1
$\begingroup$

As the title suggests, I would like to compute probability vectors with elements proportional to (unsigned) Stirling numbers of the first kind by row. For easy reference, here is the Wiki page.

enter image description here

For example, I'd like an efficient algorithm to return $p_6 \approx \frac{[120, \;274, \;225, \;85, \;15, \;1]}{720}$ when $n=6$. Previous MO questions have considered computing Sterling numbers, for example here and here. In my case, however, a reasonable approximation would suffice. I suppose that the scaling by row-sums might actually make the problem easier, as many of the elements decay rapidly with the row number $n$. It is not immediately obvious how to find such an approximation however, so I'm hoping someone here has a reference or some insight to share.

The approach of this paper might prove useful, but I have not digested it as yet.


This came up for me in a the course of a project concerning de Finetti theorems and kernel density estimation procedures, and is well outside my wheelhouse. I would be happy to hear that this question is trivial for combinatorics specialists. I have also tried the usual Google tactics, but precisely because there is a wealth of information about Stirling numbers I have found my question hard to address.

$\endgroup$

1 Answer 1

1
$\begingroup$

I was able to find a reasonably good approximation using bounds given in a tech report of Jim Pitman.

Starting on page 13 an example is worked out which computes bounds on ratios of row-adjacent Stirling numbers using stable and efficient polygamma evaluations; see expressions (33) and (34) specifically.

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

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