The Drunkard's Detour

A drunk man stands at a lamppost. Each second, he takes one step — left or right, equally likely. Will he ever return to the lamppost?

In one dimension, the answer is yes. With probability 1. Not "almost certainly" in the colloquial sense — with mathematical certainty, probability exactly 1. He will return. It might take ten steps or ten trillion, but he will return.

Now put him on a grid — a city intersection. Each step goes north, south, east, or west with equal probability. Will he return to where he started?

Still yes. Probability 1. The two-dimensional drunk always comes home.

Now put him in a three-dimensional lattice — a bird in space, hopping randomly along the axes. Will it return?

No. The probability of return drops to approximately 0.3405. The bird has a two-thirds chance of wandering into infinity, never to return.

This is Pólya's recurrence theorem (1921), and it is one of the strangest dimensional transitions in mathematics.

The One-Dimensional Case

Let's start with why the 1D drunk always returns. The probability of being at position 0 after 2n steps is

P₂ₙ = C(2n, n) / 4ⁿ

(He must take exactly n steps left and n right. There are C(2n, n) ways to arrange them, out of 2²ⁿ = 4ⁿ total paths.)

By Stirling's approximation, C(2n, n) ≈ 4ⁿ / √(πn), so

P₂ₙ ≈ 1/√(πn)

This goes to zero — the drunk is less and less likely to be at the origin at any specific time. But the expected number of returns is

∑_{n=1}^{∞} P₂ₙ ≈ ∑ 1/√(πn) = ∞

The sum diverges. The expected number of returns is infinite. And for a random walk, infinite expected returns is equivalent to certain return. (This is a beautiful fact from renewal theory: the number of returns is either finite with probability 1, or infinite with probability 1. There's no middle ground.)

The Two-Dimensional Case

In 2D, the probability of being at the origin after 2n steps is

P₂ₙ⁽²⁾ = [C(2n, n) / 4ⁿ]² ≈ 1/(πn)

Each coordinate performs an independent 1D random walk (with a twist — the steps are correlated because each step moves in exactly one coordinate). The analysis works out to the square of the 1D probability.

And ∑ 1/(πn) still diverges (harmonic series!), so the 2D walk is recurrent too.

The drunk on a grid will always find his way home.

The Three-Dimensional Collapse

In 3D, the return probability becomes

P₂ₙ⁽³⁾ ≈ (3/(2πn))^{3/2}

And ∑ n^{-3/2} converges.

That's the entire story. The difference between dimensions 2 and 3 is the difference between ∑ 1/n (diverges) and ∑ 1/n^{3/2} (converges). The drunk comes home because the harmonic series diverges. The bird doesn't because replacing 1/n with 1/n^{3/2} is enough to make the sum finite.

The expected number of returns in 3D is

∑_{n=1}^{∞} P₂ₙ⁽³⁾ ≈ 1.5164…

The 3D walker expects to return about 1.5 times, then wander off forever. The probability of ever returning works out to approximately 0.3405 — a constant sometimes called Pólya's random walk constant.

Its exact value involves a fearsome integral:

p₃ = 1 − 1/u₃

where u₃ = ∫₀^∞ ∫₀^∞ ∫₀^∞ [1/(1 − cos(x)cos(y)cos(z)/3)] dx dy dz × (1/(2π))³

This integral was computed numerically by McCrea and Whipple (1940) and has been refined to extraordinary precision since. There's no known closed form.

Verification

import random

def random_walk_return(dim, max_steps=100000, trials=50000):
    returns = 0
    for _ in range(trials):
        pos = [0] * dim
        for step in range(1, max_steps + 1):
            axis = random.randint(0, dim - 1)
            pos[axis] += random.choice([-1, 1])
            if all(p == 0 for p in pos):
                returns += 1
                break
    return returns / trials

for d in [1, 2, 3]:
    p = random_walk_return(d, max_steps=10000, trials=20000)
    print(f"Dimension {d}: P(return within 10000 steps) ≈ {p:.4f}")

Typical output:

The 1D and 2D numbers are slightly below 1 only because we cut off at 10,000 steps. Given enough time, both converge to certainty. The 3D number converges to Pólya's constant — with 10,000 steps, we've already captured essentially all of the return probability.

The Dimensional Phase Transition

This is a phase transition, as sharp as water freezing. In dimensions 1 and 2, the random walk is recurrent — it returns everywhere infinitely often. In dimensions 3, 4, 5, …, it's transient — it escapes to infinity.

Why the transition at dimension 3? Because the volume of a d-dimensional sphere of radius r grows as r^d, while the walker's distance grows as √n. The probability of being near the origin after n steps is proportional to n^{-d/2}. The sum ∑ n^{-d/2} converges if and only if d/2 > 1, i.e., d > 2.

Dimension 2 is the critical case: the walk barely returns. The probability of returning by time n grows like log(n) — slowly, grudgingly, but without bound. It's the mathematical equivalent of "technically yes, but don't hold your breath."

The Punchline

There's a famous quip, attributed to Shizuo Kakutani:

"A drunk man will find his way home, but a drunk bird will not."

It's funny because it's true, and it's true for a reason that has nothing to do with drunkenness, men, or birds. It's about the geometry of space itself. Two-dimensional space is too small for a random walker to get lost in. Three-dimensional space isn't.

And the dividing line is a single exponent in a single sum. Everything else — the topology, the lattice structure, the step distribution — is noise. The only thing that matters is whether d > 2.

This is what I love about the "Undiscovered Country." Not the grand theorems but the moments where a kindergarten question — "if I wander randomly, will I come back?" — has an answer that depends on the dimension of space in a way that feels almost metaphysical. The universe has opinions about the number 2, and it will share them with anyone patient enough to take a random walk.

This is Essay #5 in the "Undiscovered Country" series — dispatches from the frontier of combinatorics and number theory. Essay #1: The Ghost of e⁻² | Essay #2: The Loop | Essay #3: The Divisor Surprise | Essay #4: The Prime Conspiracy