Let $a_{i}\in [0,1]$,and $a_{1}\ge a_{2}\ge\cdots \ge a_{n}$,show that $$a_{1}+a_{1}a_{2}+a_{1}a_{2}a_{3}+\cdots+a_{1}a_{2}\cdots a_{n}\ge \sum_{i=1}^{n}\left(\dfrac{a_{1}+a_{2}+\cdots+a_{n}}{n}\right)^i$$

For $n=2$,we have $$a_{1}+a_{1}a_{2}\ge \dfrac{a_{1}+a_{2}}{2}+\left(\dfrac{a_{1}+a_{2}}{2}\right)^2$$ $$\Longleftrightarrow 2a_{1}+4a_{1}a_{2}-2a_{2}\ge (a_{1}+a_{2})^2$$ $$2\ge a_{1}-a_{2}$$it is clear

I have try mathematical induction,Assume that $n=k$ is true,consider $n=k+1$,then we have $$a_{1}+a_{1}a_{2}+\cdots+a_{1}a_{2}\cdots a_{n}+a_{1}a_{2}\cdots a_{n+1} \ge \sum_{i=1}^{n}\left(\dfrac{a_{1}+a_{2}+\cdots+a_{n}}{n}\right)^i+a_{1}a_{2}\cdots a_{n}a_{n+1}$$ so it must prove that $$\sum_{i=1}^{n}\left(\dfrac{a_{1}+a_{2}+\cdots+a_{n}}{n}\right)^i+a_{1}a_{2}\cdots a_{n}a_{n+1}\ge \sum_{i=1}^{n+1}\left(\dfrac{a_{1}+a_{2}+\cdots+a_{n+1}}{n+1}\right)^{i}$$ then I can't so How to prove it?


More Elementary Proof (Except for the Use of Continuity)

Let $$g\left(a_1,a_2,\ldots,a_n\right):=\sum_{k=1}^n\,\prod_{j=1}^k\,a_j-\sum_{k=1}^n\,\left(\frac{1}{n}\,\sum_{j=1}^n\,a_j\right)^k$$ for all $a_1,a_2,\ldots,a_n\in[0,1]^n$ with $a_1\geq a_2\geq \ldots\geq a_n$. We need to show that, for any $i\in\{1,2,\ldots,n-1\}$, $$g\left(a_1,a_2,\ldots,a_n\right)\geq g\left(a_1,a_2,\ldots,a_{i-1},\frac{a_i+a_{i+1}}{2},\frac{a_i+a_{i+1}}{2},a_{i+2},\ldots,a_n\right)\,.$$ This is easy because if you subtract the left-hand side by the right-hand side of the inequality above, you get $$\frac{a_i-a_{i+1}}{2}-\left(\frac{a_i-a_{i+1}}{2}\right)^2\,\left(1+a_{i+2}+a_{i+2}a_{i+3}+\ldots+a_{i+2}a_{i+3}\cdots a_n\right)\,.$$ The quantity above is greater than or equal to $$\frac{a_i-a_{i+1}}{2}\left(1-\frac{1}{2}\left(1-a_{i+1}\right)\sum_{r=0}^{n-i-1}\,a_{i+1}^r\right)=\frac{a_i-a_{i+1}}{2}\left(\frac{1+a_{i+1}^{n-i}}{2}\right)\geq 0\,.$$

For any starting point $\textbf{a}^0\in[0,1]^n$, where $\textbf{a}^0:=\left(a_1^0,a_2^0,\ldots,a_n^0\right)$ satisfies $a_1^0\geq a_2^0\geq\ldots\geq a_n^0$ with $M^0:=\max\limits_{i\in\{1,2,\ldots,n-1\}}\,\left(a^0_i-a^0_{i+1}\right)$, it can be easily seen that the pair-averaging procedure described in the paragraph above can be applied to get a sequence of points $\textbf{a}^1,\textbf{a}^2,\ldots$, in such a way that the value $M^l:=\max\limits_{i\in\{1,2,\ldots,n-1\}}\,\left(a^l_i-a^l_{i+1}\right)$, where $\textbf{a}^l:=\left(a_1^l,a_2^l,\ldots,a_n^l\right)$, satisfies $$M^{l+(n-1)}\leq \frac{1}{2}\,M^l$$ for all $l=0,1,2,\ldots$. That means $\textbf{a}^l$ tends to the limit $\bar{\textbf{a}}:=\left(\bar{a},\bar{a},\ldots,\bar{a}\right)$, with $\bar{a}:=\frac{1}{n}\,\sum_{j=1}^n\,a_j^0$, as $l$ grows to infinity. By continuity of $g$, we have $$g\left(\textbf{a}^0\right)\geq g\left(\bar{\textbf{a}}\right)=0\,.$$

Alternatively, suppose that an $n$-tuple $\left(b_1,b_2,\ldots,b_n\right)\in[0,1]^n$ with $b_1\geq b_2\geq \ldots \geq b_n$ is such that $g\left(b_1,b_2,\ldots,b_n\right)$ is minimized (observe, using the compactness argument and the continuity of $g$, that this tuple must exist). By the pair-averaging argument, we conclude that $b_1=b_2=\ldots=b_n$, whence the minimum possible value of $g$ is $0$.

