Buffon's Needle

Rule a sheet of paper with parallel lines spaced distance $d$ apart. Drop a needle of length $L \le d$ onto it at a random position and angle. What's the probability the needle crosses one of the lines? The famous, almost eerie answer — the probability involves $\pi$ — means you can flip this around and estimate $\pi$ just by dropping needles and counting crossings, no circle or diameter in sight.

Setting up the randomness

Two independent random quantities describe where the needle lands: $x$, the distance from the needle's center to the nearest line (uniform between $0$ and $d/2$), and $\theta$, the angle the needle makes with the lines (uniform between $0$ and $\pi/2$ by symmetry — steeper or shallower doesn't matter, only the acute angle to the lines does). Geometrically, the needle crosses the nearest line exactly when its half-length's vertical reach, $\tfrac{L}{2}\sin\theta$, exceeds the distance $x$ to that line: the crossing condition is $x < \tfrac{L}{2}\sin\theta$.

Turning geometry into a probability

For a fixed angle $\theta$, the chance of crossing is the fraction of the $x$-range, $[0, d/2]$, that satisfies $x < \tfrac{L}{2}\sin\theta$: that's $\dfrac{(L/2)\sin\theta}{d/2} = \dfrac{L\sin\theta}{d}$. Averaging this over all equally likely angles $\theta$ from $0$ to $\pi/2$ (integrating and dividing by the range) gives the overall crossing probability:

$$P(\text{cross}) = \frac{2L}{\pi d} .$$

In plain English: the probability of a crossing is proportional to how long the needle is relative to the line spacing, scaled down by a factor of $2/\pi$ — and that $\pi$ enters purely because averaging $\sin\theta$ over a range of angles brings in circular geometry, even though nothing about the setup looks like a circle.

Worked example: estimating $\pi$ from a simulated drop count

Suppose $L = d = 1$ (needle length equals line spacing) and you simulate 1,000 drops, observing 637 crossings. Rearranging the formula, $\pi \approx \dfrac{2L \cdot (\text{drops})}{d \cdot (\text{crossings})} = \dfrac{2 \times 1 \times 1000}{1 \times 637} \approx 3.14$. That's within 0.05% of the true value of $\pi\approx3.14159$, from nothing more than a crossing count — a genuine, if statistically noisy, Monte Carlo estimator for $\pi$, converging slowly (error shrinks like $1/\sqrt{n}$) but converging nonetheless.

Worked example: the case $L = d/2$

If the needle is half the line spacing, $L=1$, $d=2$: $P(\text{cross}) = \dfrac{2(1)}{\pi(2)} = \dfrac1\pi \approx 0.318$. So a shorter needle relative to the spacing crosses lines less often, exactly as the geometry suggests — it has less reach to cross a line from any given starting position.

What this means in practice

Buffon's needle is the historical ancestor of Monte Carlo estimation: instead of computing a quantity directly, you set up a random experiment whose success probability encodes the quantity, then estimate that probability by repeated sampling. That's the exact logic behind pricing path-dependent options or estimating tail risk today — simulate many random scenarios and let the observed frequency stand in for a probability or expectation that's too hard to compute in closed form.