6
$\begingroup$

Suppose $\sum_{n=1}^\infty \frac{1}{a_n} = A$ is summable, with $a_n > 0,$ $n = 1,2,3,\cdots.$ How can we prove that $\sum_{n=1}^\infty \frac{n}{a_1 + \dots + a_n}$ is also summable?

This question came from a problem-solving seminar, but I'm quite stuck without a push in the right direction. I tried a few things, including Cauchy-Schwarz (which says $\sum_{n=1}^\infty \frac{n}{a_1 + \dots + a_n} < \sum_{n=1}^\infty \frac{A}{n}$) and also the idea of assuming the latter series diverges and attempting to deduce the divergence of the former series from that, using facts such as $\sum a_n = \infty \implies \sum \frac{a_n}{a_1 + \cdots + a_n} = \infty$. Nothing has worked so far.

$\endgroup$
7
$\begingroup$

The easiest way is to use Carleman's Inequality, we have $$ \root{n}\of{a_1\cdots a_n}\leq \frac{a_1+\cdots+a_n}{n} $$ So $$ \sum_{n=1}^\infty \frac{n}{a_1+\cdots+a_n}\leq \sum_{n=1}^\infty \root{n}\of{\frac{1}{a_1}\cdots \frac{1}{a_n}} \leq e\sum_{n=1}^\infty \frac{1}{a_n}. $$ So the convergence of $\sum\dfrac{1}{a_n}$ implies that of $\sum\dfrac{n}{a_1+\cdots +a_n}.$

$\endgroup$
3
$\begingroup$

Here I propose an other proof that yields the best constant. Using Cauchy-Schwarz inequality we have: $$\sum_{j=1}^k j=\sum_{j=1}^k\sqrt{j^2a_j}\cdot{1\over\sqrt{a_j}}\leq \sqrt{\sum_{j=1}^k j^2a_j}\times\sqrt{\sum_{j=1}^k{1\over a_j}}$$ then $${1\over\sum_{j=1}^k{1/ a_j}}\leq{4\over{k^2(k+1)^2}} \left(\sum_{j=1}^kj^2a_j\right)$$ and this can be written in the following way $${k\over\sum_{j=1}^k{1/ a_j}}\leq{4k\over{2k+1}} \left({1\over k^2}-{1\over (k+1)^2}\right)\left(\sum_{j=1}^kj^2a_j\right)$$ which implies $${k\over\sum_{j=1}^k{1/ a_j}}< 2 \left({1\over k^2}-{1\over (k+1)^2}\right)\left(\sum_{j=1}^kj^2a_j\right)$$ and can be writen in the following way $${k\over\sum_{j=1}^k{1/ a_j}}< 2a_k+A_k-A_{k+1}$$ with $A_k={2\over k^2}\sum_{j=1}^{k-1}j^2a_j$ and $ A_1=0$. Summing these inequalities for $1\leq k\leq n$ we find $$\sum_{k=1}^n{k\over\sum_{j=1}^k{1/ a_j}}< 2\left(\sum_{k=1}^n a_k\right)-A_{n+1}, $$ and this proves the desired inequality since $A_{n+1}\geq0$.

On the other hand, if $c$ is the least constant such that if $n\geq 1$ and $a_1,\ldots,a_n>0$ we have $$\sum_{k=1}^n{k\over\sum_{j=1}^k1/a_j}\leq c \sum_{k=1}^na_k$$ then, by choosing $a_n={1\over n}$ we find $$\sum_{k=1}^n{2\over k+1}\leq c \sum_{k=1}^n{1\over k}$$ that is $2\leq c+1/\sum_{k=1}^n{1\over k}$, and this proves that $2\leq c$ since $\lim_{n\to\infty}\sum_{k=1}^n{1\over k}=+\infty$.

$\endgroup$
  • $\begingroup$ (+1) This proof is a (beautiful) hybrid of the one I gave above with telescoping and the proof given here. Many thanks for sharing! $\endgroup$ – Chris Oct 19 '14 at 21:15
1
$\begingroup$

Allow me to present a short, "magical" solution by thoughtful application of Cauchy-Schwarz.

Define $A_n := a_1 + \cdots + a_n.$ We will first prove the convergence of $\sum_{n=1}^\infty \frac{n^2}{A_n^2}a_n$. Now,

\begin{eqnarray} \sum_{n=1}^N \frac{n^2}{A_n^2}a_n &<& \sum_{n=2}^N \frac{n^2}{A_nA_{n-1}}a_n + \frac{1}{{a_1}} \\ &=& \sum_{n=2}^N \frac{a_n}{A_nA_{n-1}}n^2 + \frac{1}{{a_1}} \\ &=& \sum_{n=2}^N \left( \frac{1}{A_{n-1}} - \frac{1}{A_n}\right)n^2 + \frac{1}{{a_1}} \\ &<& \sum_{n=1}^N \frac{(n+1)^2 - n^2}{A_n} + \frac{2}{a_1} \\ &=& \sum_{n=1}^N \frac{2n + 1}{A_n} + \frac{2}{a_1} = 2\sum_{n=1}^N \frac{n}{A_n} + C \end{eqnarray}

where in the last step we added the sum $\sum \frac{1}{A_n}$ to the constant term. Now,

$$\left(\sum_{n=1}^N \frac{n}{A_n}\right) \le {\left(\sum_{n=1}^N \frac{n^2}{A_n^2}a_n\right)}^{1/2}{\left(\sum_{n=1}^N \frac{1}{a_n} \right)}^{1/2} $$

by Cauchy-Schwarz, so, writing $S_N = \sum_{n=1}^\infty \frac{n^2}{A_n^2}a_n$, we will have

\begin{equation} S_N \le B \sqrt{S_N} + C \end{equation}

for all $N$. This implies convergence of $S_N$, since $\sqrt{S_N} \le B + \frac{C}{\sqrt{S_N}}$ could not hold for large $N$ if $S_N$ diverged.

From the Cauchy-Schwarz inequality derived above, we can thus immediately infer the convergence of our original series $\sum_{n=1}^\infty \frac{n}{A_n}$.

$\endgroup$

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.