Let $X$ be a locally compact Polish space and let $\mathcal{O}(X)$ be the set of open subsets of $X$. A complex valuation on $X$ is a function $v: \mathcal{O}(X) \to \mathbb{C}$ such that $v(\varnothing)=0$ and $v(U \cup V)+v(U\cap V) = v(U)+v(V)$. Further, $v$ is called continuous if for a sequence $U_0 \subseteq U_1 \subseteq U_2 \subseteq \dots$ of open sets we have that $v(U_n)$ converges to $v(\bigcup_i U_i)$ for $n \to \infty$.

Can a continuous valuation be extended to a complex Borel measure on $X$?

For valuations with values in $[0,+\infty)$ that are monotonous, I believe this follows from Theorem 4.4 in Extension of valuations on locally compact sober spaces by Alvarez-Manilla.


1 Answer 1


In the special case $X = \mathbb{R}$, I believe the answer is yes. I'm still interested in more general cases as well.

The proof for $X = \mathbb{R}$ goes as follows. Comments, shorter proofs or references are welcome.

Suppose that $v : \mathcal{O}(X) \to \mathbb{C}$ is a continuous complex valuation. Then we can define its real part $\mathrm{Re}(v)$ and imaginary part $\mathrm{Im}(v)$ via $\mathrm{Re}(v)(U) = \mathrm{Re}(v(U))$ and $\mathrm{Im}(v)(U) = \mathrm{Im}(v(U))$. This gives two valuations with values in $\mathbb{R}$, and we can write $v = \mathrm{Re}(v) + i\, \mathrm{Im}(v)$. So if we want to show that $v$ can be extended to a complex Borel measure, we can assume without loss of generality that $v$ takes values in $\mathbb{R}$.

We now show that there are only countably many points that have a "point mass" associated to it. Let $v : \mathcal{O}(X) \to \mathbb{R}$ be a continuous valuation, and define $w(x) = v(\mathbb{R})-v(\mathbb{R}-\{x\})$ for every real number $x$. Further, we write $W = \{ x \in \mathbb{R} : w(x) \neq 0\}$. We show that $W$ is countable. If it was not, then there would be an $N \in \mathbb{N}$ such that $W’ = \{ x \in \mathbb{R} : w(x) > 1/N \}$ is still uncountable. Uncountable subsets of $\mathbb{R}$ have a limit point, so we take a limit point $\tilde{x} \in W'$ and a sequence $(x_n)_n$ in $W’$ such that $x_n$ converge to $\tilde{x}$ and such that $W’’ = \{ x_1, x_2, x_3, \dots \} \cup \{ \tilde{x} \}$ is homeomorphic to $\{0,1,1/2,1/3,\dots\} \cup \{ 0 \}$. Now set $U_0 = \mathbb{R} - W’’$ and $U_n = U_0 \cup \{x_1, \dots, x_n\}$ for each $n>0$. Then $$U_n \cap (\mathbb{R} - \{x_n\}) = U_{n-1}$$ and $U_n \cup (\mathbb{R}-\{x_n\}) = \mathbb{R}$, so we see that $v(U_{n-1}) + v(\mathbb{R}) = v(U_n) + v(\mathbb{R}-\{x_n\})$, and as a result $v(U_n) = v(U_{n-1}) + w(x_n) \geq v(U_{n-1}) + 1/N$. But because $v$ is continuous and $U_n$ is an increasing union of open subsets, $v(U_n)$ should converge, a contradiction. So $W$ is countable.

