5
$\begingroup$

I want to prove the arithmetic–geometric mean inequality. To prove that, I need the following inequality:

Suppose that $n$ is an integer which is greater than or equal to $1$ and $a_1, a_2, \dots, a_n \in \Bbb{R}$.

Then, $$n \cdot (a_1^n + a_2^n + \dots + a_n^n) \ge (a_1^{n-1} + a_2^{n-1} + \dots + a_n^{n-1}) \cdot (a_1 + a_2 + \dots + a_n).\ \cdots (1)$$


For any n positive real numbers $a_1, a_2, \cdots, a_n$, $$\frac{a_1 + a_2 + \cdots + a_n}{n} \ge \sqrt[n]{a_1 a_2 \cdots a_n}.$$ This inequality is obviously equivalent to the next inequality:

For any n positive real numbers $b_1, b_2, \cdots, b_n$, $$b_1^{n} + b_2^{n} + \cdots b_n^{n} \ge n (b_1 b_2 \cdots b_n)$$

We prove the second inequality by induction.

(i) When $n=1$, $b_1^{1} = b_1 \ge 1\cdot b_1.$

(ii)When $n=k$, suppose that the above inequality holds.

Let $c_1, c_2, \cdots, c_{k+1}$ be $k+1$ positive real numbers. Then, by the assumption above,

$$ \left\{ \begin{array}{c} c_1(c_2^{k} + c_3^{k} + \cdots + c_{k+1}^k) \ge c_1(k c_2 c_3 \cdots c_{k+1}) = k \prod_{j=1}^{k+1} c_j \\ c_2(c_1^{k} + c_3^{k} + \cdots + c_{k+1}^k) \ge c_2(k c_1 c_3 \cdots c_{k+1}) = k \prod_{j=1}^{k+1} c_j \\ \cdots \\ c_{k+1}(c_1^{k} + c_2^{k} + \cdots + c_{k}^k) \ge c_{k+1}(k c_1 c_2 \cdots c_{k}) = k \prod_{j=1}^{k+1} c_j,\\ \end{array} \right. $$ i.e., for $1 \le i \le k+1$, $$\sum_{j=1}^{k+1}\delta_{ij}^{'} c_j^k c_i \ge k \prod_{j=1}^{k+1} c_j$$ where, $\delta_{ij}^{'} = 1 - \delta_{ij}$, $\delta_{ij}$ is Kronecker's delta.

So,

$$\sum_{i=1}^{k+1} \sum_{j=1}^{k+1} \delta_{ij}^{'} c_j^k c_i \ge \sum_{i=1}^{k+1} k \prod_{j=1}^{k+1} c_j = k(k+1) \prod_{j=1}^{k+1} c_j.$$

$$\sum_{i=1}^{k+1} \sum_{j=1}^{k+1} \delta_{ij}^{'} c_j^k c_i = \sum_{i=1}^{k+1} \sum_{j=1}^{k+1} c_j^k c_i - \sum_{i=1}^{k+1} c_i^{k+1} = \sum_{j=1}^{k+1} c_j^k \cdot \sum_{i=1}^{k+1} c_i - \sum_{i=1}^{k+1} c_i^{k+1}.$$

If (1) holds, then

$$(k+1) \sum_{i=1}^{k+1} c_i^{k+1} \ge \sum_{j=1}^{k+1} c_j^k \cdot \sum_{i=1}^{k+1} c_i.$$

So,

$$k \sum_{i=1}^{k+1} c_i^{k+1} \ge \sum_{j=1}^{k+1} c_j^k \cdot \sum_{i=1}^{k+1} c_i - \sum_{i=1}^{k+1} c_i^{k+1} \ge k(k+1) \prod_{j=1}^{k+1} c_j.$$

When we divide both sides of the above inequality by k, we get

$$\sum_{i=1}^{k+1} c_i^{k+1} \ge (k+1) \prod_{j=1}^{k+1} c_j.$$

$\endgroup$
  • 2
    $\begingroup$ Well well, this is Chebyshev's inequality given that a1 > a2 > ... more in that link. $\endgroup$ – Santosh Linkha Apr 25 '14 at 11:15
  • $\begingroup$ I didn't know Chebyshev's inequality. Thank you very much for your answer. $\endgroup$ – Takaaki Apr 25 '14 at 21:31
1
$\begingroup$

There is the following reasoning for even $n$. $$n\sum_{k=1}^na_k^n-\sum_{k=1}^na_k^{n-1}\sum_{k=1}^na_k=\sum_{1\leq i<j\leq n}\left(a_i^n-a_i^{n-1}a_j-a_ia_j^{n-1}+a_j^n\right)=$$ $$=\sum_{1\leq i<j\leq n}\left(a_i^{n-1}-a_j^{n-1}\right)\left(a_i-a_j\right)\geq0.$$

For odd $n$ it's wrong.

For example, take $n=3$.

We get $3(a^3+b^3+c^3)\geq(a^2+b^2+c^2)(a+b+c),$ which is wrong for $b=c=0$ and $a=-1.$

Also, we can say that it's true for non-negative variables by Muirhead because $$(n,0,0,...,0)\succ(n-1,1,0,...,0).$$

$\endgroup$
  • $\begingroup$ What if we restrict to nonnegative $a_k$? It should work for all $n$ then. $\endgroup$ – Wojowu Mar 27 at 19:22
  • 1
    $\begingroup$ @Wojowu For non-negative variables it's true for all $n\geq1$, of course . The proof is the same. Also, we can use Chebyshov here. $\endgroup$ – Michael Rozenberg Mar 27 at 19:25

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.