14.1 Measures

Definition 14.1.1 (Measurable Space). Let $X$ be a set and $\cm \subset 2^{X}$ be a $\sigma$-algebra, then the pair $(X, \cm)$ is a measurable space.

Definition 14.1.2 (Measure). Let $(X, \cm)$ be a measurable space and $\mu: \cm \to [0, \infty]$, then $\mu$ is a (countably-additive) measure if

  1. $\mu(\emptyset) = 0$.

  2. For any $\seq{E_n}\subset \cm$ pairwise disjoint, $\mu\paren{\bigsqcup_{n \in \natp}E_n}= \sum_{n \in \natp}\mu(E_{n})$.

In which case, $(X, \cm, \mu)$ is a measure space.

If $\mu: \cm \to [0, \infty]$ instead satisfies (M1) and

  1. For any $\seqf{E_j}\subset \cm$ pairwise disjoint, $\mu\paren{\bigsqcup_{j = 1}^n E_j}= \sum_{j = 1}^{n} \mu(E_{j})$.

then $\mu$ is a finitely-additive measure.

Definition 14.1.3 (Null Set). Let $(X, \cm, \mu)$ be a measure space, then $E \in \cm$ is $\mu$-null/null if $\mu(E) = 0$.

Definition 14.1.4 (Almost Everywhere). Let $(X, \cm, \mu)$ be a measure space. A statement holds $\mu$-almost everywhere/$\mu$-a.e./a.e. if it is false on a subset of a $\mu$-null set.

Proposition 14.1.5. Let $(X, \cm, \mu)$ be a measure space, then:

  1. For any $E, F \in \cm$ with $E \subset F$, $\mu(E) \le \mu(F)$.

  2. For any $\seq{E_n}\subset \cm$, $\mu\paren{\bigcup_{n \in \natp}E_n}\le \sum_{n \in \natp}\mu(E_{n})$.

  3. For any $\seq{E_n}\subset \cm$ with $E_{n} \subset E_{n+1}$ for all $n \in \nat$, $\mu\paren{\bigcup_{n \in \nat}E_n}= \limv{n}\mu(E_{n})$.

  4. For any $\seq{E_n}\subset \cm$ with $E_{n} \supset E_{n+1}$ for all $n \in \nat$ and $\mu(E_{1}) < \infty$, $\mu(\bigcap_{n \in \natp}E_{n}) = \limv{n}\mu(E_{n})$.

  5. For any $\seq{E_n}\subset \cm$,

    \[\mu\paren{\liminf_{n \to \infty}E_n}\le \liminf_{n \to \infty}\mu(E_{n})\]
  6. For any $\seq{E_n}\subset \cm$ with $\mu\paren{\bigcup_{n \in \nat}E_n}< \infty$,

    \[\mu\paren{\limsup_{n \to \infty}E_n}\ge \limsup_{n \to \infty}\mu(E_{n})\]

Proof. (1): $\mu(F) = \mu(E) + \mu(F \setminus E) \ge \mu(E)$.

(2): For each $n \in \natp$, let $F_{n} = E_{n} \setminus \bigcup_{k = 1}^{n-1}E_{k}$, then $\seq{F_n}$ is pairwise disjoint with $\bigsqcup_{n \in \natp}F_{n} = \bigcup_{n \in \natp}E_{n}$. In which case,

\[\mu\paren{\bigcup_{n \in \nat}E_n}= \mu\paren{\bigsqcup_{n \in \nat}F_n}= \sum_{n \in \natp}\mu(F_{n}) \le \sum_{n \in \natp}\mu(E_{n})\]

(3): Denote $E_{0} = \emptyset$. For each $n \in \natp$, let $F_{n} = E_{n} \setminus E_{n - 1}$, then $\seq{F_n}$ is pairwise disjoint with $\bigsqcup_{n \in \natp}F_{n} = \bigcup_{n \in \natp}E_{n}$. In which case,

\[\mu\paren{\bigcup_{n \in \natp}E_n}= \mu\paren{\bigsqcup_{n \in \natp}F_n}= \limv{n}\sum_{k = 1}^{n} \mu(F_{n}) = \limv{n}\mu(E_{n})\]