The above argument shows that the desired inequality is true. Furthermore, the equality case is when $a_1=a_2=\ldots=a_n$.


I am not sure if there is a simpler answer to this inequality or no; but the solution I am going to present involves Schur convexity and majorization.

Suppose that for each sequence $(a_1,\dots,a_n)\in[0,1]^n$, the sequence $(a_{[1]},\dots,a_{[n]})$ is the ordered version with $a_{[1]}\geq\dots\geq a_{[n]}$. Consider the following function: $$ f(a_1,\dots,a_n)=a_{[1]}+a_{[1]}a_{[2]}+\dots+a_{[1]}a_{[2]}\dots a_{[n]}. $$ On the other hand note that $(a_1,a_2,\dots,a_n)\succ (\overline a,\dots,\overline a)$ with $\overline a=\frac{a_1+\dots+a_n}{n}$ and $\succ$ is the majorization order.

It is enough to prove that the function $f$ is Schur convex. Because $(a_1,a_2,\dots,a_n)\succ (\overline a,\dots,\overline a)$ implies: $$ f(a_1,a_2,\dots,a_n)\geq f(\overline a,\dots,\overline a), $$ which is another way of expressing the same inequality.

Proof of Schur convexity of $f$:

To prove the Schur convexity of $f$, we use Schur-Ostrowski criterion. It is enough to verify that: $$ (a_i-a_j)(\frac{\partial f}{\partial a_i}-\frac{\partial f}{\partial a_j})\geq 0 $$ for all sequences in $[0,1]^n$.

Without loss of generality we can assume $a_1\geq \dots \geq a_n$. Also it is enough to check it for $i=1$ and $j>1$ because the verification of other pairs boils down to this case. See $$ \frac{\partial f}{\partial a_1}=1+a_2+a_2a_3+\dots+a_2a_3\dots a_n\\ \frac{\partial f}{\partial a_j}=\sum_{i=j}^n\frac{\prod_{k=1}^i a_k}{a_j} $$ and (with the convention of $\prod_{k=i}^j a_k=1$ and $\sum_{k=i}^j a_k=0$ for $j<i$): $$ \frac{\partial f}{\partial a_1}-\frac{\partial f}{\partial a_j}=\sum_{i=1}^{j-1}{\prod_{k=2}^i a_k}+\sum_{i=j}^n(\prod_{k=1}^i a_k)(\frac{1}{a_1}-\frac{1}{a_j})\\ =\sum_{i=1}^{j-1}{\prod_{k=2}^i a_k}+\prod_{k=2}^{j-1} a_k(\sum_{i=j}^n\prod_{k=j+1}^i a_k)({a_j}-{a_1})\\ =\sum_{i=1}^{j-2}{\prod_{k=2}^i a_k}+\prod_{k=2}^{j-1}a_k\left(1- (\sum_{i=j}^n\prod_{k=j+1}^i a_k)({a_1}-{a_j})\right). $$ See that the last expression is positive: $$ ({a_1}-{a_j})(\sum_{i=j}^n\prod_{k=j+1}^i a_k)\leq({1}-{a_{j+1}})(\sum_{i=j}^n\prod_{k=j+1}^i a_{j+1})=1-a_{j+1}^{n-j+1}\leq 1. $$ This proves that $f$ satisfies Schur-Ostrowski criterion and implies the inequality.

  • 1
    $\begingroup$ I think $f$ should be symmetric function, but it's not so in our case. $\endgroup$ – Michael Rozenberg Aug 22 '17 at 16:09
  • 1
    $\begingroup$ Based on my definition it is symmetric since it acts on the ordered version of the sequence $\endgroup$ – Arash Aug 22 '17 at 16:11
  • $\begingroup$ Write please your $f$ for $n=2$. $\endgroup$ – Michael Rozenberg Aug 22 '17 at 16:13
  • 1
    $\begingroup$ $f(a_1,a_2)=\max(a_1,a_2)+a_1a_2$ $\endgroup$ – Arash Aug 22 '17 at 16:14
  • $\begingroup$ @ Arash I think without $1$ you can not end the proof. For $n=3$ if $a\geq b\geq c$ we have $\frac{\partial f}{\partial a}-\frac{\partial f}{\partial b}=1+b+bc-a-ac$ and $b+bc-a-ac=(b-a)(1+c)$ can be negative. By the way, I think your idea works. $\endgroup$ – Michael Rozenberg Aug 22 '17 at 17:19

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.