15
$\begingroup$

Let $f\in C^1([0,\pi],\mathbb R)$ such that $\displaystyle\int_0^\pi f(t) dt=0$

Prove that $\forall x\in [0,\pi],\displaystyle|f(x)|\leq \sqrt{\frac{\pi}{3}\int_0^\pi f'^2(t)dt}$

Failed natural attempt

$\int_0^\pi f(t) dt=0$ tells us that there is some $\beta\in [0,1]$ such that $f(\beta)=0$

Using the fundamental theorem of calculus and Cauchy-Schwarz inequality,

$\displaystyle |f(x)|=|f(x)-f(\beta)|\leq\int_x^\beta |f'(t)|dt\leq \int_0^\pi |f'(t)|dt \leq \sqrt{\pi} \sqrt{\int_0^\pi f'^2}$

It is not sharp enough.

This might have something to do with Fourier series.

$\endgroup$
  • $\begingroup$ An idea: Endow $C^1([0,\pi],\mathbb R)$ with the supremum norm and let $D:C^1([0,\pi],\mathbb R)\to L^2([0,\pi])$ denote the differential operator, with the image space being endowed with an $L^2$ Hilbert-space structure. Also, let $I:C^1([0,\pi],\mathbb R)\to\mathbb R$ denote the integral operator. Then, you basically need to show that if $f\in\operatorname{ker} I$, then $\|f\|^2\leq(\pi/3)\cdot\langle D(f),D(f)\rangle$. I'm not sure if this characterization is helpful, but may be worth a shot. $\endgroup$ – triple_sec Jun 21 '14 at 20:20
14
$\begingroup$

Without Fourier series: we have $((t-a)f(t))'=(t-a)f'(t)+f(t)$. Hence $$ \int_0^t\tau f'(\tau)d\tau=tf(t)-\int_0^tf(\tau)d\tau,\qquad \int_t^\pi(\tau-\pi)f'(\tau)d\tau=(\pi-t)f(t)-\int_t^\pi f(\tau)d\tau. $$ By adding these, we obtain $$ \pi f(t)=\int_0^\pi g(\tau)f'(\tau)d\tau,\qquad g(\tau)=\begin{cases}\tau,&\tau<t,\\\tau-\pi,&\tau>t.\end{cases} $$ Now, by Cauchy-Schwarz, $$ \pi |f(t)|\le \sqrt{\int_0^\pi |f'(\tau)|^2d\tau}\sqrt{\frac{t^3}3+\frac{(\pi-t)^3}3}\le \sqrt{\int_0^\pi |f'(\tau)|^2d\tau}\sqrt{\frac{\pi^3}{3}} $$ Dividing by $\pi$, we obtain the desired result. QED


How to guess this proof? It is easy enough: $\frac13$ can only come from $\int t^2 dt$ in this context, so (having in mind Cauchy--Schwarz) we need to estimate $f$ somehow via the integral of $tf'(t)$.

$\endgroup$
  • $\begingroup$ Note that at the interior points of the interval we obtain an estimate that is even sharper than the one requested by @G.T.R ... $\endgroup$ – Vladimir Jun 21 '14 at 21:16
  • $\begingroup$ Great Solution (+1) !! Do you think the inequality could be sharpened further? :-) $\endgroup$ – r9m Jun 21 '14 at 21:20
  • $\begingroup$ @r9m Probably not, but I am not 100% sure. $\endgroup$ – Vladimir Jun 21 '14 at 21:22
10
$\begingroup$

Here's an sketch of a proof using Fourier series. Let $f(x) = \sum_{n=1}^\infty a_n \cos(n x)$, then

$$ \int_0^\pi f'(t)^2dt = \frac{\pi}{2}\sum_{n=1}^\infty \left(n a_n\right)^2 $$ With this in mind, we apply Cauchy-Schwarz as follows: $$ \begin{aligned} |f(x)|^2 &=\left( \sum_{n=1}^\infty \left(n a_n\right)\left(\frac{\cos(n x)}{n}\right) \right)^2\\ &\le \left( \sum_{n=1}^\infty \left(n a_n\right)^2 \right) \left(\sum_{n=1}^\infty\left(\frac{\cos(n x)}{n}\right)^2 \right) \\ &\le \left( \sum_{n=1}^\infty \left(n a_n\right)^2 \right) \left(\sum_{n=1}^\infty\frac{1}{n^2} \right)\\ &= \left( \frac{2}{\pi}\int_0^\pi f'(t)^2dt \right) \left(\frac{\pi^2}{6} \right) \end{aligned} $$

$\endgroup$
  • $\begingroup$ It should be noted that by taking limits, the inequality also holds for $x=0,\pi$. (The fourier series need not converge to $f$ in the endpoints.) It's just a detail. $\endgroup$ – punctured dusk Jun 21 '14 at 20:47

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.