1

Can anyone suggest an algorithm for iterating once through all matrices with non-negative integer entries which are $2$ by $n$ with fixed row sums ($r_1$ and $r_2$) and fixed column sums ($c_1, c_2, \ldots c_n$)? For instance, for row and column sums $r_1= r_2 = c_1 = c_2 = 2$, (if I haven't make a mistake) the possible matrices would be

$\left(\begin{array}{cc} 2 & 0\\ 0 & 2 \end{array}\right)$

$\left(\begin{array}{cc} 1 & 1\\ 1 & 1 \end{array}\right)$

$\left(\begin{array}{cc} 0 & 2\\ 2 & 0 \end{array}\right)$

However, I would want to do this for general $2$ by $n$ matrices, not just $2$ by $2$.

I would also like to know if there is a quick way to compute the total number of matrices I would have to iterate over.

Mike Spivey
  • 55,550
Henry B.
  • 2,028

2 Answers2

3

If you are doing 2xn enumerating them is easy, as all you need to do is enumerate the first row, (the second row is forced) and this is just the problem of enumerating all vectors with a given sum and bounds on each element. This is easy to do recursively, but if you want the total count and a nice, easy to compute bijection with the natural numbers, see "On uniform generation of two-way tables with fixed margins and the conditional volume test of Diaconis and Efron" by Holmes and Jones (see section 2b). Larger tables are considerably more difficult.

See also the answers to these two math.stackexchange questions.

Glorfindel
  • 3,955
deinst
  • 5,646
0

This is an old post, but I ran across it looking for code able to do this for slightly larger matrices. So, in case anyone else runs across this post looking for the same, I ended up writing my own solution in python. You can find it here. It will iterate over about a million matrices per minute, so is quite adequate for smaller matrices.

hawkjo
  • 433