2
$\begingroup$

Let $\left(u_n\right)_{n\in\mathbb{N^*}}$ be the sequence defined $\forall n\in\mathbb{N^*}$ by : $$u_n=\sum\limits_{k=1}^{n}e^{-b^{-k}}$$ With $b\in(1,+\infty)$.

We can instictively say that when $n\to\infty$, since $\lim\limits_{k\to\infty}e^{-b^{-k}}\to 1$, it will tend to be like $\sum\limits_{k=1}^{n}1=n$, and thus I expect the dominant term of the asymptotic expansion of $\left(u_n\right)$ to be $n$ :

$$u_n=n+o(n)$$

And numerical simulation shows indeed that $u_n\sim n$ for large n.

Now, this is no demonstration ; and not only am I trying to prove it rigorously, but also (and mainly) to expand it a bit more and find out the next few terms (well, at least the second term) of the asymptotic expansion...

Any suggestion ?

$\endgroup$
2
$\begingroup$

Since $e^{-x}=1-x+x^2/2!-x^3/3!+\ldots$, your sum is \begin{align} \sum_{k=1}^n e^{-b^{-k}}&= n-\sum_{m=1}^\infty \frac{(-1)^{m+1}}{m!}\frac{1-b^{-mn}}{b^m-1} \\&=n-\frac{1-b^{-n}}{b-1}+\frac{1}{2}\frac{1-b^{-2n}}{b^2-1}-\frac{1}{6}\frac{1-b^{-3n}}{b^3-1}\ldots+\frac{(-1)^{m+1}}{m!}\frac{1-b^{-mn}}{b^m-1}+\ldots \end{align} Immediately, this tells us that $u_n\sim n-\sum_{m=1}^\infty \frac{(-1)^{m+1}}{m!}\frac{1}{b^m-1}$. If $b$ is large enough, then $u_n\approx_{n\to\infty} n+e^{-b^{-1}}-1\sim n -\frac{1}{b}$.

$\endgroup$
  • $\begingroup$ The expansion for $1 - b^{-m n}$ cannot be correct ($b^{-m n}$ decays exponentially). The next terms in the expansion of $u_n$ after $n - \alpha$ are simply $$\sum_{m \geq 1} \frac {(-1)^{m - 1} b^{-m n}} {m! (b^m - 1)}.$$ $\endgroup$ – Maxim Feb 2 at 18:24
  • $\begingroup$ Thank you, good job finding the third term of the expansion ! $\endgroup$ – Harmonic Sun Feb 2 at 19:47
4
$\begingroup$

If we use the Taylor series expansion of $e^x$: $$e^x=\sum_{n=0}^{\infty}\frac{x^n}{n!}$$ We have the following asymptotic estimation for your sequence (as for large $n$ we have $n \sim \infty$): $$u_n \sim \sum_{k=1}^{\infty}\sum_{n=0}^{\infty} \frac{(-b^{-k})^n}{n!}$$ $$=\lim_{n \to \infty}\sum_{k=1}^{n}\Big(1-\frac{1}{b^k}+\frac{1}{2! \cdot b^{2k}}-\frac{1}{3! \cdot b^{3k}}+...\Big)$$ $$= n-\frac{1}{b-1}+\frac{1}{2! \cdot (b^2-1)} - \frac{1}{3! \cdot (b^3-1)}+...$$ $$=n+\sum_{k=1}^{\infty}\frac{(-1)^k}{k! \cdot(b^k-1)}$$ In the above expansion I used the fact that $1+x+x^2+...=\frac{1}{1-x}$ which is valid as $|\frac{1}{b^k}| \lt 1$ for the given range of $b$.

$\endgroup$
  • $\begingroup$ Thank you very much ! $\endgroup$ – Harmonic Sun Feb 2 at 17:46
4
$\begingroup$

We have \begin{align*} u_n &= \sum_{k=1}^{n} \sum_{r=0}^\infty \frac{(-b^{-k})^r}{r!} \\ &= \sum_{r=0}^\infty \sum_{k=1}^{n} \frac{(-b^{-k})^r}{r!} \\ &= \sum_{r=0}^\infty \frac{(-1)^{r+1}}{r!}\frac{b^{-nr}-1}{b^r-1} \\ &= n + \frac{b^{-n}-1}{b-1} + \cdots \end{align*}

$\endgroup$
  • $\begingroup$ This gives a more accurate expansion than mine, but for an asymptotic expansion can we not remove the $b^{-n}$ terms as they will become zero? $\endgroup$ – Peter Foreman Feb 2 at 17:28
  • 1
    $\begingroup$ You're absolutely right, and your answer probably deserves more credit, because I answered effectively the same way, although maybe a few seconds too late haha. I wanted to be as general as possible, for then the OP can determine however s/he wanted to handle the defect terms $b^{-nr}$. $\endgroup$ – Tom Chen Feb 2 at 17:33

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.