Topic · Analysis

Buffon's Needle

Drop sticks on lined paper. π falls out.

Georges-Louis Leclerc, Comte de Buffon, asked the question in 1733 and published it in 1777: drop a needle on a floor of parallel lines and count the crossings. The ratio gives back π — a constant from circles emerging out of straight needles on straight wood. Here we walk the proof, the history and the experiment side by side.

Discover the story → Open the Explorer

First encounter

A needle. A lined page. A constant from circles.

The big idea

Rule a sheet of paper with parallel lines, all the same distance d apart. Take a needle of length ℓ ≤ d and drop it from above so it lands at a random spot and angle. The chance the needle lies across one of the lines is exactly 2ℓ / (π d). Drop a great many needles, count the crossings, and rearrange — π comes out of straight sticks, with no circle in sight.

A concrete example

Take ℓ = d. Drop a thousand needles. Roughly 2 / π ≈ 0.6366 of them should cross a line — about 636 of the thousand. The estimator gives back π ≈ 2 · 1000 / 636 ≈ 3.145. Below, six needles fall on four parallel lines; the amber ones cross.

Why it matters

This is the first known Monte Carlo method — long before the name was coined at Los Alamos in 1947. Buffon's needle bridges geometry and probability: it shows that a deterministic constant can be sampled with randomness alone. Modern proofs of probability lean on the same machinery — replace a hard integral with an expectation, sample the expectation, take the limit.

Below: drop a few thousand virtual needles and watch π emerge.

Section 01 · The setup

Parallel lines, distance d, needle of length ℓ ≤ d

Rule a floor — or a sheet of paper — with parallel lines, all spaced a constant distance d apart. Take a needle of length ℓ with ℓ ≤ d, and drop it from above. Two random variables describe the landing: the vertical position y of the needle's centre, uniform between two consecutive lines, and the angle θ between the needle and the lines, uniform on [0, π / 2] by symmetry. The needle lies across a line if its half-length projected vertically — that is, (ℓ / 2) · sin θ — exceeds the distance from its centre to the nearer line. Two parameters, one yes/no question, repeated a great many times. That is the entire experiment.

Section 02 · The probability derivation

Why π appears from a sine and an integral

Fix the angle θ. The needle crosses a line iff y < (ℓ / 2) · sin θ. Since y is uniform on [0, d / 2], that conditional probability is (ℓ / d) · sin θ. Average over θ uniform on [0, π / 2]: P(cross) = ∫₀^{π/2} (ℓ / d) · sin θ · (2 / π) dθ = (2 ℓ) / (π d). The 2 / π factor comes from the uniform density on the angle, and the integral of sin θ from 0 to π / 2 equals 1. There it is — π sitting in the denominator, smuggled in by a sine. The constant of the circle has appeared in a problem about straight lines on straight paper.

The exact probability

P(cross) = 2 ℓ / (π d)

A sine, an integral, a uniform density: π emerges from straight lines.

Section 03 · From P to π

Invert the formula: π ≈ 2 ℓ N / (d · C)

Drop N needles. Count the C that cross a line. The law of large numbers tells us C / N converges, almost surely, to the true probability P = 2 ℓ / (π d). Solve the equation for π: π ≈ 2 ℓ N / (d · C). This is the Buffon estimator. With ℓ = d and N = 1000 we expect C ≈ 636 and π ≈ 3.145. With N = 10⁶ and ℓ = d we expect about three correct digits. The convergence is slow — error decays as 1 / √N, as for every Monte Carlo method — but it is genuine, and every additional needle nudges the estimate honestly toward 3.14159…

Interactive · live needle drops

Drops N0

Crossings C0

Estimate of π

—

True · 3.14159 · Δ —

··

Section 04 · The Monte Carlo method

Sampling an integral with randomness

Buffon's needle is the prototype of what Stanisław Ulam and John von Neumann formalised at Los Alamos in 1947 and named Monte Carlo, after the casino: estimate a hard integral by sampling it with random points. The trick is the law of large numbers — averages of independent samples converge to their expectation. Today Monte Carlo runs the cores of modern computational physics (lattice QCD), financial derivatives pricing (Black–Scholes with stochastic volatility), photorealistic rendering (path tracing in ray-tracers, where every pixel is a Buffon-style sample of a light integral), Bayesian inference (Markov-chain Monte Carlo), and parts of modern machine-learning training where gradients are estimated stochastically. All of it descends from a count of crossings.

