Prove that

if \begin{align} &0<a_1,a_2,\dots,a_n \in \mathbb R, &\left(\sum_{k=1}^n a_k\right)\left(\sum_{k=1}^n \frac1{a_k}\right)\le\left(n+\frac12\right)^2\\ \end{align}

then $$\max_k \space a_k \le 4\times\min_k\space a_k$$

What I've tried was $$\left(\sum_{k=1}^n a_k\right)\left(\sum_{k=1}^n \frac1{a_k}\right)=n+\sum_{i\ne j}\frac{a_i}{a_j}=n+\sum_{i< j}\left(\frac{a_i}{a_j}+\frac{a_j}{a_i}\right)$$ which didn't help at all. Thanks.

  • 2
    $\begingroup$ Are you sure about the $\min$ and the $\max$? Shouldn't they be swapped? $\endgroup$
    – Clement C.
    Dec 28 '16 at 21:41
  • $\begingroup$ they probably should or it's not giving much information $\endgroup$ Dec 28 '16 at 21:47
  • $\begingroup$ Sorry guys, yeah those are swapped. I'll fix now. @ClementC. $\endgroup$
    – Kay K.
    Dec 28 '16 at 21:49
  • $\begingroup$ @KayK. have you tried rewriting $\sum_{i\neq j}\frac{a_i}{a_j}$ by doing common denominator and see what happens? Like patterns etc? $\endgroup$ Dec 28 '16 at 21:50
  • 1
    $\begingroup$ See also here: math.stackexchange.com/questions/1408802/… $\endgroup$ Dec 28 '16 at 22:54

Assume that $a_1\le\ldots\le a_n$. Suppose the contrary: $a_n>4a_1$.

Let $S$ be the sum $$S=\sum_{1\le i<j\le n}\left(\frac{a_i}{a_j}+\frac{a_j}{a_i}\right)$$

Let's split this sum in three "parts" to bound them below:

The first, $S_1$ is just $a_1/a_n+a_n/a_1$.


$S_2$ is the sum of the fractions where occurs $a_n$ or $a_1$, but not both:

$$\begin{align} S_2&=\sum_{k=2}^{n-1}\left(\frac{a_k}{a_1}+\frac{a_k}{a_n}+\frac{a_1}{a_k}+\frac{a_n}{a_k}\right)\\ &=\left(1+\frac{a_1}{a_n}\right)\sum_{k=2}^{n-1}\left(\frac{a_k}{a_1}+\frac{a_n}{a_k}\right)\\ &\stackrel*\ge2(n-2)\left(1+\frac{a_1}{a_n}\right)\left(\frac{a_n}{a_1}\right)^{1/2}\\ &=2(n-2)\left[\left(\frac{a_n}{a_1}\right)^{1/2}+\left(\frac{a_1}{a_n}\right)^{1/2}\right]\\ &>\frac52\cdot 2(n-2)\ge 5(n-2) \end{align}$$

Last, $S_3$ is the sum of the remaining terms:

$$S_3=\sum_{2\le i<j\le n-1}\left(\frac{a_i}{a_j}+\frac{a_j}{a_i}\right)\ge2\binom{n-2}2=n^2-5n+6$$

So $$S>\frac{17}4+5n-10+n^2-5n+6=n^2+\frac14$$

and then $$n+S>\left(n+\frac12\right)^2$$

Remarks: Throughout the proof I have been using two facts:

  • The function $f(x)=x+x^{-1}$ for $x>1$ is strictly increasing.
  • The sum of $r$ positive numbers whose product is $1$ is at least $r$. (This follows easily from AM-GM inequality).

For the inequality marked with $*$ I have used AM-GM inequality, too.


Maybe the following reasoning can help.

Let $f(a_1,a_2,...a_n)=\sum\limits_{k=1}^n a_k\sum\limits_{k=1}^n \frac1{a_k}$, $\max_{i}a_i=M$, $\min_{i}a_i=m$ and $M=xm$.

We need to prove that $x\leq4$.

Indeed, let $x>4$.

Easy to see that $\frac{\partial^2f}{\partial a_i^2}>0$ for all $i$, which says that $\max\limits_{m\leq a_i\leq M}f=\max\limits_{a_i\in\{m,M\}}f$.

Let $m$ gotten $k$ times.

Hence, $(km+(n-k)M)\left(\frac{k}{m}+\frac{n-k}{M}\right)\leq\left(n+\frac{1}{2}\right)^2$ or $k^2+(n-k)^2+k(n-k)\left(x+\frac{1}{x}\right)\leq\left(n+\frac{1}{2}\right)^2$ and since $x+\frac{1}{x}>\frac{17}{4}$, we obtain $$9k(n-k)<4n+1,$$ which is strange.

  • 1
    $\begingroup$ Some question. I think what you showed was that there is no case where there are $m$'s $k$ times and $M$'s $n-k$ times for any $0\le k\le n$ if $x>4$. But, I am not sure whether it proves that $x>4$ is not possible in any other case where other $a_k$'s are in between $m$ and $M$... $\endgroup$
    – Kay K.
    Dec 28 '16 at 22:54

Another way: When you see "reversals" of CS inequality, often Pólya-Szegö’s inequality is useful.

With $\min a_k = m, \max a_k = Cm$, using Pólya-Szegö, we have $$\sum a_k \sum \frac1{a_k} \geqslant \frac{n^2}4\left(C +\frac{1}C \right)^2$$

$$\therefore \frac{n}2\left(C + \frac1C\right) \leqslant n+\frac12 \implies C + \frac1C \leqslant 2 + \frac1n < 3 \implies C < \frac12(1+\sqrt5) < 4$$ We can strictly bound this as for $n=1$ trivially $C=1$. In fact the above establishes $C < 1.62$.


Hint: \begin{align*} \min_j a_j &\leq a_k \quad \text{ for every } 1 \leq k \leq n\\ \sum_{k = 1}^n \min_j a_j &\leq \sum_{k = 1}^n a_k\\ n \, \min_j a_j &\leq \sum_{k = 1}^n a_k \end{align*} Similarly $$ n \frac{1}{\max\limits_j a_j} \leq \sum_{k = 1}^n \frac{1}{a_k} $$ Hence $$ n^2 \frac{\min\limits_j a_j}{\max\limits_i a_i} \leq \left( \sum_{k = 1}^n a_k \right)\left( \sum_{k = 1}^n \frac{1}{a_k} \right) \leq \left( n + \frac{1}{2} \right)^2 \leq \ldots $$ This should lead to something. Also, obviously $\frac{\max\limits_i a_i}{\min\limits_j a_j} \geq 1$. (I like having different indeces $i, j$ but these are of course dummy variables)


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.