Buffon's Needle

In 1777 the French naturalist Georges-Louis Leclerc, Comte de Buffon, posed a deceptively simple question: if you drop a needle of length L onto a floor ruled with parallel lines spaced d apart (where L ≤ d), what is the probability that the needle crosses one of the lines?

The answer he derived is exact: $P = \frac{2L}{\pi d}$. $\pi$ — the ratio of a circle's circumference to its diameter — appears not because there is a circle in sight, but because integrating over all possible angles sweeps out a half-turn of rotation, and that integral evaluates to $2/\pi$.

Turn the formula around and you get something remarkable: if you actually perform the experiment and count how many needles cross, you can estimate $\pi$ from the ratio

$$\pi \approx \frac{2L \cdot (\text{needles dropped})}{d \cdot (\text{crossings})}$$

No circle required — just a floor, a stick, and patience.

This is one of the first recorded Monte Carlo methods: using random physical trials to estimate a mathematical constant. It predates electronic computers by nearly two centuries, yet it is the same idea that now powers everything from physics simulations to financial risk models.

Drop virtual needles and watch the π estimate converge in real time. Each needle lands at a random position and angle; those that cross a line are highlighted in red.

<div class="controls">
  <label>{{speed_label}}
    <select id="speed">
      <option value="1">{{opt_1}}</option>
      <option value="10" selected>{{opt_10}}</option>
      <option value="100">{{opt_100}}</option>
      <option value="1000">{{opt_1000}}</option>
    </select>
  </label>
  <button id="drop" type="button">{{btn_drop}}</button>
  <button id="reset" type="button" class="ghost">{{btn_reset}}</button>
</div>
<canvas id="canvas" width="460" height="200"></canvas>
<div class="stats" id="stats">{{press_start}}</div>

* { box-sizing: border-box; }
body { font-family: system-ui, sans-serif; color: #222; margin: 0; }
.controls { display: flex; align-items: center; gap: .6rem; flex-wrap: wrap; margin-bottom: .6rem; }
label { font-size: .88rem; color: #444; }
select { font-size: .88rem; padding: .25rem .4rem; border: 1px solid #adb1b8; border-radius: 6px; background: #f5f7f9; }
button { font: 600 14px system-ui, sans-serif; padding: .4rem .85rem;
         border: 1px solid #1d3557; background: #1d3557; color: #fff; border-radius: 8px; cursor: pointer; }
button.ghost { background: #fff; color: #1d3557; }
canvas { display: block; border: 1px solid #cdd9e3; border-radius: 8px; background: #f5f8fb; width: 100%; max-width: 460px; }
.stats { margin-top: .55rem; font-size: .92rem; line-height: 1.6; }
.stats strong { color: #1d3557; }
.pi-est { font-size: 1.15rem; font-weight: 700; color: #0a7d33; }
.pi-est.poor { color: #c92f3c; }

// Code not found

Notice how noisy the estimate is at first — after just 20 needles you might see π ≈ 2.4 or π ≈ 4.1. The law of large numbers guarantees convergence, but convergence is slow: to gain one extra decimal digit of accuracy you need roughly 100× more needles. That inefficiency is a fundamental property of Monte Carlo estimation and explains why modern methods combine random sampling with clever variance-reduction tricks.

The derivation is a beautiful two-line integral. A needle of length $L$ dropped at a random angle $\theta$ (uniform in $[0, \pi]$) has its centre at a random distance $x$ from the nearest line (uniform in $[0, d/2]$). It crosses a line exactly when $x \le \frac{L}{2}\sin\theta$.

The crossing probability is therefore:

$$P = \frac{1}{\pi} \int_0^{\pi} \frac{L \sin\theta}{d}\, d\theta = \frac{L}{d} \cdot \frac{2}{\pi} = \frac{2L}{\pi d}$$

The integral of sin over [0, π] is exactly 2 — that is where π enters. The result is a solved, closed-form formula (not an open problem, not NP-hard) discovered in the 18th century and proven rigorously by Buffon himself.

As a Monte Carlo estimator, convergence follows the Central Limit Theorem: the error after $n$ trials shrinks as $O(1/\sqrt{n})$. Each extra decimal digit of π costs 100× more trials — the same square-root barrier that limits all unbiased Monte Carlo methods. This is in sharp contrast to deterministic algorithms for π like the BBP formula, which can compute any individual hexadecimal digit of π in polynomial time in the digit index (see How Hard Is It to Compute π?).

The historical significance is separate from the computational one: Buffon's experiment is the prototype of geometric probability and inspired both the theory of stochastic simulation and, eventually, the Monte Carlo methods used on the Manhattan Project.

Buffon's needle is the founding example of a family of ideas that now appear across science and engineering:

• Monte Carlo integration: replacing an impossible analytical integral with a random average. The same principle evaluates high-dimensional integrals that no quadrature rule can handle, from quantum chromodynamics to climate models.
• Randomized algorithms: many modern algorithms — from randomized quickselect to primality testing — deliberately introduce randomness to side-step worst-case behaviour, inheriting Buffon's insight that a random sample often tells you something exact.
• Physical simulation: the Monte Carlo Method used to simulate neutron diffusion in the Manhattan Project was named after the casino precisely because it shares Buffon's spirit — random trials revealing average behaviour.
• Financial risk (VaR and stress tests): banks estimate the probability of large losses by simulating millions of random market scenarios, each "needle drop" contributing to an aggregate estimate.
• Computer graphics (path tracing): the stunning realism of modern CGI comes from Monte Carlo integration over light paths — tracing millions of random rays so that averages converge to physically accurate images.

Every one of these applications inherits the same square-root convergence rate — and the same engineering challenge of reducing variance so that fewer samples give better estimates.

Buffon's needle is a remarkable proof that geometry, probability, and arithmetic are the same thing seen from different angles. A needle lands at random, and in the long run the fraction that crosses a line encodes π — not as a coincidence, but as the inevitable consequence of integrating a uniform distribution over a half-circle of angles.

The broader lesson is the Monte Carlo principle: when a quantity is hard to compute analytically, sample it. The price is statistical noise that shrinks only as the square root of the sample count — a fundamental tax on all randomized estimation. Understanding that tax is understanding the reach and the limits of randomized algorithms, from this 18th-century parlour experiment to the central limit theorem that governs them all.