User:Camillo.delellis/sandbox

2020 Mathematics Subject Classification: Primary: 28A [MSN][ZBL]

Definition

An outer measure is a set function $\mu$ such that

• Its domain of definition is an hereditary $\sigma$-ring (also called $\sigma$-ideal) of subsets of a given space $X$, i.e. a $\sigma$-ring $\mathcal{R}\subset \mathcal{P} (X)$ with the property that for every $E\in \mathcal{R}$ all subsets of $E$ belong to $\mathcal{R}$;
• Its range is the extended real half-line $[0, \infty$];
• $\mu (\emptyset) =0$ and $\mu$ is $\sigma$-subadditive (also called countably subadditive), i.e. for every countable family $\{E_i\}\subset \mathcal{R}$ the following inequality holds:

\[ \mu \left(\bigcup_i E_i\right) \leq \sum_i \mu (E_i)\, . \] Cp. with Section 10 of [Ha] and with Section 1.1 of [EG]. The most common outer measures are indeed defined on the full space $\mathcal{P} (X)$ of subsets of $X$. Indeed observe that, if an hereditary $\sigma$-ring is also an algebra, then it must contain $X$ and hence it coincides necessarily with $\mathcal{P} (X)$.

Measurable sets

There is a commonly used procedure to derive a measure from an outer measure $\mu$ on an hereditary $\sigma$-ring $\mathcal{R}$ (cp. with Section 11 of [Ha] and Section 1.1 of [EG]).

Definition 1 If $\mu:\mathcal{R}\to [0, \infty]$ is an outer measure, then a set $M\in \mathcal{R}$ is called $\mu$-measurable if \[ \mu (A\cap M) + \mu (A\setminus M) = \mu (A) \qquad \forall A\in \mathcal{R}\, . \]

Theorem 2 If $\mu:\mathcal{R}\to [0, \infty]$ is an outer measure, then the class $\mathcal{M}$ of $\mu$-measurable sets is a $\sigma$-ring and $\mu$ is countably additive on $\mathcal{M}$, i.e. \[ \mu \left(\bigcup_i E_i\right) = \sum_i \mu (E_i) \] whenever $\{E_i\}\subset \mathcal{M}$ is a countable collection of pairwise disjoint sets.

When $\mathcal{R} = \mathcal{P} (X)$, then it follows trivially from the definition that $X\in \mathcal{M}$: thus $\mathcal{M}$ is in fact a $\sigma$-algebra. Therefore $(X, \mathcal{M}, \mu)$ is a measure space.

Caratheodory criterion

An important class of outer measures on metric spaces $X$ are the ones satisfying the so-called Caratheodory criterion (called also Caratheodory measures): for such $\mu$ the Borel sets are $\mu$-measurable.

Definition 3 An outer measure $\mu: \mathcal{P} (X) \to [0, \infty]$ on a metric space $(X,d)$ is a Caratheodory measure if \[ \mu (A\cup B) = \mu (A) + \mu (B) \] for every pair of sets $A, B\subset X$ which have positive distance (i.e. such that $\inf \{d(x,y): x\in A, y\in B\} > 0$).

Theorem 4 If $\mu$ is a Caratheodory measure, then every Borel set is $\mu$-measurable.

Cp. with Theorem 5 of [EG]

Regular and Borel regular outer measures

Several authors call regular those outer measures $\mu$ on $\mathcal{P} (X)$ such that for every $E\subset X$ there is a $\mu$-measurable set $F$ with $E\subset F$ and $\mu (E) = \mu (F)$. They moreover call $\mu$ Borel regular if the Borel sets are $\mu$-measurable and for every $E\subset X$ there is a Borel set $G$ with $E\subset G$ and $\mu (E) = \mu (G)$. Cp. with Section 1.1 of [EG].

Constructions of outer measures

Outer measures induced by measures

A common procedure to construct outer measures $\mu$ from a set function $\nu$ is the following.

Definition 5 If $\mathcal{C}$ is class of subsets of $X$ containing the empty set and $\nu : \mathcal{C}\to [0, \infty]$ a set function, for every set $A\subset X$ we define \begin{equation}\label{e:extension} \mu (A) = \inf \left\{ \sum_i \mu (E_i) : \{E_i\}_{i\in \mathbb N} \subset \mathcal{C} \mbox{ covers '"`UNIQ-MathJax67-QINU`"'}.\right\}\, . \end{equation}

Some authors define such $\mu$ on the hereditary $\sigma$-ring $\mathcal{H}$ of subsets of $X$ for which there is a countable covering in $\mathcal{C}$ (cp. with Section 10 of [Ha]). Other authors define $\mu$ on the entire $\mathcal{P} (X)$, using the convention that $\mu (A) = \infty$ when $A\not \in \mathcal{H}$ (cp. for instance with [Mu]).

Theorem 6 If $\nu$ and $\mu$ are as in Definition 5, then $\mu$ is an outer measure on $\mathcal{P} (X)$. If in addition $\mathcal{C}$ is a ring and $\nu$ is a finitely additive set function, then $\mu (E) = \nu (E)$ for every $E\in \mathcal{C}$.

Cp. with Theorem A of [Ha] (NB: the proof given in [Ha] of $\sigma$-subadditivity of $\mu$ does not use the assumption that $\mu$ is finitely additive).