## Sunday, March 11, 2012

### Buffon's needle

What a better way to start the blog but with the classical probabilistic puzzle of Buffon's needle?

The statement, as per Wikipedia:
Suppose we have a floor made of parallel strips of wood, each the same width, and we drop a needle onto the floor. What is the probability that the needle will lie across a line between two strips?
For the sake of clarity, we will consider the case where the separation of the strips is $t=1$, the needle's lenght is $l=1$ and the needle can fall equally at every point in the floor. We will call $h$ to the distance from the center of the needle to the closest point of the closest strip, and $\theta$ to the angle of the needle and the strip line.

Now let's think the situation in which the needle will cut the strip. If the angle $\theta$ is zero, then the needle is forced to rest along the strip, therefore $h=0$. If the needle is perpendicular to the strip (the angle is $\frac{\pi}{2}$, then the distance can be anything, up to $\frac{1}{2}$. We see that we can rotate a needle from an angle of zero up to $\pi$ before we get the same position of the needle and that the distance can range from zero (the center of the needle lying on a strip) to $\frac{1}{2}$ (before being closer to another strip).

We must add "as many" combinations of $\theta$ and $h$ make the needle touch a strip. Since these variables are continuous, we need to integrate them. Therefore, to "add" all the values, we fix an angle $\theta$ and think of how far left or right (in the picture above) we can move the needle without preventing it from touching the strip. So, when $\theta=0$ then the needle cannot move without leaving the strip, and thus $h=0$, its minimum value. On the other hand, when $\theta=\frac{\pi}{2}$ then $h$ is completely free to have values within its possible range from $-\frac{1}{2}$ to $\frac{1}{2}$, this is, $\frac{1}{2}$ away from the closest strip. In between, we will be able to move the needle's center within a range $[-\frac{\sin\theta}{2}, \frac{\sin\theta}{2}]$ from the closest strip. Each value must be weighted by its "importance" in the interval we are integrating. As per our assumptions, every point "weights" the same (has the same probability or density in the inverval). For the angles, the density all over the interval is $\frac{1}{\pi}$, and for the distances, $1$.
$$\int_{0}^{\pi} { \int_{-\frac{\sin\theta}{2}}^{\frac{\sin\theta}{2}} {1 dh} \frac{1}{\pi} d\theta} =\\ \int_{0}^{\pi} {\left( \frac{\sin\theta}{2} - (-\frac{\sin\theta}{2}) \right) \frac{1}{\pi} d\theta} = \\ \frac{1}{\pi} \int_{0}^{\pi} {\sin\theta d\theta} = \\ \frac{1}{\pi} \left(- \cos\pi - (-\cos0) \right) = \frac{2}{\pi}$$
Although we didn't say, the equal weighting is called uniform probability. The result would change if the needle was to be thrown with some other probability law.

This problem also gives us the change to estimate $\pi$. If we throw $N$ needles over the floor, and count how may actually met the strips and divide it by $N$, we would get an approximation to $\frac{1}{\pi}$ (by the law of large numbers). Inverting, we have estimated $\pi$. In this image (taken from here), 100.000 needle throws are simulated. If you focus on the angle, what this histogram tells you is the freedom to move the center of the needle away from the closest strip. The farthest distance would be $\frac{1}{2}$, but the author scaled the histogram to one to match it with the $\sin \theta$ function.