10

$$\lim_{x\rightarrow +\infty}\sqrt{x}e^{-x}\left(\sum_{k=1}^{\infty}\frac{x^{k}}{k!\sqrt{k}}\right)$$ Any hint will be appreciated.

Note: There is a related question on MathOverflow: Asymptotic expansion of $\sum\limits_{n=1}^{\infty} \frac{x^{2n+1}}{n!{\sqrt{n}}}$. The MO question references this question and it also links to some other posts containing similar expressions.

z3wood
  • 1,105
  • Just to let you know, \sum will give you a nicely-formatted $\sum$ symbol that deals with things appearing underneath and above it without any extra effort. – Patrick Stevens Jan 28 '17 at 13:20
  • 2
    write $\frac{1}{\sqrt{k}}=\frac{1}{\sqrt{\pi}}\int_{\mathbb{R}_+}e^{-kq^2}dq$ the sum then becomes simple and the remaining integral can be approximated by standard methods – tired Jan 28 '17 at 17:15
  • I tried you way,then it becomes $\sqrt{x}e^{-x}\sum_{k=1}^{\infty}(\frac{2}{\sqrt{\pi}}\int_{0}^{\infty}e^{-kq^{2}}dq)\frac{x^{k}}{k!}$.How to deal with this one? – z3wood Jan 29 '17 at 01:43
  • the sum becomes trivial... – tired Jan 30 '17 at 17:42
  • $\sum_{k=1}^{\infty}...=e^{xe^{-y^2}}-1$ – tired Jan 30 '17 at 21:01
  • @tired Are you sure that this works? How does one estimate the integral $\int_0^\infty (e^{xe^{-y^2}}-1),dy$ for large $x$? I don't believe Laplace's method works directly, but perhaps deforming the integration path and applying the method of steepest decent works. I have not pursued. What did you have in mind? – Mark Viola Feb 02 '17 at 00:30
  • @Dr.MV one can for example integrate by parts to get an integral which is bette suited for the Laplace method. Otherwise a matched asymptotic expansion also works (after some substitutions) – tired Feb 02 '17 at 01:41
  • strange, no work shown so much upvotes. everybody gets downvoted – jimjim Feb 02 '17 at 11:36
  • @Arjang maybe because this is an interesting problem? but in principle you are right, that is the reason why i resist to write up an answer and only give some hints – tired Feb 02 '17 at 12:29
  • A discussion on meta which is (to some extent) related to this post. – Martin Sleziak Jan 26 '20 at 07:59

1 Answers1

9

(I have overwritten my previous incorrect answer.)

For any positive constant $c > 1$ we have: \begin{align*}S(cx) = e^{-x}\sum\limits_{k \ge cx} \frac{x^k}{k!} &= e^{-x}\frac{x^{cx}}{(cx)!}\left(1 + \frac{x}{(cx+1)} + \frac{x^2}{(cx+1)(cx+2)} + \cdots\right) \\& \le e^{-x}\frac{x^{cx}}{(cx)!}\sum\limits_{k=0}^{\infty}\frac{1}{c^k} < \frac{c}{c-1}e^{-x}x^{cx}\frac{e^{cx}}{(cx)^{cx}} = \frac{ce^{-((c\log c) -c +1)x}}{c-1}\end{align*}

where, we used the inequality, $x! > e^{-x}x^{x}$ for sufficiently large $x$. Also note that $(c\log c) - c + 1 > 0$ for all $c > 0$

Again, for sufficiently large $x$: \begin{align*}T(x/c) = e^{-x}\sum\limits_{0 \le k < x/c} \frac{x^k}{k!} = e^{-x}\sum\limits_{0 \le k < x/c} \frac{c^k(x/c)^k}{k!} & \le e^{-x}\frac{(x/c)^{x/c}}{(x/c)!}\sum\limits_{0 \le k < x/c} c^k \\& < e^{-x}e^{x/c}\frac{xc^{x/c}}{c} = \frac{x}{c}e^{-\left(1 - \frac{1}{c} - \frac{\log c}{c}\right)x}\end{align*} where, we note that $(1 - \frac{1}{c} - \frac{\log c}{c}) > 0$.

Hence, we have for any $c>1$, $$\displaystyle e^{-x}\sum\limits_{x/c \le k \le cx} \frac{x^k}{k!} = 1 - S(cx) - T(x/c) \quad \underbrace{\longrightarrow}_{ x \to +\infty} \quad 1 \tag{1}$$

Now, we have the simple estimates, $$\frac{e^{-x}}{\sqrt{cx}}\sum\limits_{x/c \le k \le cx} \frac{x^k}{k!} < e^{-x}\sum\limits_{x/c \le k \le cx} \frac{x^k}{k!\sqrt{k}} < \frac{e^{-x}}{\sqrt{x/c}}\sum\limits_{x/c \le k \le cx} \frac{x^k}{k!} \tag{2}$$

and $$e^{-x}\sum\limits_{k \not\in [x/c,cx]} \frac{x^k}{k!\sqrt{k}} < T(x/c) + S(cx) \tag{3}$$ where, both the terms in the upper bound decays exponentially.

