% ================================================================= % Lecture 2 % Primary source: handwritten notes (Mathpix mmd, lines 66-216) % Fallback: kashlak.pdf (only for OCR/notation/curriculum clarity) % ================================================================= \section[Lecture 2 -- Constructing sigma-Fields and Measures]{Lecture 2 \textemdash{} Constructing \texorpdfstring{$\sigma$}{sigma}-Fields and Measures (Existence; Carath\'eodory's Extension)} \label{sec:lec02} \emph{Motivating question.} Suppose we declare the measure of a half-open interval to be its length, \(\mu((a,b])=b-a\) for \(b>a\). What \emph{else} can we then measure? The collection of all such intervals is not a \(\sigma\)-field --- e.g.\ \((a,b]\cup(c,d]\) is not in general a half-open interval --- so we need a procedure that grows the class of ``measurable'' sets and extends \(\mu\) to it. The extension is delivered by \emph{Carath\'eodory's extension theorem}, the existence half of the construction of a measure space. Uniqueness is taken up next lecture. \subsection{Semirings, rings, and fields} We climb a small ladder of set systems sitting below a \(\sigma\)-field. At each rung the candidate measure has more room to manoeuvre. \begin{definition}{Semiring}{semiring} A collection \(\Acal\) of subsets of \(\Omega\) is a \emph{semiring} if \begin{itemize} \item \(\emptyset\in\Acal\), \item \(A\cap B\in\Acal\) for all \(A,B\in\Acal\), and \item for all \(A,B\in\Acal\) the set difference splits as a finite disjoint union \[ B\setminus A \;=\; \bigsqcup_{i=1}^{n}C_i,\qquad C_i\in\Acal. \] \end{itemize} The set difference itself need not lie in \(\Acal\), but it must be \emph{expressible} as a finite union of members of \(\Acal\). \end{definition} \begin{example}[Half-open intervals form a semiring] The collection of all half-open intervals \((a,b]\subseteq\R\) (together with \(\emptyset\)) is a semiring: intersections of half-open intervals are half-open intervals, and a difference \((a,b]\setminus(c,d]\) is the union of at most two half-open intervals. \end{example} \begin{definition}{Ring}{ring} A collection \(\Acal\) of subsets of \(\Omega\) is a \emph{ring} if \begin{itemize} \item \(\emptyset\in\Acal\), and \item for all \(A,B\in\Acal\), both \(B\setminus A\in\Acal\) and \(A\cup B\in\Acal\). \end{itemize} A ring is closed under finite (set-theoretic) unions and differences. \end{definition} \begin{example}[Finite unions of half-open intervals] The collection of all \emph{finite} unions of half-open intervals \((a,b]\subseteq\R\) is a ring. It is not yet a \(\sigma\)-field --- it fails to absorb countable unions. \end{example} \begin{definition}{Field}{field} A ring \(\Acal\) is a \emph{field} (or \emph{algebra}) on \(\Omega\) if additionally \(\Omega\in\Acal\). \end{definition} \begin{remark} A field that is closed under \emph{countable} unions is a \(\sigma\)-field. The progression \[ \text{semiring}\;\subset\;\text{ring}\;\subset\;\text{field}\;\subset\;\sigma\text{-field} \] mirrors a progression in stability under set operations: pairwise intersection only, then finite unions/differences, then \(\Omega\), finally countable unions. \end{remark} \subsection{Set functions and pre-measures} Before defining a measure we collect the regularity properties a set function may enjoy. \begin{definition}{Set function and its properties}{setfunction} Let \(\Acal\) be a collection of subsets of \(\Omega\). A \emph{set function} is any map \(\mu\colon\Acal\to[0,\infty]\) (not necessarily a measure). For \(A,B\in\Acal\) we say: \begin{itemize} \item \(\mu\) is \emph{increasing} (monotone) if \(A\subseteq B\;\Rightarrow\;\mu(A)\le\mu(B)\); \item \(\mu\) is \emph{(finitely) additive} if \(\mu(A\cup B)=\mu(A)+\mu(B)\) whenever \(A,B\in\Acal\) are disjoint and \(A\cup B\in\Acal\); \item \(\mu\) is \emph{countably additive} if for every pairwise disjoint sequence \(\{A_i\}_{i=1}^{\infty}\subseteq\Acal\) with \(\bigcup_i A_i\in\Acal\), \[ \mu\!\left(\bigcup_{i=1}^{\infty}A_i\right) \;=\; \sum_{i=1}^{\infty}\mu(A_i); \] \item \(\mu\) is \emph{countably subadditive} if for every (not necessarily disjoint) sequence \(\{A_i\}_{i=1}^{\infty}\subseteq\Acal\) with \(\bigcup_i A_i\in\Acal\), \[ \mu\!