Uncertainty & dynamics

The drunk in the middle of the road

Stand on a number line at 0. Flip a coin. Heads step right, tails step left. Repeat forever.
Half the deepest lessons about randomness live inside this game.

One walker · watch the coin-flipper

Coin flips

Position

Farthest reached

Watch it long enough and you'll notice something: you drift away, but slowly. After 100 flips you're typically about 10 away. After 10,000 flips you're typically about 100 away. Distance grows as the square root of time. Not linear. Not exponential. Slow.

Now let's race a hundred of them

Below, a hundred walkers all start at the middle line and flip their own coins. The shaded band is ±√t — roughly where a typical walker is at time t. Watch the swarm respect it.

100 walkers & the √t envelope

Time step

Empirical std

theory √t = 0

The "trend" illusion. Some of those walkers spend the whole run above the line — they'll look like they've got momentum, a streak, "hot hands." They don't. They're coin flips. Every "trend" you can imagine, this process produces by accident. Now imagine you were shown one of them and asked to explain it. You'd invent a story.

The return-time paradox

Here's a fact so counter-intuitive it feels wrong. A one-dimensional walker returns to zero with probability 1. Every single walk comes home, eventually. But the average time to come home is infinite. Both true. Let's see.

First-return times · click Simulate

Median return

—

typical walk

99th percentile

—

the slow tail

Empirical mean

—

theory: ∞

Look at what dominates: half the walks come home in exactly two steps. The 99th percentile is thousands of steps out. Both facts describe the same distribution — one you cannot summarise with an "average." The mean is undefined; only the median is a fair number.

Gambler's ruin. A gambler on a fair bet is guaranteed to break even — that is a real theorem. But he has finite money and a finite life. Between now and eventually, the walk visits every negative number too. Against an infinitely-rich casino, he goes broke with probability 1. "It'll come back eventually" is mathematically true and financially useless.

Drunk man vs drunk bird

Now do the same coin-flip game in higher dimensions. Each step, pick an axis at random, move ±1 along it. Does the walker still come home?

Pólya (1921) proved: in 1-D and 2-D, yes, with probability 1. In 3-D and above, no. Kakutani summarised it: "A drunk man will find his way home, but a drunk bird may get lost forever."

Pólya's theorem in miniature

Bars = fraction of 500 walks (5,000 steps each) that returned to the origin. Diamonds = theoretical eventual-return probability. 2-D is technically 100% but climbs excruciatingly slowly (1/log t), so a finite budget under-counts it.

Bonus: the arcsine surprise

Ask: in a fair coin-flip game between two players, what fraction of the time is one player "in the lead"? Intuition says ½. Probability theory answers: the extremes are massively more likely than the middle. In a truly balanced game, one side tends to lead almost the whole time.

Fair walks look one-sided

x-axis: fraction of time each walker was above zero. If leading were 50/50 by intuition, this should peak in the middle. It doesn't. It peaks at the extremes.

Reader lens for the news. "One team dominated the whole game" is exactly what a fair, balanced contest looks like most of the time. Same for mutual funds, sales streaks and corporate winning streaks. A random walk generates dominance narratives without any underlying dominance.

Related models

🌐 Markov chains — a random walk is a Markov chain. The √t spread and the stationary-distribution story share one skeleton.
🔔 Normal / CLT — the walker's position becomes normally distributed around the start — that's the CLT.
🏺 Path dependence — the walk's history matters; Pólya urns are the related process where early luck locks in late outcomes.

← SIR epidemic Markov chains →