Prove that $\displaystyle\sum_{n=1}^\infty\frac{1}{n(1+1/2+\cdots+1/n)}$ diverges.

I think the only way to prove this is to find another series to compare using the comparison or limit tests. So far, I have been unable to find such a series.

  • 3
    $\begingroup$ How about the series $\frac{1}{n \ln n}$? Since the harmonic sum is asymptotic to $\ln n$ (more precisely, $\ln n + \gamma$, but this doesn't matter), the given series diverges iff $\sum \frac{1}{n \ln n}$ diverges. $\endgroup$ – Srivatsan Oct 9 '11 at 19:38
  • $\begingroup$ That seems appropriate, however the book has not covered this fact (the asymptotic relationship of the harmonic sum and $\ln n$). $\endgroup$ – process91 Oct 9 '11 at 19:43
  • $\begingroup$ Are you familiar with the integral test? That will tell you that $\sum_2\infty \frac{1}{n\ln n}$ diverges. It is not hard to show that $1+1/2+\cdots+1/n$ is $\lt c\ln n$ for suitable $c$, which will then give the result you want. For the integral test part, you can explicitly evaluate $\int_2^M \frac{1}{x\ln x}\,dx$. $\endgroup$ – André Nicolas Oct 9 '11 at 19:43
  • $\begingroup$ Alright, seems like that could be what the author intended, I will attempt to show that $1+1/2+\dots+1/n<c\log n$ $\endgroup$ – process91 Oct 9 '11 at 19:46

This answer is similar in spirit to Didier Piau's answer.

The following theorem is a very useful tool:

Suppose that $a_k > 0$ form a decreasing sequence of real numbers. Then $$\sum_{k=1}^\infty a_k$$ converges if and only if $$\sum_{k=1}^\infty 2^k a_{2^k}$$ converges.

Applying this to the problem in hand we are reduced to investigating the convergence of $$\sum_{k=1}^\infty \frac{1}{1 + 1/2 + \dots + 1/2^k}$$ But one easily sees that $$1 + 1/2 + \dots + 1/2^k \le 1 + 2 \cdot 1/2 + 4 \cdot 1/4 + \dots + 2^{k-1}\cdot1/2^{k-1} + 1/2^k \le k + 1.$$ Because $$\sum_{k=1}^\infty \frac{1}{k+1}$$ diverges, we are done.

  • $\begingroup$ Where can I find a proof of this theorem? Nevermind, found it - Cauchy condensation test. $\endgroup$ – process91 Oct 9 '11 at 20:15

Hints: 1. For every $k\geqslant0$, the sum of $\frac1n$ from $n=2^k$ to $n=2^{k+1}-1$ is at most $1$. 2. For every $k\geqslant0$ and every $n$ such that $2^k\leqslant n\leqslant 2^{k+1}-1$, $1+\frac12+\cdots+\frac1n\leqslant k+1$ and $\frac1n\geqslant\frac1{2^k}$. 3. For every $k\geqslant0$, the terms from $n=2^k$ to $n=2^{k+1}-1$ of the series which interests you sum to at least $\frac1{k+1}$. 4. The harmonic series diverges.


Let $H_n = 1 + \frac{1}{2} + \ldots + \frac{1}{n}$. $H_n$ is known as harmonic sum. Note that $H_{n+1} - H_n = \frac{1}{n+1}$. Now, consider $d_n = H_n - 1 - \ln(n)$. Then $d_{n+1} - d_n = \frac{1}{n+1} + \ln\left( 1 - \frac{1}{n+1} \right) < 0$, $\forall n \ge 1$ which follows from $\ln(1-x) < -x$, for $0<x<1$.

Now, since $d_{n+1} - d_n < 0$ and $d_1 = 0$, it follows that $H_n \le 1 + \ln(n)$, and thus $\sum_{n=1}^m \frac{1}{n H_n} \ge \sum_{n=1}^m \frac{1}{n( 1 + \ln(n))}$. Since the latter diverges by integral test, so does the original sum.

  • $\begingroup$ I like this method very much. I ended up using the Cauchy condensation test, but this is a very nice proof as well - thanks! $\endgroup$ – process91 Oct 10 '11 at 2:04

A possible approach thing to do is show this is greater than $$\int_{x=1}^{n+1} \frac{1}{x \log_e(x)}dx = \log_e(\log_e(n+1))$$ or some multiple of it, and show that the later diverges as $n$ increases.


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.