Let $\{a_n\}$ be a decreasing sequence and $a_n>0$ for all $n$. If $\displaystyle\lim_{n\to\infty}\frac{a_{n+1}}{a_n}=1$ and $\displaystyle\lim_{n\to\infty}\frac{a_{2n}}{a_n}=\frac{1}{2}$, how to prove or disprove that $\displaystyle\lim_{n\to\infty}\frac{a_{3n}}{a_n}=\frac{1}{3}$ ?

Thank you.

  • 6
    $\begingroup$ What did you try and what is your progress so far? $\endgroup$ – Sungjin Kim Mar 20 '17 at 16:56
  • $\begingroup$ Is it true that powers of 2 and powers of 3 come arbitrarily close to each other? $\endgroup$ – Aritro Pathak Mar 20 '17 at 18:13
  • 1
    $\begingroup$ You can't disprove this since $a_n = n^{-1}$ exhibits an explicit example of $\lim_{n \to \infty} a_{3n}/a_n = 1/3$, so the only options remaining is to 1。 prove it, or 2. show that $\lim_{n \to \infty} a_{3n}/a_n = 1/3$ may or may not hold by exhibiting another example. $\endgroup$ – Henricus V. Mar 20 '17 at 20:06
  • $\begingroup$ I think if in the sequence $a_n$, we had $\displaystyle\lim_{n\to\infty}\frac{a_{n+1}}{a_n}=1$, it is not possible that $\displaystyle\lim_{n\to\infty}\frac{a_{2n}}{a_n}=\frac{1}{2}$ because: $\endgroup$ – Amin235 Mar 20 '17 at 20:44
  • $\begingroup$ \begin{eqnarray} \displaystyle\lim_{n\to\infty}\frac{a_{2n}}{a_n}&=& \displaystyle\lim_{n\to\infty}\frac{a_{n+1}}{a_n}\times \frac{a_{n+2}}{a_{n+1}}\times \cdots \times \frac{a_{2n}}{a_{2n-1}} \\ && \\ &=&\displaystyle\lim_{n\to\infty}\frac{a_{n+1}}{a_n}\times \displaystyle\lim_{n\to\infty}\frac{a_{n+2}}{a_{n+1}}\times \cdots \times \displaystyle\lim_{n\to\infty}\frac{a_{2n}}{a_{2n-1}}\\ &&\\ &=&1\times 1\times \cdots \times 1\\ &&\\ &=&1 \end{eqnarray} $\endgroup$ – Amin235 Mar 20 '17 at 20:44

The limit of $\frac{a_{3n}}{a_n}$ need not be $\frac13$, because it need not exist.

Each $n>0$ can be written uniquely as $2^k + r$, where $2^k \le n < 2^{k+1}$. Define $a_n$ (in terms of these $k$ and $r$) piecewise as follows: $$a_n = \begin{cases} \frac{2^{1 - r/2^{k-1}}}{2^k} & \mbox{if }r < 2^{k-1} \\ \frac{1}{2^k} & \mbox{if }r \ge 2^{k-1}. \end{cases}$$

Essentially, for the first half of the range from $2^k$ to $2^{k+1}$, $a_n$ decreases from $\frac{1}{2^{k-1}}$ to $\frac1{2^k}$ geometrically, by factors of $2^{-1/2^{k-1}}$. For the second half of that range, $a_n$ stays at $\frac1{2^k}$.

It is identically true that $\frac{a_{2n}}{a_n} = \frac12$. Moreover, if $n \ge 2^k$, then $2^{-1/2^{k-1}} \le \frac{a_{n+1}}{a_n} \le 1$, so $\frac{a_{n+1}}{a_n} \to 1$.

However, when $n = 2^k$, $\frac{a_{3n}}{a_n} = \frac14$, while $\frac{a_{9n}}{a_{3n}} = \frac1{2^{5/4}}$, so $\frac{a_{3n}}{a_n}$ does not converge.

You might complain that the sequence $a_n$ is not strictly decreasing. If this is a problem, just replace $a_n$ by $a_n' = a_n\left(1 + \frac1n\right)$. We have:

  • $\frac{a'_{n+1}}{a'_n} = \frac{a_{n+1}}{a_n} \cdot \frac{1+\frac1{n+1}}{1+\frac1n} = \frac{a_{n+1}}{a_n} \left(1 - \frac1{(n+1)^2}\right)$, which still converges to 1.
  • $\frac{a'_{2n}}{a'_n} = \frac{a_{2n}}{a_n} \cdot \frac{1 + \frac1{2n}}{1 + \frac1n} = \frac{a_{2n}}{a_n} \left(1 - \frac1{2n+2}\right)$, which still converges to $\frac12$.
  • $\frac{a'_{3n}}{a'_n} = \frac{a_{3n}}{a_n} \cdot \frac{1 + \frac1{3n}}{1 + \frac1n} = \frac{a_{3n}}{a_n} \left(1 - \frac2{3n+3}\right)$, which still does not converge to anything.
  • Since we already had $a_{n+1} \le a_n$, we now have $a'_{n+1} = a_{n+1} \left(1 + \frac1{n+1}\right) \le a_n \left(1 + \frac1{n+1}\right) < a_n \left(1 + \frac1n\right) = a'_n$.

