11
$\begingroup$

My Analysis professor showed this inequality and elegantly proved it using polar coordinates, saying that it can't be done algebraically. Instead, here's how I think I have handled it: firstly we see it's true for $ab=0$; dividing by the RHS we get $$\left(\frac{a}{b}\right)^{2-\alpha}+\left(\frac{a}{b}\right)^{-\alpha}\ge1, $$ or equivalently, setting $t=a/b$, $$t^2+1\ge t^\alpha$$ which holds because the LHS is $\ge t^2\ge t^\alpha$ for $t\ge1$ and $\ge1\ge t^\alpha$ for $0\le t<1$.

Am I missing something? Are there fancy algebraic (perhaps some linear algebra inequalities) ways to prove the inequality?

$\endgroup$
  • 2
    $\begingroup$ It can't be done by... is always a dangerous thing to say in maths :) $\endgroup$ – Jack D'Aurizio Oct 3 '16 at 21:20
  • 5
    $\begingroup$ +1 for not being discouraged by prof. However, maybe a hardcore algebraists won't even like exponentiation with irrational exponent? $\endgroup$ – Hagen von Eitzen Oct 3 '16 at 21:21
  • $\begingroup$ @JackD'Aurizio: Indeed, ahahah :D $\endgroup$ – Vincenzo Oliva Oct 3 '16 at 21:25
  • 3
    $\begingroup$ Another way: Write $\alpha = 1 + \beta$ with $\beta \in [-1,1]$. Divide by $ab$ (after disposing of the trivial case $ab = 0$) to get $$\frac{a}{b} + \frac{b}{a} \geqslant \biggl(\frac{a}{b}\biggr)^{\beta}.$$ $\endgroup$ – Daniel Fischer Oct 3 '16 at 21:33
  • 3
    $\begingroup$ Or, if you assume that $a\geq b$, then $$a^2+b^2\geq a^2 = a^\alpha a^{2-\alpha} \geq a^\alpha b^{2-\alpha}.$$ $\endgroup$ – Pierre-Guy Plamondon Oct 3 '16 at 21:36
7
$\begingroup$

Turning my comment into an answer: I don't know whether you'll think this is cheating, but assuming without loss of generality that $a\geq b$, then $$a^2+b^2 \geq a^2 = a^\alpha a^{2-\alpha} \geq a^\alpha b^{2-\alpha}. $$

$\endgroup$
  • $\begingroup$ Cheating? Just clever. :D $\endgroup$ – Vincenzo Oliva Oct 3 '16 at 21:55
0
$\begingroup$

Ask him to prove, using polar coordinates, that $$a^2+b^2\ge ka^{\alpha}b^{2-\alpha}\quad \forall 0\le\alpha\le 2, \forall a,b\ge 0$$ where $$k=\frac{2}{\left(\alpha^{\alpha}(2-\alpha)^{(2-\alpha)}\right)^{1/2}}.$$ (Which is the best constant possible.)

$\endgroup$
  • $\begingroup$ Interesting, thanks. $\endgroup$ – Vincenzo Oliva Oct 6 '16 at 6:23

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.