(4): Since $\mu(E_{1}) < \infty$,

\[\limv{n}\mu(E_{n}) = \mu(E_{1}) - \limv{n}\mu(E_{1} \setminus E_{n}) = \mu(E_{1}) - \mu\paren{E_1 \setminus \bigcap_{n \in \natp}E_n}= \mu\paren{\bigcap_{n \in \natp}E_n}\]

by (3).

(5): Using (1) and (3),

\begin{align*}\mu\paren{\liminf_{n \to \infty}E_n}&= \mu\paren{\bigcup_{n \in \natp}\bigcap_{m \ge n}E_m}= \limv{n}\mu\paren{\bigcap_{m \ge n}E_m}\\&\le \limv{n}\inf_{m \ge n}\mu(E_{m}) = \liminf_{n \to \infty}\mu(E_{n})\end{align*}

(6): Using (1) and (4),

\begin{align*}\mu\paren{\limsup_{n \to \infty}E_n}&= \mu\paren{\bigcap_{n \in \natp}\bigcup_{m \ge n}E_m}= \limv{n}\mu\paren{\bigcup_{m \ge n}E_m}\\&\ge \limv{n}\sup_{m \ge n}\mu(E_{m}) = \limsup_{n \to \infty}\mu(E_{n})\end{align*}
$\square$

Theorem 14.1.6 (Dynkin’s Uniqueness Theorem). Let $(X, \cm)$ be a measurable space, $\mathcal{P}\subset \cm$ be a $\pi$-syste, and $\mu, \nu: \cm \to [0, \infty]$ be measures. If

  1. $\sigma(\mathcal{P}) = \cm$.

  2. $\mu(E) = \nu(E)$ for all $E \in \mathcal{P}$.

  3. There exists $\seq{E_n}\subset \mathcal{P}$ such that $E_{n} \upto X$ and $\mu(E_{n}) < \infty$ for all $n \in \natp$.

then $\mu = \nu$.

Proof. Let $F \in \mathcal{P}$ with $\mu(F) < \infty$ and

\[\alg(F) = \bracs{E \in \cm|\mu(E \cap F) = \nu(E \cap F)}\]

then $\alg(F) \supset \mathcal{P}$ by (b), and

  1. For any $E, E' \in \alg(F)$ with $E \subset E'$ and $F \in \mathcal{P}$,

    \begin{align*}\mu((E' \setminus E) \cap F)&= \mu(E' \cap F) - \mu(E \cap F) \\&= \nu(E' \cap F) - \nu(E \cap F) = \nu((E' \setminus E) \cap F)\end{align*}
  2. For any $\seq{E_n}\subset \alg$ with $E_{n} \subset E_{n+1}$ for all $n \in \natp$ and $F \in \mathcal{P}$,

    \[\mu\paren{\bigcup_{n \in \nat}E_n \cap F}= \limv{n}\mu(E_{n} \cap F) = \limv{n}\nu(E_{n} \cap F) = \nu\paren{\bigcup_{n \in \nat}E_n \cap F}\]

    by continuity from below (Proposition 14.1.5).

so $\alg(F)$ is a $\lambda$-system. By (a) and Dynkin’s $\pi$-$\lambda$ theorem (Theorem 13.2.5), $\alg(F) = \cm$.

Let $\seq{E_n}$ as in assumption (c), then $\mu(E_{n} \cap F) = \mu(E_{n} \cap F)$ for all $n \in \natp$ and $F \in \cm$. Thus by continuity from below (Proposition 14.1.5),

\[\mu(F) = \limv{n}\mu(E_{n} \cap F) = \limv{n}\nu(E_{n} \cap F) = \nu(F)\]
$\square$