A topic in the Open Knowledge Graph — a free, open map of 15,290 topics and the order to learn them in.
A sigma-algebra on a set Ω is a collection of subsets closed under complementation and countable unions, providing the mathematical structure needed to define probability measures rigorously. It determines which subsets can be assigned probabilities in a probability space. Sigma-algebras replace naive intuitions about 'all' subsets with a carefully chosen closure structure.
You know from set operations that unions, intersections, and complements let you build new sets from old ones. And from studying cardinality, you know that infinite sets come in radically different sizes — countably infinite like ℕ, or uncountably infinite like ℝ. A σ-algebra (sigma-algebra) combines both ideas: it is a collection of sets closed under complementation and under *countably infinite* unions, not just finite ones. The word "sigma" signals countability — the same prefix as in "sigma-notation" for sums.
Formally, a collection ℱ of subsets of a sample space Ω is a σ-algebra if: (1) Ω ∈ ℱ, (2) if A ∈ ℱ then the complement Ac ∈ ℱ, and (3) if A₁, A₂, A₃, ... is any countably infinite sequence with each Aᵢ ∈ ℱ, then ∪_{i=1}^∞ Aᵢ ∈ ℱ. From (2) and (3) together, De Morgan's laws give closure under countable intersections as well. Why require only *countable* unions rather than arbitrary unions? Because uncountably infinite unions can produce pathological sets that resist any consistent measurement. Requiring only countable closure keeps the algebra broad enough for all practical probability while excluding the paradoxes.
The pathology at stake is concrete. A remarkable result — the Vitali set construction, assuming the axiom of choice — shows that you cannot consistently assign lengths to *all* subsets of [0,1] while preserving three natural properties: non-negativity, countable additivity, and translation invariance. The resolution is not to measure everything, but to designate in advance a collection of "measurable" sets and work only with those. The σ-algebra is that designation. A set is measurable if and only if it belongs to the σ-algebra; probability is only defined for measurable sets.
Two extreme examples anchor the concept. The trivial σ-algebra {∅, Ω} is the smallest possible: only the impossible event and the certain event are measurable. It carries zero information about the sample space. The power set 2^Ω of all subsets is the largest σ-algebra and works fine when Ω is finite or countably infinite. For the real line, the correct choice is the Borel σ-algebra, generated by all open intervals — it includes all open, closed, and half-open sets, all singletons, all countable sets, and much more, while the Vitali sets are nowhere to be found. The Borel σ-algebra on ℝ is the working foundation for all measure-theoretic probability you will encounter in subsequent courses.