We now show that $v$ extends to a (signed) Borel measure on $\mathbb{R}$. Define the cumulative distribution function $F(t) = v(\mathbb{R})-v((t,+\infty))$. Continuity of $v$ shows that $F$ is right continuous. We want to construct a (signed) Borel measure that has $F$ as cumulative distribution function. Since $F$ is right continuous, a Borel measure like this exists if and only if $F$ is of bounded variation (this is explained well here). Suppose that $F$ is not of bounded variation. Then we can find $a_0 < a_1 < \dots < a_n$ such that $\sum_{i = 0}^{n-1} | F(a_{i+1}) - F(a_i) | > 1$. We claim that we can assume $a_0,\dots,a_n \notin W$, for $W$ as defined above. Then because $a_i \notin W$ for each $a_i$, we find that $F(a_{i+1}) - F(a_i) = v((a_i,a_{i+1}))$ for each $i$. To make sure that $a_i \notin W$, we can replace $a_i$ by $a_i + \epsilon_i \notin W$ for $\epsilon>0$, with $\epsilon$ small enough such that still $\sum_{i = 0}^{n-1} |v((a_i+\epsilon_i,a_{i+1}+\epsilon_{i+1}))| = \sum_{i = 0}^{n-1} | F(a_{i+1}+\epsilon_{i+1}) - F(a_i+\epsilon_i) | > 1$. This is possible because $W$ is countable and $F$ is right continuous. If we now consider the disjoint open intervals $I_1^{(0)},\dots,I_{n_0}^{(0)}$ given by $(-\infty, a_0+\epsilon_0)$, $(a_0+\epsilon_0, a_1+\epsilon_1)$, …, $(a_{n-1}+\epsilon_{n-1}, a_n + \epsilon_n)$, $(a_n+\epsilon_n,+\infty)$, then we still have $\sum_{j=1}^{n_0} |v(I_j)| > 1$.

Now $F$ again fails to be of unbounded variation over at least one of the intervals $I^{(0)}_j$. We choose precisely one such interval, and we subdivide this particular interval further in the way described above, such that we end up in total with a new set of open intervals $I_1^{(1)}, \dots, I_{n_1}^{(1)}$ that again partition $\mathbb{R}$ (up to finitely many points) and such that $\sum_{j =1}^{n_1} |v(I_j)| > 2$. By repeating this process, we make in step $k$ an open partition $I_1^{(k)}, \dots, I_{n_k}^{(k)}$ of $\mathbb{R}$ (up to finitely many points) with $A(k) = \sum_{j=1}^{n_k} |v(I_j^{(k)})| > k$.

After renumbering, we can assume that the interval in step $k$ that gets subdivided in step $k+1$, is precisely the interval $I_1^{(k)}$. Further, by choosing the intervals small enough in each step, we can assume that $v(I_1^{(k)})$ converges to a constant $C \in \mathbb{R}$ as $k$ goes to infinity. Now let $\Omega$ be the set of intervals that do not have a further subdivision, i.e. the intervals of the form $I_j^{(k)}$ for some $k$ and $j\geq 2$. Then $\sum_{I \in \Omega} v(I)$ converges unconditionally to $v(U)$ with $U = \bigsqcup_{I \in \Omega} I$. But then the series is also absolutely convergent by Riemann's Rearrangement Theorem, so $A = \sum_{I \in \Omega} |v(I)| < +\infty$. We now see that $k < A(k) = \sum_{j=1}^{n_k} |v(I_j^{(k)})| \leq A + |v(I_1^{(k)})|$. Because $|v(I_1^{(k)})|$ converges to $C$ for $k \to +\infty$, this gives a contradiction.

Because $F$ is right continuous and of bounded variation, there is a continuous Borel measure $\mu$ such that $F(t) = \mu((-\infty, t])$. It remains to show that $\mu$ restricts to $v$ on open subsets. First one can show that $\mu(\mathbb{R}) = v(\mathbb{R}) = \lim_{x\to+\infty} F(x)$. Then from the definition of $F$ it follows that $\mu((a,+\infty))=v((a,+\infty))$ for all $a \in \mathbb{R}$. Further, $$\begin{align*}\mu((-\infty,b)) &= \lim_{\epsilon \to 0} \mu((-\infty,b-\epsilon]) = \lim_{\epsilon \to 0} F(b-\epsilon) = \lim_{\epsilon \to 0} v(\mathbb{R}) - v((b-\epsilon,+\infty)) \\ &= \lim_{\epsilon \to 0} v((-\infty,b-\epsilon)) = v((-\infty,b))\end{align*},$$ where in the second last equality we use that we can take the limit over those $\epsilon > 0$ with $b-\epsilon \notin W$. Since $\mu$ and $v$ agree on the intervals $(a,+\infty)$ and $(-\infty,b)$, they agree on all open sets.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.