If the limit $\frac{a_{3n}}{a_n}$ does exist, then we may proceed as in Andras's answer to show that it must equal $\frac13$:

If $\frac{a_{3n}}{a_n} \to c$, then $\frac{a_{3^kn}}{a_n} \to c^k$ for any $k$, but we can bound $\frac{a_{3^kn}}{a_n}$ between $\frac{a_{2^\ell n}}{a_n}$ and $\frac{a_{2^{\ell+1} n}}{a_n}$ for $\ell$ such that $2^\ell < 3^k < 2^{\ell+1}$ (i.e., $\ell = \lfloor k \log_2 3\rfloor$). These two ratios converge to $\frac{1}{2^{\ell}}$ and $\frac1{2^{\ell+1}}$, so we get that $2^{-\ell/k} < c < 2^{-(\ell+1)/k}$. Taking $k$ arbitrarily large, $\frac{\ell}{k} \to \log_2 3$, so $c$ must be $2^{-\log_2 3} = \frac13$.

| cite | improve this answer | |
  • $\begingroup$ You say "multiply $a_n$ by a sequence that decays to $1$ sufficiently slowly for this not affect convergence of any of the ratios we care about." But you do not need just the convergence of the ratios, you also need the product to be strictly decreasing. I am convinced by all your argument except that part, which is still unclear. $\endgroup$ – Ewan Delanoy Mar 20 '17 at 20:39
  • $\begingroup$ Take $a_n' = a_n (1 - \epsilon_n)$, where $\epsilon_n$ is strictly decreasing and $\epsilon_n \to 0$ very very slowly. We already had $a_{n+1} \le a_n$; now we have $a'_{n+1} \le a_n' \cdot \frac{1-\epsilon_n}{1-\epsilon_{n+1}}$, and $\frac{1-\epsilon_n}{1-\epsilon_{n+1}} < 1$. $\endgroup$ – Misha Lavrov Mar 20 '17 at 20:49
  • $\begingroup$ But you've convinced me that it's worth editing the post to make this argument explicit. $\endgroup$ – Misha Lavrov Mar 20 '17 at 20:51
  • $\begingroup$ Thanks for your explanation. I've found a variation of your argument that I prefer : start by putting $a_{2^n}=\frac{1}{2^n}$, $a_{32^n}=b_n\frac{1}{2^{n-1}}$, where $(b_n)$ is any sequence taking values in $[\frac{1}{4},\frac{1}{2}]$. Then you can fill the blanks (taking $\frac{a_{2n}}{a_n}$ to be identically $\frac{1}{2}$ as you do), to satsify all the conditions. $\endgroup$ – Ewan Delanoy Mar 20 '17 at 20:53
  • $\begingroup$ Yeah, I started out thinking along those lines, but then decided to give an explicit formula - I'm not sure that made things any clearer. (Also, my $\epsilon_n$ argument in my comment above got its signs confused, so I apologize for that, but I'm fairly sure the argument in my answer is correct.) $\endgroup$ – Misha Lavrov Mar 20 '17 at 21:06

This is not an answer, but a longer train of thoughts / conjectures that might lead to a full answer.

It might be an idea to use $\displaystyle\lim_{n\to\infty}\frac{a_{2n}}{a_n}=\frac{1}{2}$ and apply it $r$ times. This gives $$ \displaystyle\lim_{n\to\infty}\frac{a_{2^rn}}{a_n}=\frac{1}{2^r} $$ which holds for all $r$. Now take some $m$ and identify $r$ such that $2^r \leq m < 2^{r+1}$. Then one has $$ \frac{1}{2^r} \geq \displaystyle\lim_{n\to\infty}\frac{a_{m n}}{a_n} > \frac{1}{2^{r+1}} $$

In particular, this holds when $m$ is chosen to be any power $3^p$. This gives a countable infinite set of inequalities of the type above, all of which must hold. They do hold if $$ \displaystyle\lim_{n\to\infty}\frac{a_{3 n}}{a_n}=\frac{1}{3} $$ implying $$ \displaystyle\lim_{n\to\infty}\frac{a_{3^p n}}{a_n}=\frac{1}{3^p} $$

The conjecture is that there is no other way that they can hold.

| cite | improve this answer | |
  • $\begingroup$ Pretty much the same as Vik78 : I downvote, and will reverse my vote if the author inserts the mention "this is not an answer but only a long comment" $\endgroup$ – Ewan Delanoy Mar 20 '17 at 19:07
  • 2
    $\begingroup$ The author says it might be and calls it a conjecture. What more do you need? $\endgroup$ – marty cohen Mar 20 '17 at 19:16
  • $\begingroup$ @martycohen I need a clear statement that this is not an answer. Just putting "It might be" or "conjecture" at some place in the text means nothing. It took me some time to realize that this wasn't an answer, and a reader with a moderate understanding of mathematics might easily mistake it for one. $\endgroup$ – Ewan Delanoy Mar 20 '17 at 20:28
  • $\begingroup$ This is not an answer. I put it also in the text. I'm sorry that longer train of thoughts / conjectures are not welcome. $\endgroup$ – Andreas Mar 21 '17 at 6:38
  • $\begingroup$ @Andreas Longer train of thoughts/ conjectures are most welcome with me. It's just that to keep a quality standard here on MSE, I think we should distinguish clearly between a comment however elaborate and an answer however quickly sketched. Answerers often make sketchy answers, either because of lack of time or because they want to force the asker to do some work. Thank you for agreeing with me, I just reversed my vote. $\endgroup$ – Ewan Delanoy Mar 21 '17 at 6:56

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.