nLab standard Borel space

Contents

Idea

Standard Borel spaces are measurable spaces whose sigma-algebras arise as the Borel sigma-algebras of Polish spaces.

They provide a setting for many constructions in probability theory, including conditional probability, disintegration, and countable products.

From the point of view of categorical probability, standard Borel spaces and Markov kernels form the Markov category BorelStoch.

Definitions

Definition using a Polish topology

A topological space is called Polish if its topology is induced by some complete and separable metric.

A measurable space (X,Σ X)(X,\Sigma_X) is a standard Borel space if there exists a Polish topology on XX whose Borel sigma-algebra is exactly Σ X\Sigma_X.

Equivalently, (X,Σ X)(X,\Sigma_X) is isomorphic, as a measurable space, to a Polish space equipped with its Borel sigma-algebra. The choice of a Polish topology is not part of the structure of a standard Borel space.

A measure space is called a standard Borel measure space if its underlying measurable space is standard Borel. In particular, a standard Borel probability space is a standard Borel space equipped with a probability measure. Completeness of the measure is not required.

Characterization using finite Boolean algebras

There is also a characterization in terms of finite Boolean algebras of measurable subsets.

A field of subsets of XX is a Boolean algebra of subsets of XX, containing ∅\varnothing and XX and closed under finite unions and complements relative to XX.

The Boolean algebra generated by finitely many sets S 0,…,S n−1⊆XS_0,\ldots,S_{n-1}\subseteq X is finite. Its atoms are the nonempty sets of the form

⋂ i=0 n−1S i ⋆, \bigcap_{i=0}^{n-1}S_i^\star,

where each S i ⋆S_i^\star is either S iS_i or X∖S iX\setminus S_i. Every element of this algebra is a union of atoms. The algebra generated by the S iS_i should be distinguished from the generating family {S 0,…,S n−1}\{S_0,\ldots,S_{n-1}\} itself.

A family of subsets separates points if, for every two distinct points, some member of the family contains exactly one of them.

Proposition. A measurable space (X,Σ X)(X,\Sigma_X) is standard Borel if and only if there exists a sequence of finite Boolean subalgebras ℱ n⊆Σ X\mathcal{F}_n \subseteq \Sigma_X, indexed by n≥0n\geq 0, with the following properties:

  1. ℱ n⊆ℱ n+1\mathcal{F}_n\subseteq\mathcal{F}_{n+1} for every nn;

  2. Σ X=σ(⋃ n≥0ℱ n)\Sigma_X=\sigma\left(\bigcup_{n\geq 0}\mathcal{F}_n\right);

  3. ⋃ n≥0ℱ n\bigcup_{n\geq 0}\mathcal{F}_n separates points of XX;

  4. whenever A nA_n is an atom of ℱ n\mathcal{F}_n for each nn and A n+1⊆A nA_{n+1}\subseteq A_n for every nn, one has

    ⋂ n≥0A n≠∅. \bigcap_{n\geq 0}A_n\neq\varnothing.

Under condition 3, the intersection in condition 4 contains at most one point. Thus conditions 3 and 4 can equivalently be replaced by the requirement that every such decreasing sequence of atoms have intersection consisting of exactly one point.

Proof. Suppose first that these conditions hold and that XX is nonempty. Let P nP_n be the finite set of atoms of ℱ n\mathcal{F}_n, equipped with the discrete topology, and put

K={(A n) n≥0∈∏ n≥0P nmiddle|A n+1⊆A n for every n}. K= \left\{ (A_n)_{n\geq 0}\in\prod_{n\geq 0}P_n \;\middle|\; A_{n+1}\subseteq A_n\text{ for every }n \right\}.

The compatibility conditions are closed conditions, so KK is a closed subspace of a countable product of finite discrete spaces. Hence KK is compact metrizable, and in particular Polish.

Send x∈Xx\in X to the sequence of atoms containing it. Point separation makes this map injective, and condition 4 makes it surjective onto KK. The coordinate cylinders generate the Borel sigma-algebra of KK, and their inverse images are the atoms of the ℱ n\mathcal{F}_n. Condition 2 therefore makes this bijection an isomorphism of measurable spaces.

Conversely, the classification of standard Borel spaces recalled below implies that every nonempty standard Borel space is measurably isomorphic to one of the following compact metrizable spaces:

  • a finite discrete space;
  • the countably infinite space {0}∪{1/m:m≥1}\{0\}\cup\{1/m:m\geq 1\};
  • Cantor space.

Each of these spaces has a countable basis of clopen sets. Enumerating that basis and taking the Boolean algebras generated by its first finitely many members gives conditions 1–3. Condition 4 follows from compactness, since the atoms are nonempty closed sets.

The empty space also satisfies the characterization, using the constant algebra {∅}\{\varnothing\}; the condition on sequences of atoms is then vacuous.

Terminology and the separation condition

Following Gray, an increasing sequence of finite fields whose union is ℱ\mathcal{F} is called a basis for the field ℱ\mathcal{F} if every decreasing sequence of atoms, one from each field, has nonempty intersection.

In this terminology, the preceding proposition says that a measurable space is standard Borel precisely when its sigma-algebra is generated by a point-separating field possessing a basis.

The separation condition is essential. For example, let X={a,b}X=\{a,b\} and Σ X={∅,X}\Sigma_X=\{\varnothing,X\}. The constant sequence ℱ n=Σ X\mathcal{F}_n=\Sigma_X has the nonempty-intersection property, but (X,Σ X)(X,\Sigma_X) is not standard Borel: its singletons are not measurable.

