I'm trying to figure out if there is some (relatively simple) formula for calculating the probability of rolling a sum of N with as many rolls as needed with a single regular six-sided die. For example:
$N=1$ is $0.16666 = 1/6 = 1/6$ (1)
$N=2$ is $0.19444 = 7/36 = 1/6 + 1/36$ (2 or 1,1)
$N=3$ is $0.22685 = 49/216 = 1/6 + 2/36 + 1/216$ (3 or 1,2/2,1 or 1,1,1)
$N=4$ is $0.26466 = 343/1296 = 1/6 + 3/36 + 3/216 + 1/1296$ (4 or 2,1/1,2/2,2 or 1,1,2/1,2,1/2,1,1 or 1,1,1,1)

Does this series converge to essentially one and is there some good formula?

It seems like pascal's triangle is involved somehow (at least for n=1-6, but I'm not sure how (if even possible) to convert it into a formula. Any help/advice is appreciated.

  • 2
    $\begingroup$ If you throw a die a lot of times and add your results as you go along, the sum increases by $3.5$ on average for each roll, so my gut says that your probability converges to $\frac{1}{3.5}\approx 0.286$ as $N$ grows. $\endgroup$ – Arthur Jan 5 '15 at 23:27

Let $q(x)=\frac{1}{6}\left(x^1+x^2+x^3+x^4+x^5+x^6\right)$. Write:

$$\begin{align}G(x)&=\sum_{n=0}^\infty q(x)^n\\&=\frac{1}{1-q(x)}\\&=\frac{6}{6-x-x^2-x^3-x^4-x^5-x^6} \end{align}$$

Then the coefficient of $x^N$ in $G(x)$ is your probability.

So you need to know something about the roots of the denominator. $1$ is one such root, but you are going to need to deal with complex roots, as well, or at least get an upper bound for the roots.

The formula is:

$$p(N)=a_1\alpha_1^N+a_2\alpha_2^N+\cdots a_6\alpha_6^N$$

For some constants $a_i$ and with the $\alpha_i$ the inverse roots of the denominator.

According to Wolfram Alpha, the roots of the denominator are $1$ and a bunch of values (one real, four complex) with absolute value greater than $1$.

So for large enough $N$, $p(N)\approx a_1$.

We can find $a_1$ by doing the usual partial fractions computation:

$$\lim_{x\to 1} G(x)(1-x)=\frac{2}{7}=\frac{1}{3.5}$$ which is what Arthur conjectured in comments above.

It turns out for $i>1$, $|\alpha_i|<\frac{3}{4}$, so we have:

$$p(N)=\frac{2}{7} + o\left(\left(\frac 34\right)^N\right)$$

We can also get a recursion:


We get $P(28)=.2857\dots.$ For $n\geq 88$ we get:

$$P(n)=0.285714285714\dots$$ which is $12$ digits of accuracy to $\frac{2}{7}=0.\overline{285714}.$

We also get that $$\frac{P(100)-\frac{2}{7}}{\left(\frac34\right)^{100}}\approx -0.016$$


$$\frac{P(1000)-\frac{2}{7}}{\left(\frac34\right)^{1000}}\approx 7.34\times 10^{-13}$$

(Computing the left side with python Fraction class to get exact value, then converted to float.)

This gives:

$$P(1000)-\frac{2}{7}\approx 8.45\times 10^{-138}$$

  • $\begingroup$ Minor correction: you mean $\lim_{x \to 1} G(x)(1-x)$ instead of $\lim_{x \to 1} q(x)(1-x)$. $\endgroup$ – Joshua P. Swanson Jul 10 '20 at 2:27

Let $P(N)$ be the probability rolling a sum of $N$ with any number of dice. If we consider the number of ways we can roll $m$ dice to sum to $N$, it should just be number of ways we can roll $(m-1)$ dice and get within $6$ of $N$, times the probability of getting the roll we need.

For example considering $N=43$, then $P(43)$ should be equal to $P(42)$ times the probability of rolling a $1$, plus $P(41)$ times the probabilty of rolling a $2$, plus ..., plus $P(37)$ times the probability of rolling a $6$.

For $N > 6$ we can define $P(N)$ recursively as: $$ \begin{align} P(N) &= \frac{1}{6}P(N-1) + \frac{1}{6}P(N-2) + \dotsb + \frac{1}{6}P(N-6) \\ &= \frac{1}{6}\Big(P(N-1) + P(N-2) + \dotsb + P(N-6)\Big) \\ &= \frac{1}{6}\sum_{i=N-1}^{N-6}P(i) \end{align} $$

We should consider $N \leq 6$ separately because there are not six previous terms: $$ P(N) = \frac{1}{6}\sum_{i=0}^{N-1}P(i) $$ (Note that we let $P(0) = 1$ since we can always roll $0$ dice for a sum of $0$.)

So for each $N$ after $6$, $P(N)$ is just average of the six previous $P$. I'm sure there should be a closed form for this, but it is outside of my ability to find this closed form.

  • 1
    $\begingroup$ I think you should start with $P(0)=1$ and your $N\leq 6$ should include $0$ in the sum. $\endgroup$ – Thomas Andrews Jan 5 '15 at 23:53
  • 1
    $\begingroup$ It should also be $\frac{1}6$, not $\frac 1N$ in the $N\leq 6$ case. $\endgroup$ – Thomas Andrews Jan 5 '15 at 23:54
  • $\begingroup$ @ThomasAndrews right you are. It will be edited. $\endgroup$ – Mike Pierce Jan 5 '15 at 23:55

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.