Fix $a$ some positive number and $n$ some positive integer,and assume that $$x_{1}+x_{2}+\cdots+x_{n}=na,x_{i}\ge 0,i=1,2,\cdots,n$$ Find this function maximum and minimum $$f=\sum_{k=1}^{n-1}(x_{k}-a)(x_{k+1}-a)$$

My try: let $$y_{i}=\dfrac{x_{i}}{a}\Longrightarrow y_{1}+y_{2}+\cdots+y_{n}=n$$ then $$f=\dfrac{1}{a^2}\sum_{k=1}^{n-1}(y_{k}-1)(y_{k+1}-1)$$ $$\Longrightarrow f=\dfrac{1}{a^2}\left(y_{1}y_{2}+y_{2}y_{3}+\cdots+y_{n-1}y_{n}-2n+y_{1}+y_{n}+n-1\right)$$ so we only need to find the minimum and maximum value of the following $$g=y_{1}y_{2}+y_{2}y_{3}+\cdots+y_{n-1}y_{n}+y_{1}+y_{n}$$

I want to use AM-GM inequality, but I can't finish.

  • $\begingroup$ Try Lagrange multiplier technique. $\endgroup$ – Mhenni Benghorbal Feb 18 '14 at 11:47
  • 2
    $\begingroup$ can you post your Lagrange multiplier full methods?because sometimes this methods is not good to solve inequality $\endgroup$ – china math Feb 18 '14 at 11:52
  • $\begingroup$ @chinamath. Where are inequalities ? $\endgroup$ – Claude Leibovici Feb 18 '14 at 12:03
  • $\begingroup$ It is said this today University walks exam questions $\endgroup$ – china math Feb 18 '14 at 12:42
  • $\begingroup$ $0<g<\dfrac{n(n+2)}{4}$ $\endgroup$ – chenbai Feb 18 '14 at 12:53

Plug $(0,n,0,\ldots,0)$ to see $\min g= 0$.
For the maximum, consider $n=2,3$ first, where the problem is immediately solved by AM-GM. The case $n=4$ is a bit more complicated, but not too hard. In the sequel we assume that $n\ge 5$. Since $g$ is continuous and is defined on a compact set, the maximum $g_0$ is attained at some point $(y_1,y_2,\ldots,y_n)$.

Claim. $y_3=\cdots=y_{n-2}$.
Proof. Suppose $y_k\not=y_{k+1}$ for some $3\le k\le n-3$. Then $g(\ldots,\frac{y_k+y_{k+1}}2,\frac{y_k+y_{k+1}}2,\ldots)>g_0$, a contradiction.

The following statement will be used frequently:

Claim. $n^2(n-5)/(n-4)^2\le(n+1)^2/4$, for $n\ge 5$.
Proof. For $n\ge 8$ it's plain since $n-5<n-4$ and $n-4\ge 4$. For $n=5,6,7$ we may check by direct computation.

Now let $y_3=u$, then $$ g_0=(y_2+y_{n-1})u+y_1y_2+y_{n-1}y_n+y_1+y_n+(n-5)u^2 $$ If $y_2\not=y_{n-1}$ and $\min\{y_1,y_n\}\not=0$, then $(y_1,y_n)\mapsto(y_1+\delta,y_n-\delta)$ will increase $g$ for some $\delta$ (maybe negative). Hence $y_2\not=y_{n-1}\Rightarrow\min\{y_1,y_n\}=0$. Similarly $y_1\not=y_n\Rightarrow\min\{y_2,y_{n-1}\}=0$. Now there are several cases:

Case 1. $y_2=y_{n-1}, y_1=y_n$. If $y_1=y_2=0$, then $$g_0=\frac{n^2(n-5)}{(n-4)^2}$$ which is no more than $(n+1)^2/4$ established in Case 2 and Case 3, a contradiction. Otherwise, if $y_1=y_n\not=0$, then by a perturbation $(y_1,y_n)\mapsto(y_1+\delta,y_n-\delta)$ the value of $g$ leaves unchanged, and the problem is reduced to Case 3; if $y_2=y_{n-1}\not=0$, it can be similarly reduced to Case 2.

Case 2. $y_2\not=y_{n-1}$. WLOG we may assume $y_n=0$.

  • Subcase 2.1 If $y_1\not=0$, we have $y_1>y_n$, hence $y_{n-1}=0$ (otherwise $(y_2,y_{n-1})\mapsto(y_2+\delta,y_{n-1}-\delta)$ will increase $g$ for $\delta>0$). So \begin{align} g_0&=y_2u+y_1y_2+y_1+(n-5)u^2\\ &=\frac{y_2(n-y_1-y_2)}{n-4}+y_1y_2+y_1+\frac{(n-5)(n-y_1-y_2)^2}{(n-4)^2} \end{align} The last expression is a convex quadratic function in $y_1$, thus attains its maximum at $y_1=0$ or $y_1=n-y_2$. When $y_1=0$, \begin{align} &\frac{y_2(n-y_2)}{n-4}+\frac{(n-5)(n-y_2)^2}{(n-4)^2}\\ =&-\frac{(n-y_2)^2}{(n-4)^2}+\frac{n(n-y_2)}{n-4}\\ \le&\cases{(5/2)^2,\phantom{n^2(n-5)/(n-4)^2}n=5\\n^2(n-5)/(n-4)^2,\phantom{(5/2)^2}n\ge 6} \end{align} and when $y_1=n-y_2$ we have $$(n-y_2)y_2+n-y_2\le \frac{(n+1)^2}{4}$$ Now we conclude that(note that $(5/2)^2<(5+1)^2/4$) $$g_0\le\max\left\{\frac{n^2(n-5)}{(n-4)^2},~\frac{(n+1)^2}4\right\}=\frac{(n+1)^2}{4}$$

  • Subcase 2.2 If $y_1=0$, then \begin{align} g_0&=(y_2+y_{n-1})u+(n-5)u^2\\ &=\Big(n-(n-4)u\Big)u+(n-5)u^2\\ &=-u^2+nu\le\frac{n^2}{4}<\frac{(n+1)^2}4 \end{align} which is strictly smaller than Subcase 2.1, hence cannot be a maximum.

Case 3. $y_1\not=y_n$. WLOG we may assume $y_{n-1}=0$.

  • Subcase 3.1 If $y_2\not=0$, by the same reasoning in Subcase 2.1 we have $y_n=0$. Now it's reduced to Subcase 2.1, where $y_{n-1}=y_n=0$.
  • Subcase 3.2 If $y_2=0$, then \begin{align} g_0&=y_1+y_n+(n-5)u^2\\ &=n-(n-4)u+(n-5)u^2\\ &\le\max\left\{n,~\frac{n^2(n-5)}{(n-4)^2}\right\}\\ &\le\max\left\{\frac{(n+1)^2}{4},~\frac{n^2(n-5)}{(n-4)^2}\right\}\\ &=\frac{(n+1)^2}{4} \end{align}

We may conclude that for $n\ge 5$, $g_0\le(n+1)^2/4$. And in fact $g_0=(n+1)^2/4$ since $g(\frac{n+1}2,\frac{n-1}2,0,0,\ldots,0)=(n+1)^2/4$.
Combining these results with the the case $n=3,4$, we may say that the maximum is $(n+1)^2/4$ for $n\ge 3$.

P.S. It's also worth mentioning that Lagrange multiplier method, mentioned in the comment, is not well applicable here. The result shows the extremal point is not a solution to the Lagrange multiplier equation.

| cite | improve this answer | |

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.