Here is one of two proofs of $AM-GM$ inequality from book: "INEQUALITIES, cuadernos de olimpiadas de matematicas" by Radmila Bulajich, Jose Antonio, Rogelio Valdez:


Let $A = \frac{a_{1}+...+a_{n}}{n}$. If $a_{1},a_{2},..,a_{n}$ are all equal, then we are done. But notice there will be at least two numbers such that $a_{i}<A$ and $a_{j}>A$. Because if all $a_{i}$s $>A$ or all $a_{i}$s $<A$ then we will get a contradiction.

Consider $n=4$. $A=(a_{1}+a_{2}+a_{3}+a_{4})/4$. Take two numbers, one less than $A$ and other one greater than $A$. Le this be $a_{1} = A-h$, $a_{2}= A+k$, with $h,k>0$. Notice that $a_{1}' = A$ and $a_{2}'=A+k-h$ will make $a_{1}+a_{2}=a_{1}'+a_{2}'$ but the product $a_{1}'a_{2}' > a_{1}a_{2}$.

$$ A = \frac{a_{1}+a_{2}+a_{3}+a_{4}}{4} = \frac{a_{1}'+a_{2}'+a_{3}+a_{4}}{4}$$

and $a_{1}'a_{2}'a_{3}a_{4} > a_{1}a_{2}a_{3}a_{4}$.

We can always repeat the same process and still create a number equal to $A$, and this process cannot be used more than $4$ times.

Same argument for $n$ numbers. We can always repeat the same process and still create a number equal to $A$, and this process cannot be used more than $n$ times.

Why is the above proves $AM-GM$? I don't quite understand the connection.

  • 2
    $\begingroup$ It's induction. If $A$ is the average of $a_1$ to $a_n$, You can have a series of $a_1 + a_2 + a_3 + ..... + a_n = A + a_2' + a_3 +....a_n$ but $Aa_2'a_3a_4...a_n> a_1a_2a_3a_4...a_n$ and then you do it again to get $A+a_2' + a_3 +.... + a_n= A+A+ a_3' + a_4 + ...+a_n$ but $A*Aa_3'a_4...a_n> Aa_2'a_3a_4...a_n$ and you keep repeating untill you get $a_1a_2...a_n<.....<A*A*....*A$ so $\sqrt[n]{a_1a_2...a_n}<\sqrt[n]{A^n} = a=\frac{a_1 + .....+ a_n}n$. $\endgroup$ – fleablood Jan 27 at 6:11
  • $\begingroup$ Ah thanks. I should have seen that. $\endgroup$ – Arief Anbiya Jan 30 at 8:15

It's a well ordering principal/induction proof.

I'll explain. But before I do I'll point out that I am going to be doing a lot of statements like "one of the values of $a_i > A$ so well assume that that one is $a_1$". As both addition and multiplication are commutative, I'm going to simply index and reindex the variable as I go along and assume that somehow by magic the variables were all lined up in the order I needed them in from the beginning.

So we have $a_1,........, a_n$. The average value of each of these is $A=\frac {a_1+ ...... + a_n}n$

In the event that the $a_i$ are all equal, they are all equal to $A$ and $\sqrt[n]{a_1a_2.....a_n} = \sqrt[n]{A^n} = A = \frac {a_1+ ...... + a_n}n$ and we are done.

If they aren't equal then at least one of them is $>A$ (and I'll assume that one that is is $a_1$) and one of the (I'll assume it is $a_2$) is $< A$.

Now if I replace $a_1$ with $A$ and $a_2$ with $a'_2= A-a_1$ we get that the sum $A + a'_2 + a_3 + .... + a_n=a_1 + a_2 + a_3+.. + a_n$ stays the same but tha product $A*a'_2a_3....a_n$ does not.

And if we let $a_1 = A-h$ and $a_2= A+k$ and $a'_2 = A-h+k$ where $h,k>0$ then we have $a_1a_2 = (A-h)(A+k) = A^2-hA+kA - hk$ and $Aa'_2 = A(A-h+k) = A^2-hA+kA = a_1a_2 - hk < a_1a_2$.

So we get $a_1a_2 < Aa'_2$ and therefore $a_1a_2a_3....a_n < Aa'_2a_3....a_n$.

Now if $a'_2 = a_i = A$ we are again done but if not one of $a'_2, a_i$ is $< A$ and another is $> A$. If so let's relabel $a'_2,a_i$ as $b_2,....., b_n$ so that $b_2 < A$ and $b_3 > A$ and we do the same argument we did above and get that

$a_1a_2a_3a_4.......a_n< Aa'_2a_3a_4....a_n= Ab_2b_3b_4.....b_n< A*A*b'_3b_4....b_n$.

We keep repeating the argument until we get

$a_1a_2a_3a_4.......a_n< Aa'_2a_3a_4....a_n< A*A*b'_3b_4....b_n < A*A*A*c'_3c_4...c_4 < ...... < A*A*A*.... *A = A^n$

So we have

$a_1a_2a_3a_4..... a_n\le A^n$ (with equality holding if and only if all $a_i = A$.

$\sqrt[n]{a_1a_2a_3a_4..... a_n}\le \sqrt[n]{A^n} = A=\frac{a_1+a_2 + .....+a_n}n$.

And that's the proof.


What the proof is doing is creating a sequence of sets of $\ n\ $ numbers \begin{align} \left\{a_{11}, a_{12},\right.&\left.\dots,a_{1n}\right\}\\ \left\{a_{21}, a_{22},\right.&\left.\dots,a_{2n}\right\}\\ &\vdots\\ \left\{a_{r1}, a_{r2},\right.&\left.\dots,a_{rn}\right\} \end{align} with the following properties:

  • $$ a_{1j}=a_j\ \ \text{ for }\ j=1,2,\dots, n$$
  • $$ a_{rj}=A \ \ \text{ for }\ j=1,2,\dots, n $$
  • $$\frac{a_{i1}+ a_{i2}+\dots+a_{in}}{n}=A \ \ \text{ for }\ i=1,2,\dots, r $$
  • $$\\ \hspace{-1em} \root{n}\of{a_{i+1,1} a_{i+1,2}\dots a_{i+1,n}}> \root{n}\of{a_{i1} a_{i2}\dots a_{in}}\ \ \text{ for }\ i=1,2,\dots, r-1\ .$$

It follows from these properties, that $$ \root{n}\of{a_1a_2\dots a_n}= \root{n}\of{a_{11}a_{12}\dots a_{1n}}> \root{n}\of{a_{r1} a_{r2}\dots a_{rn}}=A\ , $$ —that is, when $\ a_1, a_2, \dots,a_n\ $ are not all equal, their geometric mean is strictly larger than their arithmetic mean.

  • 1
    $\begingroup$ Technically they're $n$-tuples, not sets. $\endgroup$ – J.G. Jan 27 at 6:30
  • $\begingroup$ But they work better as sets with constant reordering as far as the logic of the original text goes $\endgroup$ – fleablood Jan 27 at 18:15
  • $\begingroup$ That's why I thought of them as being sets, although J.G. is correct that what I had written before I added the curly brackets were technically $n$-tuples rather than sets. $\endgroup$ – lonza leggiera Jan 27 at 22:57

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.