Buffon's Needle, Explained

After reading this you will know why dropping needles on a lined floor estimates π, how to turn a count of crossings into a number near 3.14159, and why the estimate crawls toward the truth instead of racing.

What it is, with one throw to fix the idea

Picture a floor of parallel lines, all spaced the same distance apart. Call that spacing t. Now drop a needle of length L onto the floor, with L \le t so the needle can never cross two lines at once. Sometimes the needle lands clean between two lines. Sometimes it lands across a line. Georges-Louis Leclerc, Comte de Buffon, asked in 1733 how often it crosses. The answer he found in 1777 contains π.

The crossing probability is \frac{2L}{\pi t}. Take a needle exactly as long as the spacing, so L = t. Then the probability is \frac{2}{\pi} \approx 0.6366. Drop 1000 such needles and about 637 should cross a line. Turn that around: if you counted 637 crossings out of 1000 throws, you would estimate \pi \approx \frac{2 \cdot 1000}{637} \approx 3.140. You measured π with a box of needles and no circle in sight.

The circle is hidden in the geometry of a rotating line. π enters through the angle the needle makes with the floor lines, and that is the whole trick.

When this is the right lens, and when it is not

Use Buffon's needle to see how randomness can compute a constant, and to feel the pace of Monte Carlo convergence in your bones. It is a clean example because the exact answer is known, so you can watch the error and check it against theory.

Do not use it to actually compute π. It is one of the worst π estimators ever devised. To pin down π to three correct decimals you may need millions of throws, and even then a single unlucky run can miss. Faster methods exist: a few terms of a good series beat billions of needles. Treat this as a teaching model, not a calculator.

The formula and where π comes from

Fix the needle after it lands. Two things describe how it sits: the distance x from the needle's center to the nearest line, and the acute angle \theta between the needle and the lines. By symmetry, x is uniform on [0, t/2] and \theta is uniform on [0, \pi/2].

The needle crosses the nearest line exactly when its half-length projected perpendicular to the lines reaches across the gap:

Here \frac{L}{2}\sin\theta is how far each half of the needle sticks out toward a line for a given tilt. A needle lying flat along the lines (\theta = 0) reaches nothing and almost never crosses. A needle standing perpendicular (\theta = \pi/2) reaches its full half-length L/2 and crosses most often.

Average the crossing condition over all positions and angles. The probability is the favorable area divided by the total area of the (x, \theta) rectangle:

The integral \int_0^{\pi/2}\sin\theta\, d\theta = 1, and the rest is bookkeeping. That lone π in the denominator is the angle's range \pi/2 refusing to cancel. Solve for π to get the estimator:

Here N is the number of throws and C is the number that crossed. The estimate is only as good as the ratio C/N, which is a noisy measurement of the true probability.

A worked example reproducing the demo

The demo uses the defaults: needle length equal to line spacing, so L = t = 1. With those, the estimator simplifies to \hat{\pi} = 2N/C. Suppose you drop 10000 needles and 6362 of them cross a line.

Reading the running estimate

The headline behavior is slow convergence. The standard error of the estimated probability falls like 1/\sqrt{N}, and the π estimate inherits that rate. Concretely, the error shrinks roughly as \frac{\sqrt{\pi(\pi-2)}}{2\sqrt{N}} for the L=t case, about 0.65/\sqrt{N}. Plug in numbers and you get a table you can sanity-check against your own runs.

Notice each row multiplies N by 10 and cuts the error by about 3.16, which is \sqrt{10}. To gain one more correct decimal digit (a tenfold drop in error) you need 100 times more throws. This is the plain arithmetic of 1/\sqrt{N}.

Common mistakes

The errors here are small in code but large in effect.

A second trap is division by zero: if C = 0, the estimate 2LN/(tC) blows up. With any reasonable N this never happens, but a run of 3 throws with no crossings gives nonsense. Third, watch what you sample. The angle \theta must be uniform on [0, \pi/2] and the offset x uniform on [0, t/2]. Sampling the needle's two endpoints uniformly instead biases the angle and quietly breaks the estimate.

Finally, do not read too much into one lucky run. If your first 500 throws give exactly 3.14159, that is coincidence, not accuracy. Run it again and the digits will move. Judge the method by the spread across many runs, not by one pretty number.

Related tools worth a look

If the randomness-computes-a-number idea grabs you, the Law of Large Numbers demo shows the same averaging settling onto an expected value, and the Central Limit Theorem Demo explains why the error is bell-shaped and why it shrinks like 1/\sqrt{N}. For a hands-on view of Monte Carlo estimation of areas, try the Monte Carlo Playground. To watch chance build a smooth distribution from nothing, drop balls through the Galton Board. And the Random Walk Explorer makes the same square-root scaling visible in the spread of a wandering path.

Related tools

3D Geometry Board

Points, lines, planes, spheres and polyhedra on an orbiting camera — slice solids with planes, unfold nets, drop the Dandelin spheres into a cone and drag everything.

Bézier Curve Playground

Drag control points and watch De Casteljau’s algorithm construct a Bézier curve step by step.

Central Limit Theorem Demo

Sample from any lopsided distribution and watch the averages pile up into a bell curve.

Conway’s Game of Life

Paint cells and watch Conway’s cellular automaton breed gliders, guns and oscillators.

Dynamic Geometry Board

Construct with points, lines, circles and transformations, then drag anything — the construction stays true, and a checker tells you which relations are theorems.

Fantasy Map Generator

Roll a seed and get a continent: coastlines, mountains, rivers, biomes and named towns, ready for your campaign.