# Prove that $\sum_{k=1}^\infty \left(\frac{\sqrt{k}-1}{\sqrt{k}}\right)^k$ converges using a comparison

I have been asked to prove that \begin{align*} \sum_{k=1}^\infty \left(\frac{\sqrt{k}-1}{\sqrt{k}}\right)^k \end{align*} converges. I beleive that I was able to do it using a logarithm test with two applications of L'Hopital, but I have been given a hint that it can be done easily with a comparison. However, I am at a loss for which what I can compare it to. Any help would be greatly appreciated.

$$\ln(1+x)\le x$$ for all $$x\ge-1$$ so $$k^2\left(\frac{\sqrt k-1}{\sqrt k}\right)^k=k^2\left(1-\frac{1}{\sqrt k}\right)^k=k^2\exp\left(k\ln\left(1-\frac{1}{\sqrt k}\right)\right)\le k^2\textrm e^{-\sqrt k}\underset{k\to+\infty}{\to}0,$$ hence $$\sum_{k=1}^{+\infty}\left(\frac{\sqrt k-1}{\sqrt k}\right)^k<+\infty$$.

• Where did your k^2 term come from? – Walt Aug 6 at 16:46
• I use that if $k^2a_k\to0$ when $k\to+\infty$, then $\sum_{k}a_k<+\infty$. – Will Aug 6 at 16:51

Hint $$(1-\frac{1}{\sqrt{k}})^{k}=((1-\frac{1}{\sqrt{k}})^{\sqrt{k}})^{\sqrt{k}}\rightarrow e^{-\sqrt{k}}$$

• This seems like poor reasoning. Using the same type of reasoning I could say that $(1-\frac{1}{n})^n\rightarrow 1^n$ which is certainly false. – Walt Aug 6 at 16:47
• A sloppy answer. C'mon... – Wlod AA Aug 7 at 4:21
• The result is correct. – marty cohen Aug 8 at 23:53

$$\begin{array}\\ \dfrac{\sqrt{k}}{\sqrt{k}-1} &=\dfrac{\sqrt{k}-1+1}{\sqrt{k}-1}\\ &=1+\dfrac{1}{\sqrt{k}-1}\\ \text{so}\\ \left(\dfrac{\sqrt{k}}{\sqrt{k}-1}\right)^k &=\left(1+\dfrac{1}{\sqrt{k}-1}\right)^k\\ &=\left(\left(1+\dfrac{1}{\sqrt{k}-1}\right)^{\sqrt{k}}\right)^{k/\sqrt{k}}\\ &>e^{k/\sqrt{k}} \qquad\text{since } (1+\frac1{x-1})^x > e\\ &=e^{\sqrt{k}}\\ &\gt \dfrac{\sqrt{k}^m}{m!} \qquad\text{for any } m \ge 1\text{ (from power series for } e^x)\\ &\ge \dfrac{k^2}{24} \qquad\text{choosing } m=4\\ \text{so}\\ \sum_{k=1}^n \left(\frac{\sqrt{k}-1}{\sqrt{k}}\right)^k &<\sum_{k=1}^n e^{-\sqrt{k}}\\ &<\sum_{k=1}^n \dfrac{24}{k^2}\\ \end{array}$$

and this sum converges.

Start by considering the function:

$$T(k) \equiv \left(1-\frac{1}{\sqrt{k}}\right)^k \exp(\sqrt{k}) \quad \quad \quad \text{for all real } k>0.$$

With a bit of work (which I will leave for you), you can show that $$T'(k)>0$$, so the function $$T$$ is strictly increasing. Since $$T(k) \rightarrow 1$$ this means that the function increases up to this limiting bound (which it never attains), and so for all $$k \in \mathbb{N}$$ this function is bounded by $$T(k) < 1$$. Using this inequality we obtain:

\begin{aligned} \text{Your Sum} = \sum_{k=1}^\infty T(k) \cdot \exp(-\sqrt{k}) < \sum_{k=1}^\infty \exp(-\sqrt{k}). \end{aligned}

Now, since $$\exp(-\sqrt{k})$$ is a decreasing function, for all $$k \in \mathbb{N}$$ we have:

\begin{aligned} \exp(-\sqrt{k}) = \int \limits_{k-1}^{k} \exp(-\sqrt{k}) dr < \int \limits_{k-1}^{k} \exp(-\sqrt{r}) dr. \end{aligned}

We therefore have:

\begin{aligned} \text{Your Sum} < \sum_{k=1}^\infty \exp(-\sqrt{k}) &< \sum_{k=0}^\infty \int \limits_{k}^{k+1} \exp(-\sqrt{r}) dr \\[6pt] &= \int \limits_0^\infty \exp(-\sqrt{r}) dr \\[6pt] &= \Bigg[ - 2(1+\sqrt{r}) \exp(-\sqrt{r}) \Bigg]_{r=0}^{r \rightarrow \infty} \\[6pt] &= \Bigg[ 0 - ( - 2 ) \Bigg] = 2 < \infty. \\[6pt] \end{aligned}

• T is decreasing and $e^{-\sqrt{k}} > e^{-k}$. – marty cohen Aug 8 at 23:49
• @martycohen: Thanks for pointing our the error. I have re-worked the proof to correct. – Reinstate Monica Aug 9 at 1:37

Let's use $$1+x\le e^x$$.

Then $$\frac{\sqrt{k}-1}{\sqrt{k}} = 1- \frac{1}{\sqrt{k}} \le e^{-\frac{1}{\sqrt{k}}}.$$

Take to the power $$k$$, to obtain

$$(\frac{\sqrt{k}-1}{\sqrt{k}})^k\le e^{-\sqrt{k}}.$$