0

Let $X$ be a locally compact Hausdroff space. Let $\Gamma$ be a family of compact subsets of $X$ with the following properties.

1) $\emptyset \in \Gamma$.

2) $K\cup L \in \Gamma$ whenever $K \in \Gamma$ and $L\in \Gamma$.

3) Let $C$ be compact and let $U$ be open such that $C\subset U$. Then there exists $K\in\Gamma$ such that $C\subset K\subset U$.

Let $\lambda: \Gamma \rightarrow [0,\infty)$ be a non-negative real valued function with the following properties.

1) $\lambda(K) \le \lambda(L)$ whenever $K \subset L$.

2) $\lambda(K\cup L) \le \lambda(K) +\lambda(L)$.

3) $\lambda(K\cup L) = \lambda(K) +\lambda(L)$ whenever $K\cap L = \emptyset$.

4) $\lambda(K) = inf\{\lambda(L): K\subset int(L) \subset L, L\in \Gamma \}$, where int$(L)$ is the interior of $L$.

I would like to prove the following assertion. For the definition of Radon measure, see Fubini's theorem on a product of locally compact spaces which do not have countable bases.

There exists a unique Radon measure $\mu$ on $X$ such that $\mu = \lambda$ on $\Gamma$.

The motivation came from my question. Let $X$ and $Y$ be locally compact Hausdorff spaces. Let $\Gamma$ be the family of subsets of $X\times Y$ which are finite unions of those of the form $K\times L$ where $K$(resp. $L$) is a compact subset of $X$(resp. $Y$). Then $\lambda: \Gamma \rightarrow [0,\infty)$ can be defined by using the usual product measure on $K\times L$ where $K$(resp. $L$) is a compact subset of $X$(resp. $Y$).

Makoto Kato
  • 42,602
  • 1
    I guess that can be proved as in the proof of Riesz representation theorem in Rudin's real and complex analysis. Define for each open set $V$ $\lambda(V) = \sup_{K\subset V} \lambda (K)$ and then for any set $E$, $\lambda (E) = \inf_{E\subset V} \lambda (V)$. (There are a lot of check though) –  May 23 '15 at 01:20
  • @John The details do matter. In mathemaics, just being almost correct is not enough at all. – Makoto Kato May 25 '15 at 00:57
  • Of course. That's why I say "I guess". There are, indeed, a lot to check. –  May 25 '15 at 01:39

1 Answers1

3

The following proofs are influenced by Halmos's Measure Theory, Rudin's Real and Complex Analysis, Hewitt & Ross's Abstract Harmonic Analysis I and Folland's Real Analysis.

Definition For an open subset $U$ of $X$, we write $\mu(U) = sup \{\lambda(K): K \subset U, K \in \Gamma \}$.

For any subset $E$ of $X$, we write $\mu(E) = inf \{\mu(U): E \subset U, U open \}$.

Lemma 1 $\mu(K) = \lambda(K)$ whenever $K \in \Gamma$.

Proof: Choose $\epsilon \gt 0$. There exists $L\in \Gamma$ such that $K \subset L^{\circ}$ and $\lambda(L) - \lambda(K) \lt \epsilon$. Since $\lambda(K) \le \mu(K) \le \mu(L^{\circ}) \le \lambda(L)$ and $\epsilon$ can be arbitrarily small, $\mu(K) = \lambda(K)$.

Lemma 2 Let $C \subset U \cup V$ where $C$ is compact and $U$ and $V$ are open. Then there exist compact sets $K$ and $L$ such that $C = K \cup L$ and $K\subset U$, $L\subset V$.

Proof: Since $C - U$ and $C - V$ are disjoint compact sets, There exist disjoint open sets $U_1$ and $V_1$ such that $C - U \subset U_1$ and $C - V \subset V_1$. Put $K = C - U_1$ and $L = C - V_1$. Clearly $K \subset U$ and $L \subset V$. Since $K \cup L = (C - U_1) \cup (C - V_1) = C - (U_1\cap V_1) = C$, we are done.

Lemma 3 $\mu(U_1\cup U_2)\le \mu(U_1) + \mu(U_2)$ whenever $U_1$ and $U_2$ are open sets.

Proof: Suppose $K \subset U_1 \cup U_2$, where $K\in \Gamma$. By Lemma 2, there exist compact sets $C_1$ and $C_2$ such that $K = C_1\cup C_2$, $C_1\subset U_1$, $C_2\subset U_2$. By the assumption, there exist $K_1, K_2\in \Gamma$ such that $C_1\subset K_1\subset U_1$, $C_2\subset K_2\subset U_2$. Since $K\subset K_1\cup K_2$, $\lambda(K) \le \lambda(K_1\cup K_2)\le \lambda(K_1) + \lambda(K_2)$. Hence $\lambda(K) \le \mu(U_1) + \mu(U_2)$. Hence $\mu(U_1\cup U_2) \le \mu(U_1) + \mu(U_2)$.

