6

Can someone help me to construct a linear functional in $\mathcal{C}([0,1])$ that does not attain its norm?

Actually, I want to prove that $\mathcal{C}([0,1])$ is not reflexive Banach space. Is it sufficient to construct that kind of functional?

Norbert
  • 56,803
elinor
  • 387

2 Answers2

12

Try $\varphi(f) = \int_{0}^{1/2} f(x)\,dx - \int_{1/2}^{1} f(x)\,dx$.

As for the second question, yes. For every $x \in X$ there is by Hahn-Banach a functional $\varphi \in X^{\ast}$ with $\|\varphi\| = 1$ such that $\varphi(x) = \|x\|$. Now apply this to $X^{\ast}$ and use reflexivity.

t.b.
  • 78,116
  • Sir , I am trying to see why $||\phi||=1$ . Can you give me some hints . I would like to work it out. – Theorem Nov 27 '12 at 20:55
  • $\leq$ is trivial, for $\geq$ take $f$ piecewise linear, constant equal to $1$ on $[0,1/2 - \varepsilon]$ and constant equal to $-1$ on $[1/2 +\varepsilon, 1]$. – t.b. Nov 29 '12 at 14:19
  • i don't get it why $\le$ is trivial, can u give me more details . – Theorem Nov 29 '12 at 14:39
  • triangle inequality + $\lvert \int f \rvert \leq \int \rvert f\lvert$. – t.b. Nov 29 '12 at 14:42
  • yes , got it . But are you sure for $\ge$ the function u gave will work ? – Theorem Nov 29 '12 at 17:25
  • Sir that will be ,$\phi (f)\ge \int_0^{1/2-\epsilon}|f|-\int_{1/2+\epsilon}^1|f|$ right , it doesn't seem to work ? – Theorem Nov 29 '12 at 20:18
  • And can i show that it doesn't attain its norm without using reflexivity ? – Theorem Nov 29 '12 at 20:33
  • Try again: $$f_\varepsilon(x) = \begin{cases} 1 & 0 \leq x \leq 1/2 - \varepsilon \ \text{linear} & \text{on }1/2 - \varepsilon \lt x \lt 1/2+\varepsilon \ -1 & 1/2 + \varepsilon \leq x \leq 1.\end{cases}$$ You should get $\varphi(f_\varepsilon) = 1 - \varepsilon$ while $\lVert f_\varepsilon \rVert = 1$, now let $\varepsilon \to 0$. Yes, no reflexivity needed: see here and here. – t.b. Nov 29 '12 at 22:23
6

Knowing some deeper theorems will help to see what to do.

The Riesz representation theorem says that the continuous linear functionals on $C([0,1])$ are precisely the signed Radon measures on $[0,1]$. It's not hard to see that any bounded measurable function is a continuous linear functional on the space of signed measures, i.e. is an element of $C([0,1])^{**}$. So the existence of bounded discontinuous functions on $[0,1]$ shows that $C([0,1])$ cannot be reflexive.

Nate Eldredge
  • 97,710