\left(\bigcup_{i=1}^{\infty}A_i\right) \;\le\; \sum_{i=1}^{\infty}\mu(A_i). \] \end{itemize} \end{definition} \begin{definition}{Pre-measure}{premeasure} A set function \(\mu\colon\Acal\to[0,\infty]\) on a ring \(\Acal\) is a \emph{pre-measure} if \(\mu(\emptyset)=0\) and \(\mu\) is countably additive on \(\Acal\). \end{definition} A pre-measure is exactly what a measure looks like \emph{before} the underlying set system has been closed up to a \(\sigma\)-field. Carath\'eodory's theorem will perform that closing-up. \subsection{Outer measure and \texorpdfstring{$\mu^{*}$}{mu*}-measurability} A pre-measure on a ring can be extended in a canonical way to \emph{every} subset of \(\Omega\) by approximating from above. \begin{definition}{Outer measure}{outermeasure} Let \(\mu\) be a pre-measure on a ring \(\Acal\) on \(\Omega\). The \emph{outer measure} induced by \(\mu\) is \[ \mu^{*}(E) \;=\; \inf\left\{\sum_{i}\mu(A_i) \;:\; A_i\in\Acal,\; E \subseteq \bigcup_{i}A_i\right\}, \qquad E \subseteq \Omega, \] where the infimum runs over finite or countable covers of \(E\) by elements of \(\Acal\). \end{definition} The outer measure \(\mu^{*}\) is defined on \emph{all} of \(\Pcal(\Omega)\), but in general it is only countably \emph{sub}additive there --- not additive. To recover countable additivity we restrict attention to sets that ``split'' every test set cleanly. \begin{definition}{$\mu^{*}$-measurable set}{caratheodory-measurable} A set \(B\subseteq\Omega\) is \emph{\(\mu^{*}\)-measurable} (in the sense of Carath\'eodory) if for every \(E\subseteq\Omega\), \[ \mu^{*}(E\cap B) \;+\; \mu^{*}(E\cap B^{c}) \;=\; \mu^{*}(E). \] Write \(\Mcal\) for the collection of all \(\mu^{*}\)-measurable subsets of \(\Omega\). \end{definition} \begin{figure}[h] \centering \begin{tikzpicture}[>=Stealth, scale=1.0] % Omega: outer rectangle \draw[thick, deepnavy] (-3.6,-2.0) rectangle (3.6,2.0); \node[deepnavy] at (3.25,1.7) {\(\Omega\)}; % B: a "notched" blob inside Omega (blue) \draw[thick, exampleblue] plot[smooth cycle, tension=0.9] coordinates {(-1.6,1.2) (0.6,1.5) (2.4,0.8) (1.5,-0.2) (2.2,-1.3) (0.0,-1.4) (-0.4,-0.4) (-1.8,-0.6)}; \node[exampleblue] at (1.6,0.4) {\(B\)}; \node[exampleblue] at (-3.0,-1.6) {\(B^{c}\)}; % E: a red blob crossing into and out of B \draw[thick, highlightred, fill=highlightred, fill opacity=0.18] plot[smooth cycle, tension=0.9] coordinates {(-2.4,1.4) (-0.2,1.3) (0.9,0.4) (-0.4,-0.2) (-2.2,0.0)}; \node[highlightred] at (-1.6,0.7) {\(E\)}; \end{tikzpicture} \caption{The Carath\'eodory criterion. A set \(B\subseteq\Omega\) is \(\mu^{*}\)-measurable when every test set \(E\) (red) is split additively by \(B\) and its complement: \(\mu^{*}(E)=\mu^{*}(E\cap B)+\mu^{*}(E\cap B^{c})\).} \label{fig:caratheodory-split} \end{figure} \begin{remark} Countable subadditivity of \(\mu^{*}\) already gives ``\(\ge\)'' in the defining identity, so the substantive condition is the reverse inequality: \(B\) does not waste mass at its boundary. The class \(\Mcal\) is the largest natural domain on which \(\mu^{*}\) is genuinely additive. \end{remark} \subsection{Carath\'eodory's extension theorem} We now assemble the pieces. Starting from a pre-measure on a ring, the outer measure machinery delivers a measure on a \(\sigma\)-field containing the original ring. \begin{theorem}{Carath\'eodory Extension}{caratheodory} Let \(\Acal\) be a ring on \(\Omega\) and let \(\mu\) be a pre-measure on \(\Acal\). Let \(\mu^{*}\) be the outer measure induced by \(\mu\) and \(\Mcal\) the collection of \(\mu^{*}\)-measurable sets. Then: \begin{enumerate} \item \(\mu^{*}(\emptyset)=0\), and \(\mu^{*}\) is monotone and countably subadditive on \(\Pcal(\Omega)\); \item \(\mu^{*}\) and \(\mu\) agree on \(\Acal\), i.e.\ \(\mu^{*}(A)=\mu(A)\) for every \(A\in\Acal\); \item \(\Acal\subseteq\Mcal\); \item \(\Mcal\) is a \(\sigma\)-field on \(\Omega\) and \(\mu^{*}\) restricted to \(\Mcal\) is a measure; \item consequently \[ \Acal \;\subseteq\; \sigma(\Acal) \;\subseteq\; \Mcal \;\subseteq\; \Pcal(\Omega), \] and \(\mu^{*}\big|_{\sigma(\Acal)}\) is a measure on \(\sigma(\Acal)\) extending \(\mu\). \end{enumerate} \end{theorem} \begin{remark} The chain in (5) records the precise sense in which Carath\'eodory ``extends'' \(\mu\): the original ring \(\Acal\) sits inside the generated \(\sigma\)-field \(\sigma(\Acal)\), which sits inside the larger \(\sigma\)-field \(\Mcal\) of \(\mu^{*}\)-measurable sets, which sits inside the full power set. Both \(\sigma(\Acal)\) and \(\Mcal\) carry the measure \(\mu^{*}\); the ``correct'' extension is the outer measure restricted to \(\sigma(\Acal)\). It can happen that \(\sigma(\Acal)\) is a strict subset of \(\Mcal\) (this is the gap filled by completion). \end{remark} \begin{remark} We have shown \emph{existence}: at least one measure on \(\sigma(\Acal)\) agreeing with \(\mu\) on \(\Acal\) exists. The companion question --- \emph{is the extension unique?} --- is \[ \mu_1(A)=\mu_2(A) \;\forall A\in\Acal \;\;\overset{?}{\Longrightarrow}\;\; \mu_1(B)=\mu_2(B) \;\forall B\in\sigma(\Acal), \] and is the subject of the next lecture, via Dynkin's \(\pi\)-\(\lambda\) theorem. \end{remark}