I need to find out whether this series converges or diverges:

$$\sum_{n=1}^\infty \frac 1{3^{\sqrt{n}}}$$

The $n$th term, ratio, and root tests are inconclusive, Abel's test doesn't apply (or I can't think of how to separate out part of the sequence), and I can't figure out a series to compare this to that'll work.

WolframAlpha says it converges by the comparison test, BTW. It just doesn't tell me what it compared the series to.

  • 4
    $\begingroup$ A good choice might be $a_n = \frac{1}{n^2}$. $\endgroup$ – Cameron Williams Apr 26 '15 at 3:53
  • $\begingroup$ How did you come up with that? $\endgroup$ – user234494 Apr 26 '15 at 3:55
  • $\begingroup$ It's one of the "simplest converging series," besides the geometric ones. Clearly, yours is not geometric, so among those who remain $\sum_n \frac{1}{n^2}$ would be a natural candidate. $\endgroup$ – Clement C. Apr 26 '15 at 3:56
  • 1
    $\begingroup$ As a way to prove (the other) inequality, you could write $n^2 = e^{2\ln n}$ and $3^{\sqrt{n}} = e^{\sqrt{n}\ln 3}$; and compare asymptotically the exponents. $\endgroup$ – Clement C. Apr 26 '15 at 4:03
  • 2
    $\begingroup$ Wouldn't you want to prove that $\dfrac{1}{3^{\sqrt n}} < \dfrac{1}{n^2}$ for large enough $n$ -- that is $n^2 < 3^{\sqrt{n}}$? $\endgroup$ – Andrey Kaipov Apr 26 '15 at 4:04

Hint: $\log_e n \leq \sqrt{n}$ for $n\geq 1$. Then $$3^{\log_e n} = n^{\log_e3}\leq 3^{\sqrt{n}}$$ This will imply, for n large enough $$\frac{1}{n^{\log_e3}}\geq \frac{1}{3^{\sqrt{n}}}$$ Now $\log_e3 > 1$.

  • $\begingroup$ Yes. $a^{\log b} = b^{\log a}$ for $a$ and $b$ positive. $\endgroup$ – Gautam Shenoy Apr 26 '15 at 5:29
  • $\begingroup$ Yes. I had initially thought that you didn't explicitly write this in your answer, but in fact it is there. $\endgroup$ – Shalop Apr 26 '15 at 5:30
  • $\begingroup$ I suppose it might be "common knowledge" that $\sqrt{n}$ eventually exceeds $\log n$. But this is mathematically just as challenging to prove as the fact that $3^{\sqrt{n}}$ eventually exceeds $n^2$, the answer from the comment thread. $\endgroup$ – alex.jordan Apr 26 '15 at 5:51

Here's something different, that doesn't require comparing $n^2$ with $3^{\sqrt{n}}$, or any similar comparison:


And the sequence $\{\left\lfloor\sqrt{1}\right\rfloor,\left\lfloor\sqrt{2}\right\rfloor,\left\lfloor\sqrt{3}\right\rfloor,\ldots\}$ is equal to $$\{\overbrace{1,1,1}^3,\overbrace{2,\ldots,2}^{5},\overbrace{3,\ldots,3}^{7},\ldots,\overbrace{k,\ldots,k}^{2k+1},\ldots\}\text{.}$$ This follows from understanding that consecutive perfect squares differ by increasing odd numbers. Or equivalently that the sum of consecutive odd integers $3+5+7+\cdots$ is always $1$ shy of a perfect square.

So $$\begin{align} \sum_{n=1}^{\infty}\frac{1}{3^{\sqrt{n}}}&<\sum_{n=1}^{\infty}\frac{1}{3^{\left\lfloor\sqrt{n}\right\rfloor}}\\ &=\sum_{k=1}^{\infty}\frac{2k+1}{3^{k}}\\ &=2\sum_{k=1}^{\infty}\frac{k}{3^{k}}+\sum_{k=1}^{\infty}\frac{1}{3^{k}}\\ &=2\cdot\frac{3}{4}+\frac{1}{2}\\ &=2 \end{align}$$

Not only is the sum convergent, it's less than $2$. You can get a better upper bound by leaving the initial terms alone instead of using the floor function. For instance, this same approach can be used with $$\begin{align}\frac{1}{3^{\sqrt{1}}}+\frac{1}{3^{\sqrt{2}}}+\frac{1}{3^{\sqrt{3}}}+\sum_{n=4}^{\infty}\frac{1}{3^{\left\lfloor\sqrt{n}\right\rfloor}} &=\frac{1}{3^{\sqrt{1}}}+\frac{1}{3^{\sqrt{2}}}+\frac{1}{3^{\sqrt{3}}}+1\\ &\approx1.69\ldots \end{align}$$ which is a better upper bound. (A CAS says the true value is approximately $1.34\ldots$)

For an even better approximation that you can't immediately tell is over or under, replace each term in $\sum_{n=N^2}^{(N+1)^2-1}\frac{1}{3^{\sqrt{n}}}$ (the portion of the series corresponding to one of the constant substrings in $\{\left\lfloor\sqrt{n}\right\rfloor\}$) with the average of the end terms: $\frac{1}{3^{\sqrt{N^2}}}$ and $\frac{1}{3^{\sqrt{(N+1)^2}}}$. So

$$\begin{align} \sum_{n=1}^{\infty}\frac{1}{3^{\sqrt{n}}} &\approx\sum_{N=1}^{\infty}\frac12\left(\frac{1}{3^N}+\frac{1}{3^{N+1}}\right)(2N+1)\\ &=\frac43 \end{align}$$ I'm not offering error analysis, but you can note that this does indeed get within $0.6\%$ of the exact value.


One way to show that this series converges can be to compare it to $\int_0^\infty (1/3)^\sqrt x dx$, which is convergent.

Edit - You can solve this integral by substituting $t=\sqrt x$, and then integrating the resulting integral by parts.

  • 2
    $\begingroup$ And how would convince someone that this integral is convergent? $\endgroup$ – alex.jordan Apr 26 '15 at 4:48
  • $\begingroup$ @alex.jordan: In general, for $n>0$ we have $\displaystyle\int_0^\infty e^{^{\LARGE-\sqrt[n]x}}~dx~=~n!~$ $\endgroup$ – Lucian Apr 26 '15 at 7:01
  • $\begingroup$ @alex.jordan, I would use substitution - $\sqrt x = t$. You can then solve the integral using integration by parts. $\endgroup$ – Roi Blumberg Apr 26 '15 at 8:02
  • $\begingroup$ @RoiBlumberg Oh good, then +1. Is it worth clarifying that in the answer? Maybe not if you are leaving something for OP to do. $\endgroup$ – alex.jordan Apr 26 '15 at 16:27
  • $\begingroup$ No, you're right, I will clarify that. $\endgroup$ – Roi Blumberg Apr 27 '15 at 16:08

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.