A Borel set in a topological space is a set belonging to the smallest sigma-algebra that contains all the space’s open sets. Equivalently, Borel sets are generated from open sets by complements and countable unions, with countable intersections also permitted because they follow from these operations. They connect topology with measure theory: topology describes continuity and proximity, while Borel sets provide a natural collection of sets on which measures and probabilities can be defined. They are named after the French mathematician Émile Borel.
Definition
Let be a set equipped with a topology . Its Borel sigma-algebra is
A subset is Borel precisely when .
Thus contains every open set and satisfies:
If , then .
If for every positive integer , then
Consequently,
It also contains the empty set and . The word smallest is essential: other sigma-algebras may contain all open sets, but the Borel sigma-algebra includes only sets forced by these closure requirements.
Every closed set is Borel because its complement is open. Conversely, closed sets generate the same sigma-algebra as open sets.
Borelness depends on the topology. For the discrete topology, every subset is open and hence Borel. For the indiscrete topology , the only Borel sets are and .
Examples on the real line
On the real line with its usual topology, all open, closed, and half-open intervals are Borel. For example,
Every singleton is closed, so every countable subset of is Borel. In particular, the rational numbers form a Borel set:
The irrational numbers are Borel as the complement of .
An set is a countable union of closed sets, and a set is a countable intersection of open sets. Both classes consist entirely of Borel sets. The rationals are , and the irrationals are . The Cantor set is Borel because it is closed.
All open subsets of are countable unions of intervals with rational endpoints. Consequently,
Other useful generating families include the rays , either for all real or just for rational . These descriptions are central to defining real-valued random variables and their distributions.
Closure properties and their limits
Borel sets are closed under finite unions, finite intersections, set differences, symmetric differences, and countable unions and intersections. They need not be closed under uncountable unions or intersections.
For example, every subset can be written as
Although each singleton is Borel, some subsets of are not. The restriction to countable operations is therefore substantive.
The relationship with subspaces is particularly useful. If carries the subspace topology, then
This holds even when is not Borel in . If is Borel in , every Borel subset of is also Borel as a subset of .
The Borel hierarchy
The Borel hierarchy organizes Borel sets according to the complexity of their construction from open sets. It begins with
At the next level, consists of sets and consists of sets.
More generally, for a countable ordinal , define
and let consist of the complements of sets in . The notation
denotes sets belonging to both classes.
Every Borel set occurs at some countable level. Thus, “generated by countable operations” does not mean that a fixed finite number of construction stages suffices. Some Borel sets require transfinite stages indexed by countable ordinals.
In uncountable Polish spaces—separable topological spaces admitting a compatible complete metric—the hierarchy does not collapse at any countable level. Sets of arbitrarily high countable Borel complexity exist.
Relation to measures and Lebesgue measurability
A Borel measure on is a measure defined on . Being Borel is a topological property; it does not depend on the choice of a measure.
On Euclidean space, every Borel set is Lebesgue measurable, but the converse is false. The Lebesgue sigma-algebra is the completion of the Borel sigma-algebra with respect to Lebesgue measure: it additionally includes every subset of every Borel null set.
The Cantor set illustrates the distinction. It is a Borel set of Lebesgue measure zero, so all its subsets are Lebesgue measurable. However, some of those subsets are not Borel. Indeed, the Cantor set has cardinality , the cardinality of , and hence has subsets, whereas has only Borel sets.
More generally, every Lebesgue measurable set differs from some Borel set by a null set:
where denotes symmetric difference. Thus, Lebesgue measurable sets can be represented by Borel sets up to a measure-zero discrepancy, even though the two classes are not identical.
Functions, inverse images, and projections
A function between topological spaces is Borel measurable if
It is enough to verify this condition for all open subsets of . Every continuous function is Borel measurable, since inverse images of open sets are open. This places continuity within the broader framework of measurable functions.
Direct images behave differently. Even a continuous image of a Borel set need not be Borel. In particular, a coordinate projection of a Borel subset of a product of Polish spaces can fail to be Borel.
Such projections are analytic sets, an important class extending beyond the Borel sets. Every Borel subset of a Polish space is analytic, but some analytic sets are not Borel. In contrast to arbitrary continuous maps, a homeomorphism preserves Borel sets in both directions.
Role in probability and measurable structures
For a probability space , a real-valued random variable is a measurable map
Its distribution is the probability measure on Borel sets defined by
The inverse image is therefore an event whenever is Borel. The half-lines suffice to specify the distribution through its cumulative distribution function.
A standard Borel space is a measurable space isomorphic, by a measurable bijection with measurable inverse, to a Borel subset of a Polish space. This notion retains the measurable structure while discarding the particular topology used to present it. Standard Borel spaces provide a broad setting for probability theory, including random elements taking values in spaces more general than the real line.