Probability Space

Overview

A probability space is the mathematical specification of a random experiment. It separates three questions: what can happen, which questions about the outcome are measurable, and how likely their answers are. This separation works for finite experiments and for infinite objects such as a whole sequence of coin tosses.

The word “space” refers to the model, not necessarily a geometric region. Different spaces can describe the same observed distribution, while different choices of events on one set can describe different information.

Prerequisites

One needs sets, complements, countable unions, and nonnegative series. The sigma-algebra requirement is useful even before learning integration: it tells us which combinations of observable questions remain legitimate questions. Measure theory explains the general construction, but finite examples require only sums.

Definition and conventions

A probability space is a triple $(\Omega,\mathcal F,P)$. The nonempty set $\Omega$ contains outcomes. The sigma-algebra $\mathcal F$ contains $\Omega$ and is closed under complements and countable unions. Its members are events. The map $P:\mathcal F\to[0,1]$ satisfies $P(\Omega)=1$ and, for pairwise disjoint events,

Completeness is an additional convention: every subset of a probability-zero event must then be measurable. It is not part of the definition used here. On a finite space one usually chooses all subsets, but a smaller sigma-algebra is allowed.

Intuition

Think of $\mathcal F$ as the collection of questions the model can answer. Knowing only whether an outcome lies in one fixed set generates a coarse collection of questions. Assigning probabilities then supplies a consistent weighting of those answers. Countable additivity ensures that splitting an event into countably many mutually exclusive alternatives preserves its total weight.

Examples and non-examples

Take $\Omega=\{a,b,c,d\}$, all subsets as events, and point weights $1/2,1/4,1/8,1/8$. Then

Equal weights are unnecessary. If only membership in $\{a,b\}$ is observed, the four-set sigma-algebra generated by that set gives a different, coarser model.

There is no uniform probability on the positive integers assigning the same mass $p$ to every singleton. If $p>0$, finite unions eventually exceed mass one; if $p=0$, countable additivity makes the whole space have mass zero. A list of possible outcomes without an event collection and probability assignment is therefore not yet a probability space.

Important properties

Disjoint decomposition gives $P(A^c)=1-P(A)$ and $P(\varnothing)=0$. If $A\subseteq B$, writing $B$ as $A$ plus its disjoint remainder proves monotonicity. For increasing events $A_n$, split their union into successive differences to prove $P(\bigcup_n A_n)=\lim_nP(A_n)$. This proof idea explains why countable additivity controls limits, not just finite counting.

Connections and further reading

Random variables are measurable functions on this space; their distributions transport its probabilities to a numerical space. Independence and conditioning add relations between its events. Lawler, Section 1 gives the measure-theoretic formulation, and Peking University, §1.1.4 documents the Chinese name and probability axioms.