Hence, from $(1),(2)$ and $(3)$ we have that the required limit: $$\displaystyle \lim\limits_{x \to \infty} \sqrt{x}e^{-x}\sum\limits_{k=1}^{\infty} \frac{x^k}{k!\sqrt{k}} \in \left(\frac{1}{\sqrt{c}},\sqrt{c}\right)$$ for arbitrary $c > 1$, i.e., the required limit is $1$ (as pointed out by user tired in the comments).

I suppose for any function $\alpha(x)$ with atmost polynomial growth and continuous at $x = 1$ we can slightly modify the above argument and show, $$\lim\limits_{x \to \infty} e^{-x}\sum\limits_{k=1}^{\infty} \alpha\left(\frac{k}{x}\right)\frac{x^k}{k!} = \alpha(1)$$

r9m
  • 17,938
  • 1
    It's nice to see you've returned. And (+1) -Mark – Mark Viola Feb 01 '17 at 19:26
  • @Dr.MV Thanks! :-) – r9m Feb 01 '17 at 20:31
  • Just curious ... How does one justify the inequality $$e^{-x}\sum_{j\ge 1 }\frac1{\sqrt{j}} \sum_{k>j}\frac{x^k}{k!}\le \sum_{j\ge 1 }\frac1{\sqrt{j}} \sum_{k>j}\frac{1}{k!}$$ – Mark Viola Feb 02 '17 at 00:14
  • Also, I have $$\sum_{j=1}^\infty \frac{1}{\sqrt{j}}\sum_{k=j+1}^\infty \frac1{k!}=\sum_{k=2}^\infty \frac1{k!}\sum_{j=1}^{k-1}\frac{1}{\sqrt{j}} \le \sum_{k=2}^\infty \frac1{k!}\sum_{j=1}^{k-1}(1)=1$$which is tighter than the bound you referenced. How did you obtain it? – Mark Viola Feb 02 '17 at 00:26
  • @Dr. MV i think the correct limit is one – tired Feb 02 '17 at 01:41
  • @tired Do you mean that this solution is in error? I haven't been able to justify the one estimate I referenced. – Mark Viola Feb 02 '17 at 02:14
  • @Dr.MV $$ \int_{0}^{\infty}dy(e^{xe^{y^2}}-1)=x\int_0^{\infty}dyy^2e^{-y^2}e^{xe^{-y^2}}=x\sum_{n\geq0}\frac{1}{n!}\int_0^{\infty}dyy^{2n+2}e^{xe^{-y^2}}\sim xe^{x} \sum_{n\geq0}\frac{1}{n!}\int_0^{\infty}dyy^{2n+2}(e^{-xy^2}+x\mathcal{O}(y^4))\sim xe^{x}\int_{0}^{\infty}dy(y^2+\mathcal{O}(y^4))e^{-xy^2}(1+\mathcal{xO(y^4)})=\frac{\sqrt{\pi}}{2}\frac{e^{x}}{\sqrt{x}}\left(1+\mathcal{O}\left(\frac{1}{x^2}\right)\right) $$

    from which i conclude that the limit is $1$

    – tired Feb 02 '17 at 02:15
  • @Dr.MV sorry if there are some mistakes (it is three o'clock in the morning here) but the general reasonig should be correct – tired Feb 02 '17 at 02:17
  • http://www.wolframalpha.com/input/?i=(e%5E(-20)Sqrt%5B20%5D)*Sum%5B20%5Ek%2F(k!+Sqrt%5Bk%5D),%7Bk,1,2000%7D%5D – tired Feb 02 '17 at 02:24
  • @Dr.MV i'm also happy for any feedback..this one is thougher then i thought ;) – tired Feb 02 '17 at 02:31
  • @r9m it is strongly suggested by numerical experiments (and a semiheuristic argument of mine) that the correct limit is $1$. Nevertheless (+1) for the intersting technique – tired Feb 02 '17 at 12:33
  • 1
    @r9m I don't believe that $e^{-x}x^k$ is bounded by $1$ for each $k>j$, is it? Fix $x$ and surely for $k$ large enough, $x^k>e^x$ – Mark Viola Feb 02 '17 at 14:08
  • @Dr.MV You are right .. :( God knows how I missed that .. I'll try to fix the argument. – r9m Feb 02 '17 at 15:17
  • 2
    @tired I have edited my answer, hopefully I didn't screw up again! :-) – r9m Feb 03 '17 at 01:53