Given an alphabet of size $n$, how many strings of length $c$ contain every single letter of the alphabet at least once?

I first attempted to use a recurrence relation to work it out:

$$ T(c) = \left\{ \begin{array}{cr} 0 &\mbox{ if $c<n$} \\ n! &\mbox{ if $c = n$} \\ T(c-1) \cdot n \cdot c &\mbox{ if $c > n$} \end{array} \right. $$

As there's no strings that contain every letter if c < n, and if c = n then it's just all permutations. When c > n you can take any string of size (c-1) that contains all letters (of which there are $T(c-1)$ to choose from), you choose which letter to add (of which there are $n$ choices) and there are $c$ different positions to put it. However, this gives out results that are larger than $n^c$ (the total number of strings), so it can't be right, and I realised it was because you could count some strings multiple times, as you can make them taking different inserting steps.

Then I thought about being simpler: you choose n positions in the string, put each letter of the alphabet in one of those positions, then let the rest of the string be anything:

$$ {c\choose{n}} \cdot n! \cdot n^{c-n} $$

But again this counts strings multiple times.

I've also considered using multinomial coefficients, but as we don't know how many times each letter appears in the string it seems unlikely they would be much help. I've also tried several other methods, some complicated and some simple, but none of them seem to work.

How would you go about working out a formula for this? I'm sure there's something simple that I'm missing.

  • 1
    $\begingroup$ Count instead those that do not contain all letters. You will need to use inclusion exclusion. $\endgroup$
    – user119256
    Feb 5, 2014 at 16:02
  • $\begingroup$ You can use multinomial coefficients the following way: sum the multinomial coefficients over the ordered partitions of $c$ into $n$ non-zero naturals. $\endgroup$
    – Arthur
    Feb 5, 2014 at 16:06

2 Answers 2


Let $W(c,n)$ denote the number of words of length $c$ from an alphabet of $n$ letters. Then $W(c,n)=n^c$.

Out of these, the number of words of the same size that do not contain one of the letters is $W(c,n-1)=(n-1)^c$. The number of ways of choosing which letter is missing is $\binom{n}{1}$.

The number of words of the same size that do not contain two letters is $W(c,n-2)=(n-2)^c$. The number of ways of choosing which two letters are missing is $\binom{n}{2}$... and so on ...

Now we use inclusion-exclusion principle: (subtract the number of words missing one of the letters, then add the number missing two of the letters, subtract the number missing three of the letters,...)

We get:


This is




Another way could be: Denote $S_c^n$ the number of ways to partition the word of length $c$ into $n$ pieces. Then we just need to choose which letter goes to each of the $n$ pieces. This number is $n!$. So the number of words we are looking for is


The numbers $S_c^n$ are called Stirling's numbers of the second kind.

  • $\begingroup$ Thanks, your explanation was great, and I can't believe I forgot about Stirling numbers $\endgroup$
    – John
    Feb 5, 2014 at 17:03
  • $\begingroup$ The second way doesn't convince me. You can split into pieces; but then you have to distribute letters on each position of each piece, and the letter "for the piece" might repeat, o apear on some other piece, and ... $\endgroup$
    – vonbrand
    Feb 5, 2014 at 23:40
  • 1
    $\begingroup$ @vonbrand It doesn't convince you because you haven't understood it. The parts of the partition represent the portions that are going to have a constant letter. Portions ofthe word with constant letter are the partition in each case. That is why we partition in $n$ part, the same number as the letters we want to use. $\endgroup$
    – user119256
    Feb 6, 2014 at 4:01

As you said, there are no solutions if c<=n.

then all results will contain all n letters and c-n random letters so the combinations are:


But string have fixed positions so we need all the permutations out of this:


For each character (c-n) there are possible duplicates of letters for which ordering does not meter, so have to divide to that.

((c-n).n).c!/ (c-n)! //not sure if this is correct

This is just a line of reasoning I hope to get you further, not sure if it is all correct.

  • $\begingroup$ You are counting double strings that have all letters in certain n positions but that also have all n letters in another n positions. $\endgroup$
    – user119256
    Feb 5, 2014 at 16:33

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .