Measure Spaces

The triple of outcomes, allowed events and sizes that makes probability rigorous and expectations well defined.

A measure space pairs events with sizes: sigma algebra closes under complements and unions while measure adds disjoint parts

Why Does This Exist?

Probability over finite coin flips is arithmetic. Probability over real-valued inputs, stopping times and stochastic processes needs to say which sets are events and what "size" means when single points have size zero. Without that, expectations and limits rest on air.

A measure space $(\Omega, \mathcal{F}, \mu)$ answers all three: outcomes, allowed events closed under logic, and sizes that add over disjoint parts. Probability is the case $\mu(\Omega) = 1$ .

This page defines the triple. Length itself is in Lebesgue measure, and the finite-intuition version is in probability foundations.

The one idea you need first

A sigma-algebra $\mathcal{F}$ contains $\Omega$ , closes under complements, and closes under countable unions. Events must belong to it before they get probabilities.

Think of It Like This

A club with membership rules

$\Omega$ is everyone in town. $\mathcal{F}$ is the list of recognized clubs: the whole town counts, the complement of a club counts, and any countable union of clubs counts. $\mu$ is headcount share.

You cannot ask for the share of an unrecognized gathering; it is not an event. Where the analogy stops: headcounts are finite, while measures handle infinite disjoint unions by countable additivity.

How It Actually Works

A measure $\mu: \mathcal{F} \to [0,\infty]$ has $\mu(\emptyset) = 0$ and $\mu(\sqcup E_k) = \sum \mu(E_k)$ for disjoint measurable $E_k$ . Consequences: monotonicity, continuity from below ( $E_n \uparrow E$ implies $\mu(E_n) \to \mu(E)$ ), and continuity from above with one finite set.

A probability space adds $\mu(\Omega) = 1$ , usually written $P$ . A space is complete when subsets of null sets are measurable, and $\sigma$ -finite when $\Omega$ is a countable union of finite-measure sets.

Worked example

Fair die: $\Omega = \{1,\ldots,6\}$ , $\mathcal{F}$ all subsets, $P(E) = |E|/6$ . Then $P(\{2,4,6\}) = 3/6 = 0.5$ . Disjoint $A = \{1,2\}$ , $B = \{3\}$ : $P(A \sqcup B) = 2/6 + 1/6 = 0.5$ . Continuity from below: $E_n = \{1,\ldots,n\}$ for $n \le 6$ has measures $1/6, 2/6, \ldots$ rising to $1$ .

Watch Out For

Assigning probabilities to non-events

Symptom: asking for $P$ of a Vitali-type set and getting paradoxes where parts sum above the whole. Fix: restrict to $\mathcal{F}$ ; in practice Borel sets cover every event built from intervals and limits.

Assuming finite additivity extends for free

Symptom: interchanging infinite sums and measures on decreasing sequences with infinite mass, reporting a positive limit for an empty intersection. Fix: continuity from above needs some $E_n$ of finite measure; check it.

The Quick Version

A measure space is outcomes plus a sigma-algebra plus a countably additive size.
Probability is measure with total one; expectation integrates against it.
Sigma-algebras close under complement and countable union, matching logical combinations.
Continuity from below and above moves limits outside measures under mild conditions.
Check measurability and finiteness before swapping infinite operations.