This is the solution to problem 9 from Fifty Challenging Problems in Probability by Frederick Mosteller (1987). The problem is paraphrased below; for reference, it is inspired by the original book.

In the gambling game of craps, we roll two fair six-sided dice and observe the total. If we roll a 7 or an 11, we win! We can also win in a more elaborate way:

  1. Throw anything except 2, 3, or 12. Our throw is called a point.
  2. Keep throwing. If we throw the point, we win! If we throw a 7, we lose. Otherwise, we just keep throwing.

What is the probability of winning a game of craps?

Let’s dive in!

Looking at some winning paths

When solving this, I first wrote down some winning sequences of totals:

  • 7;
  • 11;
  • 4, 4;
  • 5, 2, 6, 2, 8, 12, 11, 5;
  • and so on.

Assuming we win, we can either do it on the first throw or with a point.

Probability of winning on the first throw

The probability of winning on the first throw is the same as the probability of rolling a 7 or an 11 immediately. How do we find those? Let’s draw a 6 by 6 table where each row represents the outcome of die D1, each column represents the outcome of die D2, and each cell represents their total.

D1/D2123456
1234567
2345678
3456789
45678910
567891011
6789101112

Since there are 36 cells, the probability of rolling a total of \(x\) is equal to the count of \(x\) in the table, divided by 36. In more standard notation:

\[ X \sim \begin{pmatrix} 2 & 3 & 4 & 5 & 6 & 7 & 8 & 9 & 10 & 11 & 12 \\ \frac{1}{36} & \frac{2}{36} & \frac{3}{36} & \frac{4}{36} & \frac{5}{36} & \frac{6}{36} & \frac{5}{36} & \frac{4}{36} & \frac{3}{36} & \frac{2}{36} & \frac{1}{36} \end{pmatrix}. \]

The probability of winning in the first throw is thus \(P(X = 7) + P(X = 11) = \frac{8}{36}.\)

Probability of winning with a point

To win this way, we first roll to get a point. Then we roll until we get the point again – or 7. A valid sequence can be described as \(n, m_1, m_2, \dots, m_{k}, n\), where \(m_i \notin \{7, n\}\) for all \(i = 1, \dots, k\).

What is the probability of such a sequence? Obviously, it depends on \(n\). We note that \(n\) must be part of a valid set of totals \(V = \{4,5, 6,8,9, 10\}\). The probability of winning with a point is the sum of probabilities of winning with each \(v \in V\). We can summarize this as follows:

\[ P_\mathrm{point-win} = \sum_{v \in V} P(v) P(\textrm{roll many times without seeing 7, nor } v) P(v). \]

The two \(P(v)\) terms describe the probability of rolling \(v\) initially and as the final roll of the sequence.

How do we compute the middle term? Let’s call it \(P_S(v)\). We have to roll zero or more totals, none of which are in \(\{7, v\}\). We can write:

\[ P_S(v) = \sum_{i=0}^\infty P(\textrm{roll neither 7, nor } v)^i. \]

Each term in the sum is the probability of a roll sequence of length \(i\). The sum checks all sequence lengths \(i\). We notice that \(P(\textrm{roll neither 7, nor } v)\) is always the same for a given \(v\) regardless of \(i\). Since probabilities are between 0 and 1, the entire sum is a geometric series of the form:

\[ \sum_{i=0}^\infty q^i = \frac{1}{1 - q}. \]

Let’s compute this individually for all values \(v\in V\). I’ll use #\(v\) to denote the number of times \(v\) appears in the totals table. Let \(q(v) = P(\textrm{roll neither 7, nor } v) = \frac{36 - 6 - \#v}{36}\).

\(v\)#\(v\)\(q(v)\)\(P_S(v) = \sum_{i=0}^\infty q(v)^i\)
43\(\frac{3}{4}\)\(4\)
54\(\frac{13}{18}\)\(\frac{18}{5}\)
65\(\frac{25}{36}\)\(\frac{36}{11}\)
85\(\frac{25}{36}\)\(\frac{36}{11}\)
94\(\frac{13}{18}\)\(\frac{18}{5}\)
103\(\frac{3}{4}\)\(4\)

The probabilities are the same for \(\{4, 10\}\), \(\{5, 9\}\), and \(\{6, 8\}\) due to symmetry in the totals table. What’s left is to compute \(P_\mathrm{point-win}\):

\[ P_\mathrm{point-win} = 2 \cdot \left( \left(\frac{3}{36}\right)^2 \cdot 4 + \left(\frac{4}{36}\right)^2 \cdot \frac{18}{5} + \left(\frac{5}{36}\right)^2 \cdot \frac{36}{11}\right) = \frac{134}{495}. \]

Add the two together and we get our final win probability:

\[ P(\mathrm{Win}) = P(X = 7) + P(X = 11) + P_\mathrm{point-win} = \frac{8}{36} + \frac{134}{495} \approx 0.49293. \]

We’re at a slight disadvantage, but it’s almost even odds!

Thanks for reading! If you liked this post, you can support me on Ko-fi ☕. More FCPP solutions coming soon :)