12
$\begingroup$

Part of an exercise to prove Holder's inequality in Rudin involves proving Young's Inequality... That is, given $\frac{1}{p}+\frac{1}{q} = 1$, prove $$ab \leqslant \frac{a^p}{p} + \frac{b^q}{q}.$$

Here's my attempt at a proof:

Let $$f(x) = \frac{x^p}{p} + \frac{b^q}{q} -bx$$
then, $$f'(x) = x^{p-1} -b$$ so that $f$ attains a minimum at $$x=b^{\frac{1}{p-1}}.$$ Since $\frac{p}{p-1} = q$, this is equivalent to saying that $f$ attains its minimum when $x^p = b^q$. Now, we need show that $$f(b^{\frac{1}{p-1}}) = 0.$$ We compute
$$ \begin{align} f(b^{\frac{1}{p-1}}) &= \frac{(b^{\frac{1}{p-1}})^p}{p} + \frac{b^q}{q} - bb^{\frac{1}{p-1}} \\ &= \frac{b^q}{p} + \frac{b^q}{q} - b^{\frac{1}{p-1} +1} \\ &= \frac{b^q(p+q)}{pq} - b^{\frac{1}{p-1} +1} \\ &= b^q - b^{\frac{1}{p-1} +1} \\ &= b^q - b^q \\ & = 0 \end{align}$$ where $b^{\frac{1}{p-1} +1} = b^q$ since $$\frac{1}{p-1} +1 = \frac{1}{p-1} +\frac{p-1}{p-1} = \frac{p}{p-1} = q.$$ Thus, $f(x) = 0$ only when $x^p = b^q$. This is the global minimum of $f$ since $f^{''} \geq 0$ and analysis of concavity. Therefore, if $x > b^{\frac{1}{p-1}}$, $f(x) > 0$. That is, if $x^p > b^q$, the inequality holds. A similar analysis for $g(y) = \frac{a^p}{p} + \frac{y^q}{q} -ay$ shows that $g(y) > 0$ if $y^q > a^p$. Combining these two statements yields that, if $a^p \neq b^q$, the inequality holds, so we're done.

Is this a valid proof?

If so, if anyone could provide any alternative proofs, I'd be more than interested to see them.

$\endgroup$
  • 2
    $\begingroup$ Excellent! The standard proof is to use the concavity of the log function and show the inequality almost immediately. But this is creative! $\endgroup$ – Mark Viola Mar 31 '15 at 4:28
  • 1
    $\begingroup$ I think that (inequality) would be suitable for this question. (Probably the most suitable.) However, you have already used maximum of 5 tags. $\endgroup$ – Martin Sleziak Mar 31 '15 at 7:16
  • $\begingroup$ @Martin Sleziak changed accordingly. $\endgroup$ – Anthony Peter Mar 31 '15 at 7:17
4
$\begingroup$

As mentioned above, the standard proof uses concavity. You said you were interested in other proofs so here is how I learned it:

We will first assume $a,b \ne 0$ as this then becomes trivial. An equivalent statement to the one we want to prove is the following:

$$\frac{a}{b^{q-1}} \le \frac{1}{p} \cdot \frac{a^p}{b^q} + \frac{1}{q}$$

(divide Young's Inequality by $b^q$).

Let $t = \frac{a^p}{b^q}$. Then $t^{\frac{1}{p}} = \frac{a}{b^{q-1}}$, since $\frac{q}{p} = q - 1$ , we want to show that

$$t^{\frac{1}{p}} - \frac{1}{p} t \le \frac{1}{q}$$

Let $f(t) = t^{\frac{1}{p}} - \frac{1}{p} t$. Then $f(1) = 1 - \frac{1}{p} = \frac{1}{q}$. If we can show that this is the maximum of $f(t)$ when $t > 0$, we are done.

$$f'(t) = \frac{1}{p} t^{\frac{1}{p} - 1} - \frac{1}{p} = \frac{1}{p} \left(t^{-\frac{1}{q}} - 1\right)$$

When $0 < t < 1$, we see that $f'(t) > 0$ and similarly when $t > 1$, we see that $f'(t) < 0$. Thus, $\frac{1}{q}$ is a local max of $f(t)$ and so we have shown that

$$t^{\frac{1}{p}} - \frac{1}{p} t \le \frac{1}{q}$$

Now substituting back in $t = \frac{a^p}{b^q}$ we get Young's Inequality to come out with some algebraic manipulations.

$\endgroup$

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.