Hi everyone,

my problem seems quite simple: I have a set $\Gamma$ along with a nice $\sigma$-algebra $\mathscr{B}$. Then I have a vector space of bounded measurable functions $A \subset \mathscr{B}_{\infty} ( \Gamma )$, and a convex function $F : A \to [0, + \infty]$ which is lower semicontinuous with respect to pointwise convergence, and it satisfies:

  • $F(0)=0$;
  • $F( \lambda f ) = \lambda F( f)$ for every $f \geq 0$, $\lambda \geq 0$.
  • $F( f) \geq \inf_{\Gamma} f$

Can I find a nonnegative (sigma-additive) measure $\mathfrak{m}$ on $\Gamma$ such that $\int_{\Gamma} f d \mathfrak{m} \leq F(f)$ and $ \mathfrak{m} (\Gamma) >0 $?

[This reminds me a bit Hahn-Banach theorem: recalling Daniell's integral I'm asking a linear functional $L$ that stays between $F$ and $\inf_{\Gamma} f$ (that is concave) and with additional condition that $L(f_n) \to 0$ everytime that I have a sequence $f_n$ decreasing to $0$.]

  • 1
    $\begingroup$ Might want to clarify: I'm guessing you want a nontrivial positive countably additive measure. By "positive measure" one usually means "nonnegative measure", and since you're assuming $F \ge 0$, the trivial measure $m \equiv 0$ always works. As you mentioned, you can use Hahn-Banach to a nontrivial positive finitely additive measure such that $\int_\Gamma fdm \le F(f)$: consider $f \equiv 1$ and the subspace $S = \{\lambda F(f) : \lambda \in \mathbb{R}\}$. Then $\lambda f \mapsto \lambda$ then extends to a nontrivial (since $F(f) \ge 1$) continuous linear functional dominated by $F$. $\endgroup$
    – Dan
    Sep 27, 2012 at 17:42
  • $\begingroup$ Thanks, now I corrected the statement... Yes, I mean a sigma additive measure; I wrote positive instead of nonnegative meaning that I wanted $\mathfrak{m} (\Gamma ) >0$ $\endgroup$ Sep 28, 2012 at 14:37
  • $\begingroup$ To be more clear, an abstract version could be: giving a functional F like this on a space X, and a subspace $A \subseteq X^*$, where $X^*$ means the algebraic dual, what condition on $F$ should I have to get the existence of a nonzero element of $A$ that is less than or equal to $F$ ? $\endgroup$ Sep 28, 2012 at 14:43

1 Answer 1


No. Some counterexamples follow: First, fix $V \in \mathcal{B}$, and let $A$ be the set of $\mathcal{B}$-measurable functions which vanish on $V$. Let $F \equiv 0$. Then $F$ satisfies all of the assumptions, since $\inf_\Gamma f \le 0$ for all $f \in A$.

Stranger examples show that even assuming $F(f) > 0$ for some $f$ won't save you. Suppose $\Gamma = \mathbb{R}$, with Lebesgue $\sigma$-algebra $\mathcal{B}$. Let $A$ be the set of bounded Lebesgue-measurable functions with compact support. Then again $\inf_\Gamma f \le 0$ for all $f \in A$. Let $\Lambda$ denote the set of finitely additive measures on $\mathcal{B}$ such that $\int f d\lambda = 0$ for all bounded continuous functions $f$ and $\lambda \in \Lambda$. By Theorem 3.4 of Yosida-Hewitt, $\Lambda \backslash \{0\} \neq \emptyset$. Define $F(f) := \sup_{\lambda \in \Lambda}\int f d\lambda$ (alternatively, $F(f) := |\int f d\lambda|$ for certain fixed $\lambda \in \Lambda \backslash \{0\}$ would work). Then $F(f) \ge 0$ for all $f \in A$, and $F(f) > 0$ for certain discontinuous $f \in A$. But if $m \ge 0$ is a countably additive measure with $\int f dm \le F(f)$ for all $f \in A$, then $\int f dm = 0$ for all nonnegative bounded continuous $f \in A$, which implies $m \equiv 0$.

Yosida & Hewitt, "Finitely additive measures": http://www.ams.org/journals/tran/1952-072-01/S0002-9947-1952-0045194-X/S0002-9947-1952-0045194-X.pdf

I suppose the moral of the story is that the set of countably additive measures is not a very well-behaved subset of $\mathcal{B}_\infty^*(\Gamma) = ba(\Gamma)$. With this in mind, I suspect that if $\Gamma$ is a nice enough topological space with $\mathcal{B}$ its Borel $\sigma$-algebra, and if you assume there exists a continuous function $f \in A$ for which $F(f) > 0$, then your conclusion should hold.

  • $\begingroup$ Thanks!! I put my question in general terms because I tought it was helpful, but in fact, in my "real" problem I have also the condition that $F( 1) = 1$ where $1(x) = 1$ for all $x \in \Gamma$. $\endgroup$ Sep 30, 2012 at 16:32
  • $\begingroup$ And yes, by the way I can look at $\Gamma$ as a topological space (actually also a metric space, if I want) and the $\sigma$-algebra is that of Borelian sets... $\endgroup$ Sep 30, 2012 at 16:34
  • $\begingroup$ In that case, if $A \subset C(\Gamma)$, where $C(\Gamma)$ is the space of continuous functions on $\Gamma$, and say $\Gamma$ is a compact Hausdorff space, then your conclusion follows from Hahn-Banach, since the dual $C(\Gamma)^*$ consists only of countably additive measures. But without assuming $A \subset C(\Gamma)$, we may be stuck. You could still find a nontrivial countably additive measure $m$ such that $\int f dm \le F(f)$ for all $f \in A \cap C(\Gamma)$, but from that we can't deduce an inequality for $f \in C(\Gamma) \backslash A$ without a stronger continuity assumption on $F$. $\endgroup$
    – Dan
    Sep 30, 2012 at 17:24

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.