Interactive · estimate vs sample size

π · 3.14159

estimate

log N

Total drops N1,000

Final estimate—

Section 05 · Lazzarini's 1901 'experiment'

Six decimal places from 3408 drops — too good to be true

In 1901 the Italian mathematician Mario Lazzarini published an experiment in which he claimed to have computed π ≈ 3.1415929 from just 3408 needle drops — six correct digits, agreeing exactly with the famous Chinese-era approximation 355 / 113. That accuracy from a Buffon trial would require something like 10¹⁴ drops by honest convergence. The trick: Lazzarini's design used N = 213 · 16 = 3408 drops with a needle-to-line ratio ℓ/d = 5/6, chosen so that, the moment the running count happened to hit C = 1808 crossings, the estimator π ≈ 2·(ℓ/d)·N/C = 2·(5/6)·(3408/1808) = 355/113 landed on the famous fraction by construction. He almost certainly stopped at the right moment. The episode is famous as the cleanest cautionary tale in experimental statistics — and as evidence that Buffon's needle, honestly conducted, converges painfully slowly. Pierre-Simon Laplace had already extended the problem in 1812 to a square grid of lines, where two crossings (horizontal and vertical) can be counted independently and the variance is halved; Laplace's variant is the one a serious experimenter would actually run.

Lazzarini, 1901 · honest convergence

N drops	typical \|Δπ\|
100	~0.1
1 000	~0.03
10 000	~0.01
100 000	~0.003
1 000 000	~0.001
3 408 (Lazzarini)	claimed 0.0000003

The last row is not a fact about randomness: it is a fact about a man stopping at exactly the right needle.

Section 06 · Other geometric probabilities

Bertrand's paradox and the trap of 'random'

Buffon's needle works because the words 'uniform position, uniform angle' fix the sample space precisely. Other geometric-probability problems are not so kind. Joseph Bertrand published a paradox in 1889: what is the probability that a random chord of a unit circle is longer than the side of the inscribed equilateral triangle? Three natural procedures — pick two random endpoints on the circle, pick a random midpoint inside the disk, pick a random distance from the centre — give three different answers (1/3, 1/4, 1/2). All three are correct given their own definition of 'random chord'; the paradox is that the phrase, without a procedure, is ambiguous. Edwin Jaynes argued in 1973 for the 1/2 answer on grounds of maximum-entropy invariance, but the deeper lesson stands: geometric probability is only well-defined once the sampling procedure is fully specified. Buffon's needle is famous partly because that specification is so transparent — and the answer, 2 ℓ / (π d), does not depend on which natural procedure you pick.

Bertrand 1889 · three answers to one question

procedure	P(longer than side)
two random endpoints on the circle	1 / 3
random midpoint uniform in the disk	1 / 4
random distance from centre	1 / 2

Same words, three sample spaces, three answers. Buffon's needle is famous in part because its sampling procedure is so unambiguous, and the answer the same.

Take it further

Drop the needles.

The Explorer hands you the parameters: needle length, line spacing, drop rate. Watch the running estimate of π crawl toward the true value, watch the 1 / √N envelope tighten, and feel for yourself how slow Monte Carlo really is — and how reliable.

→ Open the Explorer

Where you meet it

How and where this technique lives in the world today.

Monte Carlo integration

Buffon's needle is the historical seed of Monte Carlo methods — now used everywhere from financial pricing (Black–Scholes) to particle-physics simulation (Geant4) and rendering (path tracing).

Physics simulation

Random sampling of high-dimensional integrals in lattice QCD, nuclear engineering and reactor design extends Buffon's logic to millions of dimensions.

Computer graphics

Stratified sampling of light rays in modern path tracers (Pixar, Cycles, Unreal Lumen) carefully descends from Buffon's needle.

Statistics teaching

Standard introductory probability demonstration; still performed in undergraduate stats labs around the world.