1

Check if the function $$ F\colon L^2[0,1]\to L^2[0,1], (F(x))(t)=\sin x(t) $$ is Gâteaux- resp. Fréchet-differentiable at $x=0$.

I started checking if the function is Gâteauch-differentiable at $x=0$ with

$$\lim\limits_{s\to 0}\frac{F(x+sh)(t)-F(x)(t)}{s}=\lim\limits_{s\to 0}\frac{\sin (sh)(t)}{s}$$

But now I do not know how to continue the calculation...

Could anabody pls help me?

Mikasa
  • 67,374

1 Answers1

0

By using the mean value theorem, you have that $$\frac{\sin(x+sh)(t)-\sin(x)(t)}{s}=\frac{\cos(u(t))sh(t)}{s}$$

where $u(t)\in[x(t),(x+sh)(t)]$ or $u(t)\in[(x+sh)(t),x(t)]$. Now if you let $s\rightarrow 0$ you have from the last equality that $$\lim_{s\rightarrow 0}\frac{\sin(x+sh)(t)-\sin(x)(t)}{s}=\cos(x(t))h(t)$$

Hence, if $D_GF(x)$ denotes you Gâteux derivative, you have $$D_GF(x)h=\cos(x)h$$

Tomás
  • 22,559
  • Can you please say what is wrong with my solution – Tomás Jan 30 '13 at 11:28
  • In general this is an standard argument. Now try to veirfy if it is Frechet differentialbe and if you have trouble leave a comment here. – Tomás Jan 30 '13 at 12:08
  • There is a more tractable way to do this. Take a look in Theorem 1.9 page 14: http://books.google.com.br/books?id=dVSTUben3QoC&pg=PA15&hl=pt-BR&source=gbs_toc_r&cad=4#v=onepage&q&f=false – Tomás Jan 30 '13 at 12:49
  • I try to solve the problem by myself. Assumption: $x_n\to 0$ a.e. $(\lvert\cos x_n h-h\rvert^2){n\in\mathbb{N}}$ is a sequence in $L^2[0,1]$, which converges a.e. against $\lvert\cos (0)h-h\rvert^2=0$, i.e. the nullfunction. Moreover, $\lvert\cos x_n h-h\rvert^2\leq (\lvert\cos x_n h\rvert+\lvert h\rvert)^2$. So one can apply Lebesgue's theorem, which says $\lim\limits{n\to\infty}\int\limits_0^1\lvert\cos x_n(t)h(t)-h(t)\rvert^2, dt=\int\limits_0^1\lim\limits_{n\to\infty}\lvert\cos x_n(t)h(t)-h(t)\rvert^2, dt=0$ forall $h\in L^2[0,1]$. And so the supremum above converges against 0. Hope ok –  Jan 30 '13 at 16:01
  • That's correct, you did it. – Tomás Jan 30 '13 at 16:06
  • The only thing I am not totally sure about is why the sequence $(\lvert\cos(x)h-h\rvert^2)$ converges a.e. against 0. I mentioned that but its not totally clear to me. –  Jan 30 '13 at 16:57
  • This is because $\cos(x_n)$ converges almost everywhere to the function $1$. – Tomás Jan 30 '13 at 16:59
  • And why a.e.? And not everywhere? The condition is $x_n\to 0$ but not a.e. - or? –  Jan 30 '13 at 17:04
  • Because $x_n$ converges only a.e. – Tomás Jan 30 '13 at 17:29
  • Really? Does the sequence criterion of continuity say this? At Wikipedia there is no word about that the sequence converges a.e. –  Jan 30 '13 at 17:32
  • Theorem: Let $u_n,u\in L^p$ and $u_n\rightarrow u$ in $L^p$. Then there exists a subsequence of $u_n$ (note ralabeled) such that $u_n\rightarrow u$ almost everywhere and $|u_n|\leq g$ where $g\in L^p$ is a positive function. This is Theorem 4.9. of Brezis's book of functional analysis. – Tomás Jan 30 '13 at 17:55
  • Ok, now I asked enough. :-) Great thanks to you! –  Jan 30 '13 at 17:57