Lemma 4 $\mu(U_1\cup U_2) = \mu(U_1) + \mu(U_2)$ whenever $U_1$ and $U_2$ are disjoint open sets.

Proof: Suppose $K_1\subset U_1$ and $K_2\subset U_2$, where $K_1, K_2\in \Gamma$. Since $K_1$ and $K_2$ are disjoint and $K_1\cup K_2\in \Gamma$, $\lambda(K_1) + \lambda(K_2) = \lambda(K_1\cup K_2) \le \mu(U_1\cup U_2)$. Hence $\mu(U_1) + \mu(U_2) \le \mu(U_1 \cup U_2)$. On the other hand, by Lemma 3, $\mu(U_1\cup U_2)\le \mu(U_1) + \mu(U_2)$ and we are done.

Lemma 5 Let $E_1, E_2, \cdots$ be a sequence of subsets of $X$. Then $\mu(\cup E_n) \le \Sigma \mu(E_n)$.

Proof: We may assume $\mu(E_n) \lt \infty$ for all $n$. Choose $\epsilon \gt 0$. There exist open sets $U_n \supset E_n$ such that $\mu(U_n) \lt \mu(E_n) + \epsilon/2^n, n = 1,2,3,\cdots$. Put $U = \cup U_n$. Suppose $K\subset U, K\in \Gamma$. Since $K$ is compact, $K\subset U_1\cup \cdots \cup U_m$ for some $m$. By Lemma 3, $\lambda(K) \le \mu(U_1\cup \cdots \cup U_m) \le \mu(U_1) + \cdots \mu(U_m) \le \Sigma \mu(E_n) + \epsilon$. Hence $\mu(U) \le \Sigma \mu(E_n) + \epsilon$. Since $\cup E_n \subset U$, $\mu(\cup E_n) \le \mu(U) \le \Sigma \mu(E_n) + \epsilon$. Hence $\mu(\cup E_n) \le \Sigma \mu(E_n)$.

Definition A subset $E$ of $X$ is said to be $\mu$-measurable if it satisfies the following condition.

For every subset $S$ of $X$, $\mu(S) \ge \mu(S\cap E) + \mu(S-E)$.

We denote the family of $\mu$-measurable sets by $\mathcal M$.

Lemma 5 shows that $\mu$ is a Caratheodory outer measure defined on the power set of $X$. It is well-known that $\mathcal M$ is a $\sigma$-algebra and $\mu$ is countably additive on it. See for example Halmos's Measure Theory for the proofs.

Lemma 6 A subset $E$ of $X$ is $\mu$-measurable if

$\mu(U) \ge \mu(U\cap E) + \mu(U - E)$

for all open sets $U$ such that $\mu(U) \lt \infty$.

Proof: Let $S$ be any subset of $X$. We need to prove that $\mu(S) \ge \mu(S\cap E) + \mu(S-E)$. If $\mu(S) = \infty$, this is trivially true. So we may assume $\mu(S) \lt \infty$. Let $U$ be an open set such that $U \supset S$ and $\mu(U) \lt \infty$. Then we have $\mu(U) \ge \mu(U\cap E) + \mu(U - E) \ge \mu(S\cap E) + \mu(S-E)$. Hence $\mu(S) \ge \mu(S\cap E) + \mu(S-E)$.

Lemma 7 Every open subsets of $X$ is $\mu$-measurable.

Proof: Let $V$ be an open subset of $X$. By Lemma 6, it suffices to prove that $\mu(U) \ge \mu(U\cap V) + \mu(U - V)$ for every open set $U$ such that $\mu(U)\lt\infty$. Choose $\epsilon\gt 0$. There exists $K \in \Gamma$ such that $K \subset U \cap V$ and $\lambda(K) \gt \mu(U \cap V) - \epsilon$. Since $U - K$ is open, there exists $L \in \Gamma$ such that $L \subset U - K$ and $\lambda(L) \gt \mu(U - K) - \epsilon$. Since $K \cup L \subset U$ and $K\cap L = \phi$, we have $\mu(U) \ge \lambda(K) + \lambda(L) \gt \mu(U \cap V) + \mu(U - K) - 2\epsilon \ge \mu(U \cap V) + \mu(U - V) - 2\epsilon$. Letting $\epsilon \rightarrow 0$, we have $\mu(U) \ge \mu(U\cap V) + \mu(U - V)$ as desired.

Proposition There exists a unique Radon measure $\mu$ on $X$ such that $\mu = \lambda$ on $\Gamma$.

Proof: The uniqueness is clear. The existence follows from Lemma 1 and Lemma 7.