I believe, based on numerical evidence, that $$\sum_{n=2}^\infty \frac{x^n}{(\log n) n!} \sim \frac{\exp(x)}{\log(x)}$$ as $x\to\infty$. However, I am not sure how to prove this. What would be a good way to approach this problem?


Consider the related finite sum $$ s(x)=\sum\limits_n\frac{x^n}{n!}\mathbf 1_{x/2\leqslant n\leqslant 2x}=\mathrm e^x\cdot\mathbb P(x/2\leqslant N_x\leqslant 2x), $$ where $N_x$ is a Poisson random variable with parameter $x$. A large deviation estimates ensures that, for every $x$ large enough, $\mathbb P(x/2\leqslant N_x\leqslant 2x)=1-p(x)$ where $0\lt p(x)\leqslant\mathrm e^{-cx}$ for some positive $c$.

The series $f(x)$ to be estimated is $f(x)=g(x)+h(x)$ where $g(x)$ sums the terms such that $x/2\leqslant n\leqslant2x$ and $h(x)$ sums the other terms. Note that $0\leqslant h(x)\leqslant\mathrm e^x p(x)/\log2$ and $s(x)/\log(2x)\leqslant g(x)\leqslant s(x)/\log(x/2)$, hence $$ \mathrm e^{x}\frac{1-\mathrm e^{-cx}}{\log x+\log2}\leqslant f(x)\leqslant\frac{\mathrm e^x}{\log x-\log2}+\mathrm e^x\frac{\mathrm e^{-cx}}{\log2}. $$ Finally, $$ f(x)=\frac{\mathrm e^x}{\log x}\,\left(1+O\left(\frac1{\log x}\right)\right). $$ Note: With some more care, one can replace the $O\left(\frac1{\log x}\right)$ error term by $O\left(x^{a-1/2}\right)$, for every positive $a$.


Here is a possible way but is not a complete solution. (This is a bit too long for a comment and hence have community wikied it)

From Stirling, we have that $$n! \sim \sqrt{2 \pi n} \left(\dfrac{n}{e} \right)^n$$ Equivalently, $\exp(x) \sim \sqrt{2 \pi} \dfrac{x^{x+1/2}}{x!}$. Hence, $$\exp(x) \sim \sum_{n= x- \sqrt{x}}^{x+\sqrt{x}} \dfrac{x^n}{n!}$$ (The above step needs proof.)

Hence, $$\sum_{n= x- \sqrt{x}}^{x+\sqrt{x}} \dfrac{x^n}{n! \log (n)} \sim \sum_{n= x- \sqrt{x}}^{x+\sqrt{x}} \dfrac{x^n}{n! \log (x)} \sim \dfrac{\exp(x)}{\log(x)}$$ where we have used the fact that $\log(x \pm \sqrt{x}) \sim \log(x)$.

  • $\begingroup$ For each positive $a$, the sum from $x-a\sqrt{x}$ to $x+a\sqrt{x}$ is equivalent to a multiple of $e^x$, not to $e^x$. $\endgroup$
    – Did
    Nov 5 '12 at 9:09
  • $\begingroup$ @did Isn't that absorbed in $\sqrt{2 \pi}$? Anyway let me get back to it tomorrow morning to try and fix this. $\endgroup$
    – user17762
    Nov 5 '12 at 9:21

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.