5
$\begingroup$

Using the axiom of choice and a Hamel basis for a normed space, one can prove the existence of everywhere defined discontinuous linear functionals.

My question: Does there exist a discontinuous everywhere defined linear functional on $L^p(\mathbb{R}^n)$, $1\leq p <+\infty$, representable by a measurable function $g : \mathbb{R}^n \to \mathbb{R}$?

In other words, the question is whether the following false: If $g$ is measurable and $$ \forall f\in L^p, \quad \left| \int f g \right| < +\infty, $$ then $g \in L^{p*}$?

I have tried to argue by counterpositive with no luck: The idea is to assume that $g\notin L^{p*}$ and construct an $f\in L^p$ such that $\big|\int fg \big|= +\infty$.

$\endgroup$
  • $\begingroup$ Try using Banach-Steinhaus to show that $g \in L^{p\ast}$. $\endgroup$ – Daniel Fischer Nov 6 '14 at 13:56
  • $\begingroup$ That did the trick! This begs the question: does there exist an explicit discontinous but everywhere defined linear functional on $L^p$? $\endgroup$ – Jas Ter Nov 6 '14 at 14:13
  • $\begingroup$ Grrmbl. I don't know of an explicit example, and am not optimistic about the possibility of constructing one. $\endgroup$ – Daniel Fischer Nov 6 '14 at 14:18
  • $\begingroup$ Hold on. Am I applying Banach-Steinhaus in the wrong way? BS assumes that the functionals in the given collection (here $g$ alone) are continuous. Then, I cannot use B-S to prove that $g$ is continuous. $\endgroup$ – Jas Ter Nov 6 '14 at 14:31
  • $\begingroup$ Cut off $g$ to obtain $g_n \in L^1 \cap L^\infty$. If you let $n\to\infty$, $g_n \to g$ pointwise, and $\int fg_n\,d\lambda \to \int fg\,d\lambda$ by dominated convergence. $\endgroup$ – Daniel Fischer Nov 6 '14 at 14:35
5
$\begingroup$

If $1 \leqslant p < +\infty$, and $g \colon \mathbb{R}^n \to \mathbb{R}$ (or $\mathbb{R}^n \to \mathbb{C}$) is a measurable function such that for all $f \in L^p(\mathbb{R}^n)$ the integral

$$\int_{\mathbb{R}^n} fg \,d \lambda$$

exists, then it follows that $g \in L^{p^{\ast}}(\mathbb{R}^n)$, where $p^{\ast}$ is the conjugate exponent to $p$, hence the linear functional $T \colon f \mapsto \int fg\,d\lambda$ is continuous on $L^p(\mathbb{R}^n)$.

We can prove that using the Banach-Steinhaus theorem: For $n \in \mathbb{N}$ define

$$g_n(x) = \begin{cases} \qquad 0 &, \lVert x\rVert > n \\ \quad\;\; g(x) &, \lVert x\rVert \leqslant n \text{ and } \lvert g(x)\rvert \leqslant n \\ \frac{n}{\lvert g(x)\rvert}\cdot g(x) &, \lVert x\rVert \leqslant n < \lvert g(x)\rvert.\end{cases}$$

Then $g_n \in L^1(\mathbb{R}^n) \cap L^{\infty}(\mathbb{R}^n) \subset L^{p^{\ast}}$, and we have $\lim\limits_{n \to +\infty} g_n(x) = g(x)$ for all $x\in \mathbb{R}^n$. The family of continuous linear functionals $T_n \colon f \mapsto \int fg_n\,d\lambda$ is pointwise bounded, since

$$\int_{\mathbb{R}^n} f(x)g_n(x)\,d\lambda \xrightarrow{n\to +\infty} \int_{\mathbb{R}^n} f(x) g(x)\,d\lambda$$

by the dominated convergence theorem. The Banach-Steinhaus theorem asserts that the family $\{ T_n : n \in \mathbb{N}\}$ is equicontinuous, i.e. $\sup\limits_{n\in \mathbb{N}} \lVert T_n\rVert < +\infty$. It then follows that $T$ is continuous and

$$\lVert T\rVert = \lVert g\rVert_{L^{p^{\ast}}(\mathbb{R}^n)} \leqslant \sup_{n\in \mathbb{N}} \lVert T_n\rVert.$$

An application of the monotone convergence theorem shows that in fact

$$\lVert T\rVert = \lVert g\rVert_{L^{p^{\ast}}(\mathbb{R}^n)} = \sup_{n\in \mathbb{N}} \lVert T_n\rVert = \lim_{n\to +\infty} \lVert g_n\rVert_{L^{p^{\ast}}(\mathbb{R}^n)},$$

since $\lvert g_n(x)\rvert^{p^{\ast}}$ converges monotonically to $\lvert g(x)\rvert^{p^{\ast}}$.

The same result with basically the same proof works for an arbitrary $\sigma$-finite measure space, one just replaces the balls of radius $n$ with centre $0$ used here by a sequence of sets with finite measure exhausting the space.

| cite | improve this answer | |
$\endgroup$

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.