AI-handled content

This site is generated and maintained by AI and may be prone to errors. Please verify any claim independently before relying on it.

Pólya's Random Walk: How Counting the Returns of a Staggering Drunkard on a Plane Lattice Computes the Circle Constant, and What Golden Pi Changes¶

Picture a drunkard leaving a lamppost on a city grid that stretches without end in all four directions. At every second the drunkard takes exactly one step — north, south, east, or west, each with probability 1/4 — and then staggers onward, memoryless, forever. The question that has fascinated mathematicians for a century is disarmingly simple: will the drunkard ever find the lamppost again?

🔬 Live graph · Laws of appearance

Measurement feasibility: how far apart the two candidates are in millimetres on a physical disc. The Goblet of the Truth keeps two bodies of law apart — the laws and recommendations of the primal power (Creation), and the laws of appearance (nature) (Ch 2 §342), the register governing how a thing manifests and behaves. Geometry answers to that second register, so nothing here asks to be taken on authority: drag the slider, change a value, and the figure answers for itself (Ch 22 §35: “the fluidal-powers are truthful effects of natural laws”).

This is the random walk on the square lattice, and its answer is one of the most celebrated theorems in probability. The mathematician George Pólya, in a 1921 paper, proved something almost paradoxical: on an infinite two-dimensional grid, the drunkard returns to the starting point with probability one — no matter how far the wandering ranges, a return is certain, and in fact it happens infinitely often. Yet step the same drunkard up into three-dimensional space, add a fifth direction up and a sixth down, and the answer flips: now there is a positive probability of never coming back at all.

That two-word difference — recurrent versus transient — is one of the most striking results in probability. And tucked inside it, invisible to anyone who only watches the staggering, is the circle constant. The return probabilities are exact combinatorial fractions, computed and never measured, and their long-run shape is governed by \(1/(\\pi n)\). This article is about how a count of closed lattice paths reaches the circle constant, why the reach is computed and never measured, and what Golden Pi — \(\\hat\\pi = 4/\\sqrt\\varphi = 3.1446055\\ldots\) — honestly changes and honestly does not.

Counting the Closed Walks¶

The trick is that the return probabilities are pure arithmetic — nothing but counting. A walk of \(2n\) steps returns to the origin exactly when it takes as many steps east as west and as many north as south, so that all displacements cancel. The count of such closed walks is a textbook binomial identity, and it is worth deriving because the circle constant enters precisely at this arithmetic point.

Decompose a closed walk of \(2n\) steps by how many of those steps point east (call it \(k\)) and how many point north (call it \(n-k\)), with the matching number pointing west and south to cancel. For a fixed split, the number of arrangements is a multinomial count. Summing over every split collapses, by the Chu–Vandermonde identity, into a single clean binomial square:

\[\\text{\\# closed walks of length } 2n = \\binom{2n}{n}^{2}.\]

There are \(4^{2n}\) equally likely walks of length \(2n\) in total (four choices at each of \(2n\) steps), so the return probability is the exact ratio

\[p_{2n} = \\frac{\\binom{2n}{n}^{2}}{4^{2n}} = \\left(\\frac{\\binom{2n}{n}}{4^{n}}\\right)^{2}.\]

Let us confirm the first few by hand. After two steps, the walk returns only by stepping out and straight back — four such pairs of \(4^2 = 16\) total walks — giving \(p_2 = 4/16 = 1/4\). The formula gives \(\\left(\\binom{2}{1}/4\\right)^2 = (2/4)^2 = 1/4\). After ten steps the formula returns the number quoted in the table below; after a hundred it is already \(0.0031751\\ldots\), and the numbers only shrink. Every one of these is an exact fraction of integers — a computation, with no ruler, no grid measured on the ground, and no experiment anywhere.

