \mu_ek_BT/q \) |
| Stokes–Einstein | \( D = k_BT/6\pi\eta r \); rotation \( k_BT/8\pi\eta r^3 \) |
| Green–Kubo | \( D = \int_0^\infty\langle v(0)v(t)\rangle dt \) |
Key Takeaways
What’s Next
- Non-Equilibrium Processes: transport laws, Onsager relations, the Boltzmann equation and the Fokker–Planck description of the same Brownian particle.
- Bose–Einstein Condensation: the last equilibrium topic of this chapter.
- Kinetic-theory transport (mean free path, viscosity, diffusion of gases): Kinetic Theory of Gases.
- The same equation as heat flow: Partial Differential Equations: Laplace, Wave and Heat.