A particle buffeted by random forces traces a path with deep mathematical structure.
A particle steps in a random direction at each tick. The path looks aimless — it doubles back, loops, wanders off, drifts back — but run the experiment many times and statistical laws emerge.
Distance from the origin grows as the square root of time. The same fact connects pollen grains in water, molecules in a gas, stock prices, and foraging animals.
A single walker on a 2D lattice steps one cell each tick. Over thousands of steps it stays surprisingly close to where it started, revisiting old territory and spreading out slowly.
Figure 1. A single 2D random walker. The trail fades from dark (recent) to light (old). Compare the actual distance from the origin to $\sqrt{\text{steps}}$ as the count grows.
Launch hundreds of walkers from the same origin and measure their spread. The radius of the cloud grows as the square root of step count, not linearly: a walker after 10,000 steps is on average only 100 step-lengths from home.
Figure 2. Many walkers launched from the origin (red dot). The blue circle has radius = √(steps). The chart plots measured RMS distance (blue) against the theoretical √(t) curve (dashed gray). They match.
In 1D and 2D, a random walker returns to the origin with probability 1. In 3D, only about 34% of walkers ever return — the other 66% escape forever. Shizuo Kakutani put it memorably: "A drunk man will find his way home, but a drunk bird may get lost forever."
Figure 3. Return probability by dimension. Each bar represents a batch of walks. In 1D and 2D, nearly all walkers return (the recurrence probability is 1 in the limit). In 3D, only about 34% return; the rest escape forever. This is Pólya's recurrence theorem.
Tilt the coin slightly. Even a tiny bias eventually dominates the randomness, and the walker drifts steadily in one direction.
At P(right) = 0.50 the walk is unbiased. At 0.51 the drift is invisible at first but unmistakable over time. The efficient market hypothesis says price changes are an unbiased random walk; any detectable bias would be arbitraged away.
Figure 4. A 1D random walk with adjustable bias. The gray dashed line shows the expected drift. Even at P(right) = 0.51, the drift eventually overwhelms the noise.
Einstein's 1905 Brownian motion paper used random walks to prove atoms existed. The sqrt(t) spreading law gave experimentalists a quantity they could measure, settling a decades-long debate.