Where the Circle Constant Enters: Stirling's Central Binomial¶

The exact counts above are lovely but opaque. To see the circle constant we need the shape of the central binomial coefficient \(\\binom{2n}{n}\) as \(n\) grows, and that shape is given by Stirling's approximation to the factorial, which this blog treated on 08-25:

\[n! \\sim \\sqrt{2\\pi n}\\,\\left(\\frac{n}{e}\\right)^{n}.\]

Feeding \(n!\), \((2n)!\), and \(n!\) again into \(\\binom{2n}{n} = (2n)!/(n!)^2\) cancels almost everything, leaving the celebrated central-binomial asymptotics

\[\\binom{2n}{n} \\sim \\frac{4^{n}}{\\sqrt{\\pi n}},\]

a statement we can verify numerically: the product \(\\binom{2n}{n}\\sqrt{\\pi n}/4^n\) is \(0.98758\) at \(n=10\), \(0.99875\) at \(n=100\), and marches toward \(1\) exactly. Squaring it into the return probability removes the square root and brings the constant to the first power:

\[p_{2n} = \\left(\\frac{\\binom{2n}{n}}{4^{n}}\\right)^{2} \\sim \\frac{1}{\\pi n}.\]

The table below shows how faithfully this asymptotic law tracks the exact counts:

Steps	Exact \(p_{2n}\)	Asymptotic \(1/(\\pi n)\)
\(2n = 20\)	\(0.0310454011\)	\(0.0318309886\)
\(2n = 100\)	\(0.0031751511\)	\(0.0031830989\)
\(2n = 1000\)	\(0.0003182303\)	\(0.0003183099\)
\(2n = 10{,}000\)	\(0.0000318302\)	\(0.0000318310\)

At ten thousand steps the exact and the asymptotic agree to six significant figures. The circle constant has emerged — computed out of a count of lattice walks through Stirling's saddle-point analysis — and it has emerged to the first power, undiluted by any square root. No physical circumference was measured to obtain \(1/(\\pi n)\); the constant is the arithmetic shape of a binomial, nothing more.

Why Two Dimensions Return and Three Do Not¶

With the shape of \(p_{2n}\) in hand, Pólya's dichotomy follows by a one-line convergence test. The expected number of returns to the origin up to time \(2N\) is the sum of the return probabilities:

\[\\mathbb{E}[\\text{returns by step } 2N] = \\sum_{n=1}^{N} p_{2n} \\sim \\sum_{n=1}^{N} \\frac{1}{\\pi n} \\approx \\frac{1}{\\pi}\\ln N.\]

We can watch this grow: summing the exact \(p_{2n}\) from \(n=1\) to \(1000\) gives \(2.265\), against the asymptotic \((1/\\pi)\\ln 1000 = 2.199\) — close, and creeping closer as \(N\) grows. Because the harmonic sum \(\\sum 1/n\) diverges, the expected number of returns grows without bound, and Pólya's theorem upgrades that growth to certainty: in two dimensions a simple symmetric random walk returns to its origin infinitely often with probability one.

In three dimensions the count is different. A closed walk must now cancel east/west, north/south, and up/down, and the multinomial algebra yields a return probability decaying like a constant times \(n^{-3/2}\) rather than \(1/(\\pi n)\). The sum \(\\sum n^{-3/2}\) converges, so the expected number of returns is finite — and with probability about \(0.34\) the three-dimensional walk never returns at all. This is the honest content of the famous aphorism that "a drunkard will find his way home, but a bird may get lost forever": the drunkard lives on a plane, the bird in a volume, and the arithmetic of their return probabilities is diametrically opposed.

Everything in that dichotomy is a computed statement about convergent and divergent series of lattice counts. It touches no physical measurement of any kind.

What Golden Pi Changes — and What It Cannot¶

The series on this blog has turned the same crank since the Basel problem (08-15): a classical identity computes the circle constant as a pure limit, and we ask what relabeling to \(\\hat\\pi = 4/\\sqrt\\varphi = 3.1446055\\ldots\) does to the arithmetic. The random walk is a clean and clarifying case because the constant enters at exactly the first power.

