Mathematics Stack Exchange is a question and answer site for people studying math at any level and professionals in related fields. Join them; it only takes a minute:

Sign up
Here's how it works:
  1. Anybody can ask a question
  2. Anybody can answer
  3. The best answers are voted up and rise to the top

For what values for m does $$\sum \limits_{k=2}^{\infty}\frac{1}{(\ln{k})^m}$$ converge? What about $$\sum_{k=2}^{\infty}\frac{1}{(\ln(\ln{k}))^m}$$ or more generally $$\sum_{k=2}^{\infty}\frac{1}{(\ln(\cdots (\ln{k}))\cdots)^m}$$ ?

share|cite|improve this question
Better watch where you start your sums. $\log 1=0$. – Gerry Myerson Jul 30 '12 at 8:45
thanks Gerry, done! – Arjang Jul 30 '12 at 21:48
Not quite there yet - what's $\log\log\log2$? and what about the title? – Gerry Myerson Jul 30 '12 at 23:11
I suggest defining $\ln^+ n = \max\{\ln n, 1\}$ and then replacing all $\ln$s with $\ln^+$s. – JeffE Jul 31 '12 at 2:54
up vote 7 down vote accepted

For your first series

Let $\alpha,\beta \in \mathbb{R}$ $$\sum_{k=2}^\infty \frac{1}{k^\alpha \ln(k)^\beta} $$ is convergent iif $(\alpha>1)$ or $(\alpha=1, \ \beta>1)$

So in your case it is not convergent. As

$$\frac{\sqrt{k}}{\ln(k)^\beta} \rightarrow \infty$$ we can find $k_0$ such that for all $k \geq k_0$ $$ \frac{\sqrt{k}}{\ln(k)^\beta} > 1 $$ which yields $$ \frac{1}{\ln(k)^\beta} > \frac{1}{\sqrt{k}} $$ The right-hand side behind divergent, your series is divergent. These series are called Bertrand series. See here for proofs.

As for your second series, just repeat the same argument. Take $\ln(k) > k_0$ you will get $$ \frac{1}{\ln(\ln(k))^\beta} > \frac{1}{\sqrt{\ln(k)}} $$ The right-hand side was just proven to be divergent. Repeat the process for your generalized version.

Edit: You have to be careful though about the integer that will start the series. For your first series it is obvisouly $k=2$ because $\ln(1)=0$ which yields division by zero. For the second series the division by zero occurs at $e$ and $\ln(\ln(k))$ is not defined for $k=1$ so start at $k=3=\lfloor e \rfloor+1$. For you generalized series just start at $k=\lfloor e^{e^{\cdots^e}} \rfloor +1$

share|cite|improve this answer
I guess, the OP was asking for a proof... – draks ... Jul 30 '12 at 8:51
Just added the proof. – vanna Jul 30 '12 at 8:59

For the first series,we can use Cauchy's condensation criterion when $p>0$ (when $p\leq 0$ the series is divergent):

If $\{a_n\}$ is a decreasing sequence of positive numbers, the series $\sum_ka_k$ is convergent if and only if $\sum_k2^ka_{2^k}$ is convergent.

We have with $a_k=\frac 1{(\ln k)^m}$ that $2^ka_k=\frac{2^k}{k^m(\ln 2)^m}$ and it diverges.

An alternative way is to note that $\ln k\leq k^{1/m}$ for a $m$ large enough (which depend on $m$), and use the divergence of harmonic series.

We have for $x\geq 1$ that $\log x\leq x$. Denote $f_N$ the $N$-th iterate of the logarithm. For a fixed $N$, we can find $n(N)$ such that for $k\geq n(N)$, $f_{N-1}(k)\geq 1$. Hence $f_N(k)\leq \log k$ for $k$ large enough, which shows that the series is divergent for all $m$ and all $N$.

The problem here is that the logarithm grows too slowly. Taking iterates doesn't make the thing better, and the exponent doesn't change anything.

share|cite|improve this answer

Your Answer


By posting your answer, you agree to the privacy policy and terms of service.

Not the answer you're looking for? Browse other questions tagged or ask your own question.