Problem 56
Medium Classical Probability

Banach's matchbox problem. A mathematician carries two boxes…

Banach's matchbox problem. A mathematician carries two boxes of matches, each initially containing $n$ matches. Each time he needs a match, he chooses one of the boxes at random. Find the probability that when he first finds a box empty, the other box contains exactly $r$ matches $(r=0,1,\dots,n)$.

Explanation of Method

Each choice is a fair coin toss: left box or right box, probability $\tfrac12$ each. He finds the left box empty when he chooses it for the $(n+1)$-th time (it had only $n$ matches). If at that moment the right box still has $r$ matches, then he took $n-r$ matches from it. So among his first $2n-r$ choices exactly $n$ were "left" and $n-r$ were "right", and the $(2n-r+1)$-th choice is "left". This is a binomial count; doubling it for the symmetric case (right box found empty) gives the answer.

Step 1: The random choices

Call the boxes L and R. Every time a match is needed, he picks L or R with probability $\tfrac12$ each, independently. Each box starts with $n$ matches.

Important: he finds a box empty only when he opens it and there is nothing inside. So box L is found empty at the moment L is chosen for the $(n+1)$-th time.

Step 2: The event "L is found empty, R has $r$ matches"

At that moment:

So the first $2n-r$ choices contain exactly $n$ L's and $n-r$ R's (in any order), and choice number $2n-r+1$ is L.

Step 3: Its probability

The number of orders of $n$ L's and $n-r$ R's is $\binom{2n-r}{n}$. Each particular sequence of $2n-r+1$ choices has probability $\left(\tfrac12\right)^{2n-r+1}$. Hence

\[ P(\text{L found empty, R has } r) = \binom{2n-r}{n}\left(\frac12\right)^{2n-r+1}. \]

Step 4: Add the symmetric case

By symmetry, "R found empty, L has $r$" has the same probability, and the two cases cannot happen together. Therefore

\[ \boxed{P(r) = 2\binom{2n-r}{n}\left(\frac12\right)^{2n-r+1} = \binom{2n-r}{n}\frac{1}{2^{\,2n-r}}, \qquad r=0,1,\dots,n.} \]

Step 5: Example — $n=5$

$r$$P(r)$Value
$0$$\binom{10}{5}/2^{10}=252/1024$$0.2461$
$1$$\binom{9}{5}/2^{9}=126/512$$0.2461$
$2$$\binom{8}{5}/2^{8}=56/256$$0.2188$
$3$$\binom{7}{5}/2^{7}=21/128$$0.1641$
$4$$\binom{6}{5}/2^{6}=6/64$$0.0938$
$5$$\binom{5}{5}/2^{5}=1/32$$0.0312$
total$1$ ✓

Surprisingly, $r=0$ and $r=1$ are equally likely and are the most likely values: when one box is found empty, the other is usually almost empty too.

Step 6: Why $P(0)=P(1)$

\[ \frac{P(1)}{P(0)} = \frac{\binom{2n-1}{n}/2^{2n-1}}{\binom{2n}{n}/2^{2n}} = \frac{2\binom{2n-1}{n}}{\binom{2n}{n}} = 1, \]

because $\binom{2n}{n} = \binom{2n-1}{n}+\binom{2n-1}{n-1} = 2\binom{2n-1}{n}$.

Step 7: A common variant

If the question were "when one box becomes empty (the last match is taken), the other has $r$ matches", the first $2n-r-1$ choices would contain $n-1$ L's, and the answer changes to $\binom{2n-r-1}{n-1}\big/2^{\,2n-r-1}$ for $r\ge1$. The statement here says "first finds an empty box", which is Banach's original version.


Result

Answer
Probability$P(r)=\binom{2n-r}{n}\Big/2^{\,2n-r},\quad r=0,1,\dots,n$
Example $n=5$$P(0)=P(1)\approx0.246$, $P(5)\approx0.031$