Theorem 14.7.6 (Carathéodory’s Extension Theorem, [Theorem 1.14, Fol99]). Let $X$ be a set, $\alg \subset 2^{X}$ be a ring, and $\mu_{0}: \alg \to [0, \infty]$ be a premeasure, then there exists a measure space $(X, \cm, \mu)$ such that

  1. $\cm \supset \alg$.

  2. $\mu|_{\alg} = \mu_{0}$.

  3. $\mu$ is complete.

  4. For any measure space $(X, \cn, \nu)$ satisfying (1) and (2),

    1. $\nu|_{\cm \cap \cn}\le \mu_{\cm \cap \cn}$.

    2. For any $E \in \cm \cap \cn$ with $\mu(E) < \infty$, $\mu(E) = \nu(E)$.

    3. If $\mu$ is $\sigma$-finite, then $\nu = \mu$.

Proof. Let

\[\mu^{*}: 2^{X} \to [0, \infty] \quad E \mapsto \inf\bracs{\sum_{n \in \natp}\mu_0(E_n) \bigg | \seq{E_n} \subset \alg, \bigcup_{n \in \natp}E_n \supset E}\]

then $\mu^{*}$ is an outer measure by Proposition 14.7.2. By Carathéodory’s theorem, there exists a measure space $(X, \cm, \mu)$ such that $\mu = \mu^{*}|_{\cm}$.

(1): Let $E \in \alg$ and $F \subset X$. If there exists $\seq{F_n}\subset \alg$ with $\bigcup_{n \in \natp}F_{n} \supset F$, then

\[\sum_{n \in \natp}\mu_{0}(F_{n}) = \sum_{n \in \natp}\mu_{0}(F_{n} \cap E) + \sum_{n \in \natp}\mu_{0}(F_{n} \setminus E) \ge \mu^{*}(F \cap E) + \mu^{*}(F \setminus E)\]

for all such $\seq{F_n}$, so $\mu^{*}(F) = \mu^{*}(F \cap E) + \mu^{*}(F \setminus E)$. Otherwise, $\mu^{*}(F) = \infty$ and the result holds directly.

(2): Let $E \in \alg$ and $\seq{E_n}\subset \alg$ such that $\bigcup_{n \in \natp}E_{n} \supset E$, then

\[\sum_{n \in \natp}\mu_{0}(E_{n}) = \sum_{n \in \natp}\mu_{0}(E_{n} \cap E) + \sum_{n \in \natp}\mu_{0}(E_{n} \setminus E) \ge \sum_{n \in \natp}\mu_{0}(E_{n} \cap E) = \mu_{0}(E)\]

As this holds for all such $\seq{E_n}$, $\mu^{*}(E) = \mu_{0}(E)$.

(U): Let $\seq{E_n}\subset \alg$, then by continuity from below (Proposition 14.1.5),

\[\nu\paren{\bigcup_{n \in \natp}E_n}= \limv{n}\nu\paren{\bigcup_{k = 1}^nE_k}= \limv{n}\mu\paren{\bigcup_{k = 1}^n E_k}= \mu\paren{\bigcup_{n \in \natp}E_n}\]

Thus if $E \in \cm \cap \cn$ with $\bigcup_{n \in \natp}E_{n} \supset E$, then

\[\nu(E) \le \nu\paren{\bigcup_{n \in \natp}E_n}= \mu\paren{\bigcup_{n \in \natp}E_n}\le \sum_{n \in \natp}\mu_{0}(E_{n})\]

As this holds for all such $\seq{E_n}$, $\nu(E) \le \mu^{*}(E) = \mu(E)$.

Suppose that $\mu(E) < \infty$, then there exists $\seq{E_n}\subset \alg$ such that $\bigcup_{n \in \natp}E_{n} \supset E$ and $\sum_{n \in \natp}\mu_{0}(E_{n}) < \infty$. In which case,

\begin{align*}\mu\paren{\bigcup_{n \in \natp}E_n}&= \nu\paren{\bigcup_{n \in \natp}E_n}= \nu(E) + \nu\paren{\bigcup_{n \in \natp}E_n \setminus E}\\&\le \nu(E) + \mu\paren{\bigcup_{n \in \natp}E_n \setminus E}\\ \mu(E)&\le \nu(E)\end{align*}

so $\nu(E) = \mu(E)$ whenever $\mu(E) < \infty$.

Finally, if $\mu$ is $\sigma$-finite, then there exists $\seq{E_n}\subset \cm$ such that $E_{n} \upto X$ and $\mu(E_{n}) < \infty$ for all $n \in \natp$. For each $n \in \natp$, there exists $\bracsn{E_{n, k}}_{k = 1}^{\infty} \subset \alg$ such that $E_{n} \subset \bigcup_{k \in \natp}E_{n, k}$ and $\mu\paren{\bigcup_{k \in \natp}E_{n, k}}< \infty$. Let $F_{n} = \bigcup_{j = 1}^{n} \bigcup_{k \in \natp}E_{j, k}$, then $\seq{F_n}\subset \sigma(\alg) \subset \cm \cap \cn$ with $F_{n} \upto X$ and $\mu(F_{n}) < \infty$ for all $n \in \natp$. For any $E \in \cm \cap \cn$, by continuity from below (Proposition 14.1.5),

\[\nu(E) = \limv{n}\nu(E \cap F_{n}) = \limv{n}\mu(E \cap F_{n}) = \mu(E)\]

so $\nu|_{\cm \cap \cn}= \mu|_{\cm \cap \cn}$.$\square$