How to prove that if $\sum _{n=1}^{\infty }a_n\:$ converges then $\sum _{n=1}^{\infty }a_na_{2n}\:$ converges?

Note: $a_n \in \mathbb R$

I tried to prove it using cauchy criterion The idea was to use the partial sums.

let $\epsilon$>0

There exists $N_1$ such that for any $n>N_1$ and $p\ge1$

$$\left|\sum \:\:\:\:_{k=n+1}^{n+p}a_k\:\right|<\epsilon \:$$ There exists $N_2$ such that for any $n>N_2$ and $p\ge1$

$$\left|\sum \:\:\:\:_{k=n+1}^{n+p}a_{2k}\:\right|<\epsilon \:$$ We will take $N=max\left\{N_1,N_2\right\}$ and then for any $n>N$ and $p\ge1$:

$$\left|\sum \:\:_{k=n+1}^{n+p}a_ka_{2k}\:\right|\le \left|\sum \:\:\:_{k=n+1}^{n+p}\left(max\left\{a_k,a_{2k}\right\}\right)^2\:\right|\le ...<\epsilon $$

But I didn't manage to show the part with the three dots. Maybe it's not the way to prove this. Can I get help please ?

  • $\begingroup$ @zyx can you expand on that? $\endgroup$ – user290300 Aug 26 '16 at 22:12
  • $\begingroup$ The answers have already disposed of the question, so I did not pursue it. @madmatician $\endgroup$ – zyx Aug 27 '16 at 13:06

There might be a problem. Let $\gamma=-\frac{1}{2}+i\frac{\sqrt{3}}{2}$ be a non-trivial cube root of $1$. $$a_n= \frac{\gamma^n}{\sqrt{n}}$$ Then $\sum_n a_n$ converges but $$ \sum_{n\geq 1} a_n a_{2n} = \sum_{n\geq 1} \frac{\gamma^{3n}}{\sqrt{2}n}= \sum_n \frac{1}{\sqrt{2}n}=+\infty $$

For the real case set $b_n={\rm Re\;} a_n$ then $\sum_n b_n$ converge. However, $${\rm Re} (\gamma^n) {\rm Re} (\gamma^{2n})= ){\rm Re} (\gamma^n) {\rm Re} (\gamma^{-n}) =({\rm Re} (\gamma^n))^2 \geq 1/4$$ so $\sum_n b_n b_{2n}\geq \frac{1}{4\sqrt{2}} \sum_n \frac{1}{n}=+\infty$.

  • $\begingroup$ Note that it's true via Dirichlet test when $a_{2n}$ is decreasing (and real). $\endgroup$ – Gabriel Romon Aug 26 '16 at 22:17
  • $\begingroup$ Nicely done! What if the requirement was $a_n \in \mathbb R$? (I have been thinking about a counterexample but failed at finding one) $\endgroup$ – user290300 Aug 26 '16 at 22:17
  • $\begingroup$ @madmatician The statement in my question is true. It was at my final exam yesterday. $a_n \in \mathbb R$ $\endgroup$ – idan di Aug 26 '16 at 22:19
  • 3
    $\begingroup$ For a real example, take $a_n = \sin(2 n \pi/3)/\sqrt{n}$. $\endgroup$ – Robert Israel Aug 26 '16 at 22:23

As H.H. Rugh's answer points out, the result doesn't hold in general. Here is a proof for the case $a_n>0$:

First note that $a_na_{2n}\leq \frac{a_n^2+a_{2n}^2}{2}$ for all $n$, and $0<a_n^2<a_n$ for all sufficiently large $n$ since $a_n\to 0$, hence $$ \sum_{n=1}^{\infty}a_na_{2n}\leq \frac{1}{2}\sum_{n=1}^{\infty}a_n^2+\frac{1}{2}\sum_{n=1}^{\infty}a_{2n}^2\ll \sum_{n=1}^{\infty}a_n+\sum_{n=1}^{\infty}a_{2n}\ll \sum_{n=1}^{\infty}a_n<\infty$$

  • 4
    $\begingroup$ No need for $a_n>0$. Just absolute convergence will do $\endgroup$ – user290300 Aug 26 '16 at 22:21
  • $\begingroup$ Sure, that works too. $\endgroup$ – carmichael561 Aug 26 '16 at 22:21
  • $\begingroup$ @carmichael561 How can the statement be a false statement ? I had it at my exam yesterday. We dont know if $a_n$ is positive for all $n$ $\endgroup$ – idan di Aug 26 '16 at 22:23
  • 4
    $\begingroup$ It is not unheard of for exams to contain mistakes. $\endgroup$ – Robert Israel Aug 26 '16 at 22:24
  • 3
    $\begingroup$ Also note that since $a_n\to 0,$ $|a_na_{2n}| \le |a_n|$ for large $n.$ $\endgroup$ – zhw. Aug 26 '16 at 22:24


$$ a_{3k}=0, a_{3k+1}=\frac{1}{\sqrt{3k+1}}, a_{3k+2}=-\frac{1}{\sqrt{3k+2}} $$ then $\sum a_{n}$ converges but not $\sum a_{n}a_{2n}$.


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.