It is known that the dual of $L^\infty (X, m)$ is the space of complex, finite, finitely additive measures on $X$ that are absolutely continuous with respect to $m$.

I would like to know whether one can get better properties of the measures in the dual if $X$ is a topological space and $m$ is Borel, regular, $\sigma$-finite. Most importantly, I would like to know whether these measures become countably-additive.

If $\omega \in L^\infty (X, m) ^*$ then there exist $\mu$, a complex, finite, finitely additive measure on $X$ that are absolutely continuous with respect to $m$, such that $\omega (f) = \int _X f \ \Bbb d \mu$.

Since the space $C_0 (X)$ of continuous functions that vanish at infinity embeds continuously into $L^\infty (X, m)$, it follows that $\omega$ restricts to a continuous linear functional on $C_0 (X)$, and can be therefore represented by some finite, regular measure $\nu$. A consequence of this is that $\int _X f \ \Bbb d \mu = \int _X f \ \Bbb d \nu \quad \forall f \in C_0 (X)$. Does this allow us to conclude that $\mu$ is countably additive?

(Quite frankly, I'd say that it doesn't, otherwise we could use the same technique to show that if $X$ is normal then the measures that form the dual of $C_b (X)$, which a priori are only finitely additive, should be countably additive. Nevertheless, I'm still hoping that $L^\infty (X, m)$ gets some nice properties from the topological properties of $m$.)

Alternatively, could one define a $L^\infty$ space of functions vanishing at infinity, call it $L_0 ^\infty$, and hope that its dual be some happy mix of the spaces $L^\infty (X, m)$ and $rca(X)$?


Even if we have a nice topological measure space, there is no way for us to expect any "regularity" of the measure from the Riesz-Representation theorem. Any possible nice properties of the resulting measure will come from the behavior of the linear functional in question. Intuitively, I think of the measures that are not countably additive as coming from linear functionals that do not arise as the restriction of a functional from $L^1$ (not exactly sure how correct this intuition is though).

Here's an example that should work. Let $X=[0,1]$ endowed with the Borel $\sigma$-algebra, $\mathfrak{M}=\mathcal{B}([0,1])$, and the Lebesgue measure, $\mu$. Clearly, $X$ is a topological space and $\mu$ is a Borel regular $\sigma$-finite measure.

Then consider the subspace of $L^\infty$, denoted by $S$, consisting of equivalence classes $[f]$ such that there is a measurable representative with the limit existing as $x\to 0$. Define the linear functional $L:S\to\mathbb{R}$ via $$L([f])=\lim_{x\to 0}\tilde{f}(x),$$ where $\tilde{f}$ is the associated representative. Now we show that $L$ is well defined, suppose that $\tilde{f}=\tilde{g}$ for $\mu$-a.e. $x\in X$, and that both $f_0:=\lim_{x\to 0}\tilde{f}(x)$ and $g_0:=\lim_{x\to 0}\tilde{g}(x)$ exist. Let $E:=\{x\in X~:~\tilde{f}(x)\neq\tilde{g}(x)\}$, which satisfies $\mu(E)=0$. Now fix $\varepsilon > 0$ and find $\delta > 0$ such that $|x|<\delta$ implies $|\tilde{f}(x)-f_0|<\varepsilon/2$ and $|\tilde{g}(x)-g_0|<\varepsilon/2$. Since $E$ contains no intervals, we see that there exists some point $x_0\in X$ with $|x_0|<\delta$ with $\tilde{f}(x_0)=\tilde{g}(x_0)$. So we deduce $$|f_0 - g_0|=|f_0-\tilde{f}(x_0)+\tilde{g}(x_0)-g_0|\leq|f_0-\tilde{f}(x_0)|+|g_0-\tilde{g}(x_0)|<\varepsilon.$$ Now by taking $\varepsilon\to 0$ we see that $f_0=g_0$, and so $L$ is well-defined. Similarly, we can show that $|L([f])|\leq\|[f]\|_{L^\infty}$.

Now use the Hahn-Banach theorem to extend $L$ to a linear functional $\Lambda\in(L^\infty(X,\mathfrak{M},\mu))^*$. Let $\lambda$ be the measure from Riesz-Representation theorem representing $\Lambda$, i.e. $$\Lambda(f)=\int_{[0,1]}f~d\lambda,\qquad f\in L^\infty(X,\mathfrak{M},\mu).$$ Now we have that $\lambda((0,1/n))=1$ for all $n\in\mathbb{N}$ (this follows from how we define the measure $\lambda$ in the proof of the representation theorem as $\lambda(E)=\Lambda(\chi_E)$, as well as the uniqueness). Now we obtain a contradiction if $\lambda$ were countably additive: if so, then we would have $\lambda(\emptyset)=1$, which clearly does not make sense.

To expand a bit on why your remark on the embedding of $C_0(X)$ into $L^\infty$ does not imply that the measure is countably additive, we can use the same example above. At first, you might think that the measure above is a Dirac delta at $0$, but in fact, we have that $\lambda(\{0\})=0$, but $\lambda((0,1])=1$. On the other hand, if we restrict $L$ to $C_0(X)=C(X)$ we obtain a linear functional $L:C(X)\to\mathbb{R}$ given by $L(f)=f(0)$. As a result, we deduce that the resulting measure from the Riesz-Representation theorem for $C(X)$ will be the Dirac delta measure at zero, $\nu =\delta_0$. So we see that the domain of the linear functional makes a very big difference on the measure which will represent it.

Regarding your last question about the dual of $L^\infty_0(X,\mathfrak{M},\mu)$, I'm not really sure what the dual space would look like right now.


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.