We have proved that the limit of $\sum_{k=0}^n r^2k^m / (1+r)^{k+1}$ when n approaches infinity is $\sum_{k=1}^m S(m,k)k!/r^{k-1}$ where S(m,k) is the second kind of stirling number.

Is there a simple asymptotic or approximate formula for the result $\sum_{k=1}^m S(m,k)k!/r^{k-1}$ with $m$ fixed and $r$ near $1$. ?


For the question to make sense, you have to specify what the asymptotics is with respect to. For example, which variables are fixed and which are going to infinity. If $r>0$ is fixed and $m\to\infty$ (and probably in some other cases too), you are better off analyzing your initial sum rather than the Stirling version. The largest term is around $k= m/\ln(1+r)$ and the terms near that have a Gaussian shape with standard deviation $m^{1/2}/\ln(1+r)$. Euler-Maclaurin summation for the main part plus crude bounds for the tails will give it to you.

  • $\begingroup$ A good suggestion. In fact and for our actual applications, $m$ is usuallly fixed and usually less than $10$, and $r$ will vary near 1, often not more than $10$ and not less than $1/2$. Therefore we do want an approximate result for $\sum_{k=1}^m S(m,k)k!/r^{k-1}$. $\endgroup$ – liaomingxue Dec 30 '12 at 11:30
  • 1
    $\begingroup$ So you have an exact expression with usually less than 10 terms. Why do you think there should be something better? $\endgroup$ – Brendan McKay Dec 30 '12 at 14:03

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.