3

I'm looking for ways to compute the coefficients of the power series $$ \sum_{n=0}^\infty\frac{(-1)^n}{n!^s}=\sum_{k=0}^{\infty}c_k s^k $$ (a prior version of the question asked whether such an expansion exists at all, at least in the asymptotic sense as $s\to0^+$; thanks to the answer by @Svyatoslav, now I know that it does).

For now, I know that $\color{blue}{c_0=\frac12}$ and $\color{blue}{c_1=\frac14\log\frac2\pi}$. The value of $c_0$ is obtained e.g. in this answer, and I've just used this approach to get the value of $c_1$. Roughly, for $f(s)=\sum_{n=0}^\infty(-1)^n e^{-\lambda_n s}$, we define $$ f_0(t)=\sum_{n=0}^\infty\mathbf{1}_{(\lambda_{2n},\lambda_{2n+1})}(t),\qquad f_{n+1}(t)=\int_0^t f_n(x)\,dx $$ and, under certain conditions, if we have $f(s)\asymp\sum_{k=0}^{(\infty)}c_k s^k$ as $s\to 0^+$, then $$ c_n=\lim_{t\to\infty}\left(\frac{f_{n+1}(t)}{t}-\sum_{k=0}^{n-1}\frac{c_k t^{n-k}}{(n-k+1)!}\right). $$ We replace $t$ by $\lambda_m$ (and $m\to\infty$) and use $f_n(\lambda_m)=(1/n!)\sum_{k=0}^{m-1}(-1)^k(\lambda_m-\lambda_k)^n$.

With $\lambda_n=\log n!$, (even) the computation of $c_1$ this way results in a limit with sums of logarithmic terms, which is not too hard, but a little bit tedious, to evaluate. And $c_2$ is out of my patience for now ;)

metamorphy
  • 43,591

1 Answers1

2

Denoting $\,S(s)=\displaystyle \sum_{n=0}^\infty\frac{(-1)^n}{n!^s}=\sum_{n=0}^\infty(-1)^ne^{-s\ln\Gamma(n+1)}$ we can use the Lindelöf summation formula for alternating series (also, for example, here, formula 46). Decomposing the numerator $$S(s)=\int_{-\infty}^\infty\frac{e^{-s\ln\Gamma\big(\frac12+ix\big)}}{2\cosh\pi x}dx=\int_{-\infty}^\infty\frac{dx}{2\cosh\pi x}-\frac s2\int_{-\infty}^\infty\frac{\ln\Gamma\big(\frac12+ix\big)}{\cosh\pi x}dx+O(s^2)$$ $$=\frac12-\frac s4\int_{-\infty}^\infty\frac{\ln\big|\Gamma\big(\frac12+ix\big)\big|^2}{\cosh\pi x}dx+O(s^2)=\frac12-\frac s4\int_{-\infty}^\infty\frac{\ln\frac\pi{\cosh\pi x}}{\cosh\pi x}dx+O(s^2)$$ $$=\frac12-\frac14\ln\Big(\frac\pi2\Big)\,s+O(s^2)$$ I'm not sure whether the next asymptotic term can be found in a closed form: $$c_2=\frac14\int_{-\infty}^\infty\frac{\ln^2\Gamma\big(\frac12+ix\big)}{\cosh\pi x}dx=\frac14\Re\,\int_{-\infty}^\infty\frac{\ln^2\Gamma\big(\frac12+ix\big)}{\cosh\pi x}dx$$

Svyatoslav
  • 20,502
  • 1
    Shouldn't the $1/8$ still be $1/4$ in your last expression? $c_n =\operatorname{Re}( c_n)$. – Gary Apr 11 '24 at 05:59
  • @Gary, Yes, this is a typo. Thank you! – Svyatoslav Apr 11 '24 at 11:15
  • 1
    You probably meant "numerator" instead of "nominator" – jjagmath Apr 11 '24 at 11:45
  • @jjagmath, yes, thank you! – Svyatoslav Apr 11 '24 at 11:54
  • 1
    Thanks a lot! (I'm about to accept the answer, upvoted it earlier. At least, now we have a computationally more attractive alternative.) – metamorphy Apr 13 '24 at 10:12
  • Thank you for the interesting problem! I apologise for using a ready formula - not enough skill to derive something similar from the scratch, though I usually try :) – Svyatoslav Apr 13 '24 at 13:02
  • @metamorphy, I'm sorry to disturb you. Just solved an interesting problem of asymptotics - https://math.stackexchange.com/questions/4984411/asymptotic-expansion-when-e-m-formula-fails/4986930#4986930 Maybe it will be interesting for you, too. Have a nice day! – Svyatoslav Oct 20 '24 at 03:51
  • 1
    @metamorphy, I'm sorry to disturb you again. There is a problem of getting asymptotics - https://math.stackexchange.com/questions/5073077/does-int-01-fracdx-leftx-sqrta1-xa1-right2-have-a-closed-f (at $a\to\infty$). The evaluation of the first several terms confirms the probable nice form of the asymptotic, though I'm not able to find an adequate approach to reveal its inherent symmetry. Maybe this problem will interest you. I'm convinced that such teasing problems should not be left unsolved :) I wish you a nice day! – Svyatoslav Jun 17 '25 at 17:34