0
$\begingroup$

Let $a_1, a_2, ..., a_9$ be a sequence of numbers satisfying $0<p \le a_i \le q$ for each $i = 1,2,...,9.$ Prove that $$\frac{a_1}{a_9} + \frac{a_2}{a_8}+...\frac{a_9}{a_1} \le 1 + \frac{4(p^2+q^2)}{pq} .$$

I'm having quite a hard time with this one. Writing out everything on the LHS of the inequality given above, we have $$\frac{a_1}{a_9} + \frac{a_2}{a_8}+ \frac{a_3}{a_7} + \frac{a_4}{a_6}+ \frac{a_5}{a_5}+ \frac{a_6}{a_4}+ \frac{a_7}{a_3}+ \frac{a_8}{a_2}+ \frac{a_9}{a_1},$$ which is equivalent to $$\frac{a_1}{a_9} + \frac{a_2}{a_8}+ \frac{a_3}{a_7} + \frac{a_4}{a_6}+ 1+ \frac{a_6}{a_4}+ \frac{a_7}{a_3}+ \frac{a_8}{a_2}+ \frac{a_9}{a_1}.$$ Supposing the sequence increases monotonically, to the left of 1 all terms are closer to $\frac{p}{q}$ and to the right of 1 all terms are closer in value to $\frac{q}{p}$. Therefore, the long list of fractions above is approximately equivalent to $1 + 4(\frac{p}{q} + \frac{q}{p}) = 1+\frac{4(p^2+q^2)}{pq}$. While I am underestimating on the left of $1$, I am overestimating on the right of $1$. But since clearly $\frac{q}{p} > 1$, the overall expression $1+\frac{4(p^2+q^2)}{pq}$ will be greater than the actual total sum.

If the sequence were decreasing monotonically, we could still use a similar approach and end up with the same answer. However, I'm worried about my solution because I can't seem to justify my assumption that the sequence is either monotonically increasing/decreasing — please help!

$\endgroup$
  • 3
    $\begingroup$ It's easy enough to show that $\frac{a_i}{a_{10-i}}+\frac{a_{10-i}}{a_{i}} \leq \frac{p}{q}+\frac{q}{p}$ by looking at the derivative. Proceed from there. $\endgroup$ – Anon Jun 18 '18 at 16:29
  • $\begingroup$ @W.Z. sorry if this should be obvious but the derivative of what? $\endgroup$ – jjhh Jun 18 '18 at 16:46
  • $\begingroup$ Is it right that the sequence of numbers is not said to be increasing or decreasing? $\endgroup$ – Andreas Jun 18 '18 at 16:47
  • $\begingroup$ @Andreas — that's why I'm not sure if my solution is valid... $\endgroup$ – jjhh Jun 18 '18 at 16:49
  • 1
    $\begingroup$ Let $f(x)=x + \frac{1}{x}$ on the domain $[1, \frac{q}{p}]$. It's clearly maximized when $x=\frac{q}{p}$. $\endgroup$ – Anon Jun 18 '18 at 16:52
1
$\begingroup$

Let $n$ be an arbitrary positive integer. Define $$f\left(a_1,a_2,\ldots,a_n\right):=\sum_{i=1}^n\,\frac{a_i}{a_{n+1-i}}$$ for $a_1,a_2,\ldots,a_n\in [p,q] \subseteq (0,\infty)$. Observe that $\frac{p}{q}\leq \frac{a_i}{a_{j}}\leq \frac{q}{p}$ for every $i,j\in\{1,2,\ldots,n\}$. Therefore, for $i=1,2,\ldots,\left\lfloor\frac{n}{2}\right\rfloor$, $$\frac{q}{p}\left(\frac{a_i}{a_{n+1-i}}-\frac{p}{q}\right)\left(\frac{a_{n+1-i}}{a_i}-\frac{p}{q}\right)\geq 0\text{ or }\frac{a_i}{a_{n+1-i}}+\frac{a_{n+1-i}}{a_i}\leq \frac{p}{q}+\frac{q}{p}\,.$$ Hence, $$f\left(a_1,a_2,\ldots,a_n\right)\leq \epsilon_n+\sum_{i=1}^{\left\lfloor\frac{n}{2}\right\rfloor}\,\left(\frac{p}{q}+\frac{q}{p}\right)=\epsilon_n+\left\lfloor\frac{n}{2}\right\rfloor\left(\frac{p}{q}+\frac{q}{p}\right)\,,$$ where $\epsilon_n:=0$ for even $n$, and $\epsilon_n:=1$ for odd $n$. The maximum value is achieved if and only if $$\left\{a_i,a_{n+1-i}\right\}=\{p,q\}$$ for every $i=1,2,\ldots,\left\lfloor\frac{n}{2}\right\rfloor$.


Interesting, Related Problem

Let $n$ be a positive integer, and $\sigma$ a permutation on the set $\{1,2,\ldots,n\}$. Define $$f_\sigma\left(a_1,a_2,\ldots,a_n\right):=\sum_{i=1}^n\,\frac{a_i}{a_{\sigma(i)}}\,,$$ for $a_1,a_2,\ldots,a_n\in [p,q]\subseteq (0,\infty)$. Show that the maximum possible value of $f$ is $$\sum_{j=1}^k\,\Biggl(\epsilon_{t_k}+\left\lfloor\frac{t_k}{2}\right\rfloor\left(\frac{p}{q}+\frac{q}{p}\right)\Biggr)=n+\left(\sum_{j=1}^k\,\left\lfloor\frac{t_k}{2}\right\rfloor\right)\,\frac{(p-q)^2}{pq}\,,$$ where $\epsilon_{r}:=0$ for each even integer $r$ and $\epsilon_r:=1$ for each odd integer $r$. Here, $t_1,t_2,\ldots,t_k$ are the lengths of the cycles in the decomposition of $\sigma$ into a product of disjoint cyclic permutations. The OP's problem is a special case of my generalized version where $n=9$ and $$\sigma=(1\;\;9)(2\;\;8)(3\;\;7)(4\;\;6)(5)\,,$$ so that $k=5$, $t_1=t_2=t_3=t_4=2$, and $t_5=1$

$\endgroup$
0
$\begingroup$

The hint.

Let $f(a_1,a_2,...,a_9)=\sum\limits_{k=1}^9\frac{a_k}{a_{k-1}}$, where $a_0=a_9$.

Thus, $$\frac{\partial^2f}{\partial a_k^2}>0,$$ which says that $f$ is a convex function of $a_k$ for all $k$.

Thus, $$\max_{a_k\in[p,q]}f=\max_{a_k\in\{p,q\}}f$$

$\endgroup$

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.