From set systems to σ-algebras, Borel sets, and the structure that makes measurement possible
A σ-algebra (sigma-algebra) on a set X is a collection F of subsets of X that is closed under complementation and countable unions, and that contains X itself. More precisely, F must satisfy three axioms: (1) X ∈ F, (2) if A ∈ F then X \ A ∈ F, and (3) if A₁, A₂, A₃, … ∈ F then ∪ₙ Aₙ ∈ F. These three simple requirements have far-reaching consequences for what we can measure.
The idea behind a σ-algebra is to specify which subsets of X we consider "measurable" — that is, which subsets we are allowed to assign a size or probability to. Not every collection of subsets works: if we want our measure to behave well under natural set operations, we need the closure properties that a σ-algebra provides. Without them, contradictions arise (as we will see in the lesson on non-measurable sets).
The simplest σ-algebra on any set X is {∅, X}, called the trivial σ-algebra. The largest is the power set P(X), which contains every subset. Between these extremes live the σ-algebras that matter most in practice, including the Borel σ-algebra on ℝ and the Lebesgue σ-algebra.
The three axioms of a σ-algebra immediately imply several additional closure properties. Since F is closed under complements and countable unions, De Morgan's laws give us closure under countable intersections as well: if each Aₙ ∈ F, then ∩ₙ Aₙ = (∪ₙ Aₙᶜ)ᶜ ∈ F. We also get closure under set difference (A \ B = A ∩ Bᶜ) and symmetric difference.
The requirement of countable unions — rather than just finite ones — is what distinguishes a σ-algebra from a mere algebra of sets. This distinction is crucial for analysis. Many natural constructions in analysis involve taking limits, and limits correspond to countable operations on sets. For instance, a set where a sequence of functions converges is expressed as a countable intersection of countable unions, which requires σ-algebra structure.
An important technical fact is that the intersection of any family of σ-algebras is again a σ-algebra. This allows us to define the σ-algebra generated by a collection of sets C as the smallest σ-algebra containing C — namely, the intersection of all σ-algebras that contain C. This generated σ-algebra, denoted σ(C), is a foundational construction in measure theory.
Use the interactive demonstration below to build σ-algebras on small finite sets. Add sets to your collection and watch the closure properties enforce themselves — the demo will automatically include all complements, unions, and intersections required to form a valid σ-algebra.
Key insight: Even on a small finite set, the σ-algebra generated by a single non-trivial subset already contains four elements. On infinite sets, generated σ-algebras can be enormously complex — the Borel σ-algebra on ℝ cannot be explicitly listed, yet every open set, closed set, countable intersection of open sets (Gδ set), and countable union of closed sets (Fσ set) belongs to it.
The Borel σ-algebra on ℝ, denoted B(ℝ), is the σ-algebra generated by the open subsets of ℝ. Equivalently, it is generated by the open intervals (a, b), or by the half-lines (−∞, a], or by any other collection that generates the standard topology. Every set you encounter in everyday analysis — open sets, closed sets, countable sets, intervals, Gδ sets, Fσ sets — is a Borel set.
Despite its vastness, the Borel σ-algebra is actually "small" in a precise sense: it has the same cardinality as ℝ (namely, 2^ℵ₀ = 𝔠), while the power set of ℝ has cardinality 2^𝔠. This means there are far more subsets of ℝ than Borel sets. The Lebesgue σ-algebra is strictly larger than B(ℝ), containing certain non-Borel sets that are still Lebesgue-measurable.
Borel σ-algebras can be defined on any topological space, not just ℝ. In ℝⁿ, the Borel σ-algebra is generated by open rectangles (products of open intervals). In probability theory, the Borel σ-algebra on [0, 1] serves as the standard framework for defining random variables and their distributions.
One might wonder: why not just measure every subset of ℝ? The answer is that this is impossible if we want our measure to satisfy three natural properties simultaneously: (1) the measure of an interval [a, b] equals b − a, (2) the measure is translation-invariant, and (3) the measure is countably additive. Vitali's construction (1905) shows that under the Axiom of Choice, there exist subsets of ℝ that cannot be assigned a measure consistent with these three properties.
The σ-algebra provides a principled way to restrict attention to those subsets thatcan be measured consistently. By working within a σ-algebra, we ensure that all our set-theoretic operations — unions, intersections, complements, limits — stay within the realm of measurable sets. This is the trade-off at the heart of measure theory: we sacrifice universality (measuring every set) to gain consistency (a well-behaved measure).
In probability theory, the σ-algebra plays an equally important role: it represents the collection of events to which we can assign probabilities. A random variable is measurable precisely when the preimage of every Borel set is an event in our σ-algebra. This connection between measure theory and probability is one of the great unifying themes of modern mathematics.