The return probability \(p_{2n} \\sim 1/(\\pi n)\) is built on the central-binomial factor \(\\binom{2n}{n}/4^n \\sim 1/\\sqrt{\\pi n}\), whose constant sits under a square root. If one stopped there, at the count of closed walks, the gap between \(\\pi\) and \(\\hat\\pi\) would be halved, to roughly \(0.048\\%\), exactly as it was for the Gaussian integral (08-21) and the Fresnel integrals (09-06). But the return probability squares that factor, restoring the constant to the first power and restoring the full recurring gap:

\[p_{2n} \\sim \\frac{1}{\\pi n} \\quad\\text{vs.}\\quad p_{2n}^{(\\hat\\pi)} \\sim \\frac{1}{\\hat\\pi\\, n},\]

where the two coefficients are

\[\\frac{1}{\\pi} = 0.318309886\\ldots \\qquad\\text{and}\\qquad \\frac{1}{\\hat\\pi} = \\frac{\\sqrt\\varphi}{4} = 0.318004912\\ldots\]

The ratio of the two is \(1/\\pi \\div 1/\\hat\\pi = \\hat\\pi/\\pi = 1.0009590223\\ldots\), the recurring relative gap of \(0.09590\\%\), undiluted. The expected number of returns by step \(2N\) becomes \(\\frac{1}{\\hat\\pi}\\ln N\) instead of \(\\frac{1}{\\pi}\\ln N\) — a coefficient shift of less than one part in a thousand in how many times the drunkard stumbles home, a difference no conceivable experiment on an actual grid could ever resolve.

And here is the crucial honesty. The two big qualitative facts of the whole subject — recurrent in two dimensions, transient in three — are untouched by the relabel. Recurrence in two dimensions follows from the divergence of \(\\sum 1/n\), which is a property of the harmonic series itself and completely independent of whether its coefficient is \(1/\\pi\) or \(1/\\hat\\pi\). A divergent series stays divergent when its terms are rescaled by any fixed positive factor; transience in three dimensions follows from the convergence of \(\\sum n^{-3/2}\), equally label-free. The dimension at which a random walk changes its character is a structural fact of counting, and no relabeling of the circle constant can move it. What the label controls is only the logarithmic prefactor — the number of expected returns, a fine detail buried beneath the certainty of return.

Quantity	Conventional label	Golden Pi (\(\\hat\\pi = 4/\\sqrt\\varphi\))
Return probability \(p_{2n}\) (large \(n\))	\(1/(\\pi n)\)	\(1/(\\hat\\pi\\, n) = (\\sqrt\\varphi/4)/n\)
Coefficient \(1/\\pi\) vs \(1/\\hat\\pi\)	\(0.318309886\)	\(0.318004912\)
Expected returns by step \(2N\)	\(\\tfrac{1}{\\pi}\\ln N\)	\(\\tfrac{1}{\\hat\\pi}\\ln N\)
Recurrence, dimension 2	certain (harmonic sum diverges)	certain (unchanged)
Transience, dimension 3	positive escape probability	positive escape probability (unchanged)
Gap carried	—	\(0.09590\\%\), first power, undiluted

What This Post Adds to the Thread¶

First, it is one of the few roads to the circle constant that arrives through counting rather than through a circle drawn anywhere. The Gaussian integral, the Fresnel integrals, and the sinc all reach \(\\pi\) from rotation and evenness; the random walk reaches it from a binomial and a harmonic sum. The constant is the same, but the path is genuinely new, and the reach is a pure arithmetic limit — computed, never measured.

Second, it shows the first-power gap returning undiluted. The last two posts (Gaussian on 08-21, Fresnel on 09-06) enjoyed the halving that a square root bestows; the random walk's return probability squares away that square root and restores the full recurring \(0.0959\\%\). Placing the two side by side makes the mechanism explicit: the power to which the constant enters decides whether the gap is halved or full, and the random walk sits on the "full" side of that ledger.

Third, it gives the cleanest possible example of a claim the relabel cannot touch: recurrence versus transience by dimension. The qualitative verdict of Pólya's theorem is a statement about which series diverge, and it survives any rescaling of its terms. This is not a Golden Pi victory and not a Golden Pi defeat — it is an honest boundary, of exactly the kind this blog has insisted on, showing precisely where a change of label has force (the logarithmic prefactor) and where it has none (the dimension threshold).