0
$\begingroup$

Consider the power series $\sum_{k=0}^\infty \sqrt k \frac{x^k}{k!}$. It is easily seen that its radius of convergence is $\infty$. I'm looking for an elementary proof of the asymptotic expansion $$\sum_{k=0}^\infty \sqrt k \frac{x^k}{k!}=e^x\left(\sqrt x - \frac{1}{8 \sqrt x} + o\left(\frac{1}{\sqrt x} \right)\right)$$ as $x$ goes to $\infty$.

Using Stirling's estimate and an asymptotic property of power series, one may derive $$\sum_{k=0}^\infty \sqrt k \frac{x^k}{k!}\sim \frac{1}{\sqrt{ 2\pi}}\sum_{k=0}^\infty \frac{(ex)^k}{k^k}$$

so the question boils down to finding an estimate of $$\sum_{k=0}^\infty \frac{x^k}{k^k}$$

I find this answer quite unconvincing since it makes heavy use of asymptotics of special functions and is not very rigorous .

I'd be satisfied if someone showed how to derive the simpler estimate $$\sum_{k=0}^\infty \sqrt k \frac{x^k}{k!}\sim e^x \sqrt x$$

$\endgroup$
1

1 Answer 1

4
$\begingroup$

This answer deals with different approaches to the asymptotics of the series in question. Here I will try to give a probabilistic argument.


This series has the following probabilistic interpretation: If $N_t$ is a Poisson distribution of rate $t > 0$ then we have

$$ \mathbb{E}[\sqrt{N_t}] = \sum_{k=0}^{\infty} \sqrt{k} \, \frac{t^k}{k!}e^{-t}. $$

Together with some basic inequalities we can get the leading order of the asymptotics. For instance, applying Jensen's inequality with the concave function $x \mapsto \sqrt{x}$ tells that

$$ \mathbb{E}[\sqrt{N_t}] \leq \sqrt{\mathbb{E}[N_t]} = \sqrt{t} $$

whereas writing $ \mathbb{E}[\sqrt{N_t}] = t \sum_{k=0}^{\infty} \frac{t^k}{k!} \frac{1}{\sqrt{k+1}} e^{-t} = t \mathbb{E}\left[ (N_t + 1)^{-1/2} \right] $ and applying Jensen's inequality with the convex function $x \mapsto (x+1)^{-1/2}$ gives

$$ \mathbb{E}[\sqrt{N_t}] \geq \frac{t}{\sqrt{\mathbb{E}[N_t] + 1}} = \frac{t}{\sqrt{t+1}}. $$

In terms of the original series, this reads as

$$ \frac{x}{\sqrt{x+1}} e^x \leq \sum_{k=0}^{\infty} \sqrt{k} \, \frac{x^k}{k!} \leq \sqrt{x} e^x, $$

which is enough to derive the asymptotics $\sim \sqrt{x} e^x$. Higher order terms can also be extracted by utilizing various concentration behaviors of $N_t$.

$\endgroup$
2
  • 1
    $\begingroup$ So this is a duplicate ! Thanks for providing the link to the original question, as well as a new approach. $\endgroup$ Nov 2, 2017 at 9:43
  • $\begingroup$ @GabrielRomon Thank you for accepting this! Also I would consider it a border case, and personally I think it is a good chance to invite other users to ponder about it and provide self-contained and not-hard-to-swallow arguments. All the answers in the link (including my answer there) are either not self-contained or kind of hand-waving, which are probably not what you wanted. $\endgroup$ Nov 2, 2017 at 9:48

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.