I would like a clarification to an exercise in Brezis' Functional Analysis. Exercise 1.19(2) says

Let $E$ be a normed vector space. Let $F:\mathbb R \rightarrow (-\infty, +\infty]$ be a convex lower semi continuous function such that $F(0)=0$ and $F(t)\geq t$ $\forall t\in \mathbb{R}$. Set $\phi (x)=F(\|x\|)$. Prove that $\phi$ is convex, lower semi continuous and that $\phi^{\ast}(f)=F^{\ast}(\|f\|)$ $\forall f\in E^{\ast}$.

Now I can prove everything except for the last statement, where I can't understand something. The book defines the conjugate function of $\phi:E\rightarrow (-\infty, +\infty]$ to be $\phi^{\ast}:E^{\ast}\rightarrow (-\infty, +\infty]$ with $$ \phi^{\ast} (f) =\sup_{x \in E}\{ \langle f , x \rangle - \phi (x)\} $$

With this in mind, the $F^{\ast}$ of the exercise should take as argument elements of $\mathbb{R}^{\ast}$, that is continuous linear functionals $g:\mathbb R\rightarrow \mathbb R$. In this case it takes $\|f\|$, where $f\in E^{\ast}$, so $f:E\rightarrow \mathbb{R}$. I can't understand how this can happen. Surely $\|f\|$ can't be just the norm of $f$ since it would not be linear. So my question is: what is $F^{\ast}(\|f\|)$ supposed to mean in this context?

Please don't offer any hints on how to solve the exercise!

  • 2
    $\begingroup$ Every number $a$ defines a linear function $\Bbb R\stackrel{\times a}{\longrightarrow} \Bbb R$. $\endgroup$ – Quang Hoang Sep 20 '14 at 18:08
  • $\begingroup$ @QuangHoang Of course it would be something simple. Thanks! $\endgroup$ – John Doe Sep 20 '14 at 18:27

Since $F$ is convex and $F(0) = 0$ is a global minimum, $F$ is monotonically decreasing on $(-\infty, 0]$ and monotonically increasing on $[0, \infty)$. Therefore, for any $x$, $y \in \mathbb{R}$ and any $\lambda \in [0, 1]$,\begin{align*} \phi(\lambda x + (1 - \lambda)y) & = F(\|\lambda x + (1 - \lambda)y\|) \le F(\lambda\|x\| + (1 - \lambda)\|y\|) \\ & \le \lambda F(\|x\|) + (1 - \lambda)F(\|y\|) = \lambda\phi(x) + (1 - \lambda)\phi(y).\end{align*}Let $M \in \mathbb{R}$ and let $H = \{x : F(x) < M\}$. Then $\{x : \phi(x) < M\} = [0, \infty) \cap (H \cup -H)$ is closed since $H$ is closed, as $F$ is lower semicontinuous. Thus, $\phi$ is lower semicontinuous.

Finally,\begin{align*}\phi^*(f) & = \sup_{x \in E} (fx - F(\|x\|)) = \sup_{x \in E} (\pm\|f\|x - F(\|x\|) \\ & = \sup_{x \in E} (\|f\|x - F(\|x\|)) = \sup_{x \ge 0} (\|f\|x - F(\|x\|)),\end{align*}since $\|f\|x - F(\|x\|)$ is strictly less than $0$ for $x < 0$, which is less than the $0$, which is the value achieved when $x = 0$. Therefore,$$\phi^*(f) = \sup_{x \ge 0} (\|f\|x - F(x)) = F^*(\|f\|),$$where the final equality holds since $\|f\|x - F(x)$ is strictly negative when $x < 0$, and hence less than the value at $0$, which is $0$.

  • $\begingroup$ The explanation doesn't make sense at $\Vert f\Vert x$ with $x\in E$. $\endgroup$ – Sean Aug 28 '18 at 0:56

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.