Gray’s term standard space, defined by the existence of a generating field with a basis, does not impose point separation and is therefore broader than standard Borel space as defined here.

Preston likewise distinguishes standard Borel spaces from more general measurable spaces. His Theorem 6.1 characterizes standard Borel spaces as separable type B spaces. Here his measurable-space use of separable means that every singleton is measurable. For countably generated measurable spaces, this is equivalent to point separation.

Examples

  • The real line, equipped with its Borel sigma-algebra, is standard Borel.

  • Every finite or countably infinite set, equipped with the discrete sigma-algebra consisting of all its subsets, is standard Borel. This includes the empty set.

  • Every Borel subset of a Polish space, equipped with the induced sigma-algebra, is standard Borel. Its subspace topology need not itself be Polish; another Polish topology can induce the same sigma-algebra.

The Borel isomorphism theorem states that any two uncountable standard Borel spaces are isomorphic as measurable spaces. Consequently, up to measurable isomorphism, standard Borel spaces are classified by cardinality: there is one for each finite cardinality, one countably infinite example, and one uncountable example.

In particular, ℝ n\mathbb{R}^n for 1≤n<∞1\leq n \lt \infty, Cantor space, and every real interval containing more than one point are measurably isomorphic to ℝ\mathbb{R}.

Categories of standard Borel spaces

Properties

Measurable-space properties

  • Finite and countable products of standard Borel spaces are standard Borel, with the product sigma-algebra, as in Meas.

  • Finite and countable disjoint unions of standard Borel spaces are standard Borel, with the coproduct sigma-algebra.

  • Measurable subsets of standard Borel spaces, with the induced sigma-algebra, are standard Borel.

  • Measurable retracts of standard Borel spaces are standard Borel.

  • The sigma-algebra of a standard Borel space is countably generated and separates points. In particular, every singleton is measurable.

  • Every measurable bijection between standard Borel spaces has a measurable inverse.

Conditional distributions and Bayesian inversion

Let (X,Σ X,μ)(X,\Sigma_X,\mu) be a standard Borel probability space and let 𝒢⊆Σ X\mathcal{G}\subseteq\Sigma_X be any sub-sigma-algebra. There exists a regular conditional distribution

r:(X,𝒢)⟶(X,Σ X) r:(X,\mathcal{G})\longrightarrow(X,\Sigma_X)

such that, for every A∈Σ XA\in\Sigma_X and G∈𝒢G\in\mathcal{G},

∫ Gr(x,A)μ(dx)=μ(G∩A). \int_G r(x,A)\,\mu(dx)=\mu(G\cap A).

Thus x↦r(x,A)x\mapsto r(x,A) is a version of the conditional expectation of the indicator of AA given 𝒢\mathcal{G}. No countable-generation assumption on 𝒢\mathcal{G} is needed.

In particular, every Markov kernel k:X→Yk:X\to Y between standard Borel spaces admits a Bayesian inverse relative to any probability measure μ\mu on XX. Writing

ν(B)=∫ Xk(x,B)μ(dx) \nu(B)=\int_X k(x,B)\,\mu(dx)

for the induced probability measure on YY, there is a Markov kernel k †:Y→Xk^\dagger:Y\to X satisfying

∫ Ak(x,B)μ(dx)=∫ Bk †(y,A)ν(dy) \int_A k(x,B)\,\mu(dx) = \int_B k^\dagger(y,A)\,\nu(dy)

for all measurable A⊆XA\subseteq X and B⊆YB\subseteq Y. The inverse is unique up to ν\nu-almost sure equality.

Countable tensor products and the Giry monad

  • The category BorelStoch admits countable infinite tensor products, given on objects by the usual countable products of measurable spaces. The corresponding extension property for probability measures is supplied by the Kolmogorov extension theorem.

  • The Giry monad preserves standard Borel spaces. More explicitly, if XX is standard Borel, let PXP X be the set of probability measures on XX, equipped with the sigma-algebra generated by the evaluation maps

    μlongmapstoμ(A),A∈Σ X. \mu\longmapsto\mu(A), \qquad A\in\Sigma_X.

    Then PXP X is standard Borel. Consequently, the Giry monad restricts to a monad on BorelMeas, whose Kleisli category is BorelStoch.

Characterizations of BorelMeas

The Loomis–Sikorski duality theorem gives a contravariant equivalence between BorelMeas and the category of countably presented countably complete Boolean algebras, with homomorphisms preserving countable joins. A standard Borel space is sent to its sigma-algebra of measurable subsets, and a measurable map is sent to its inverse-image homomorphism.

Using this duality, Chen proves that BorelMeas is biinitial in the 2-category of countably complete, countably extensive Boolean categories, with functors preserving countable limits and countable coproducts.

See:

References

For standard Borel spaces, their classification, and the distinction from more general type B spaces:

  • Chris Preston, Some Notes on Standard Borel and Related Spaces (2008), arXiv:0809.3066. See especially the introduction and Sections 2–6; Section 12 discusses conditional distributions.

For the finite-field basis approach and its applications to probability theory:

  • Robert M. Gray, Probability, Random Processes, and Ergodic Properties, second edition, Springer, 2009. A revised first edition is freely available from the author; its Sections 2.2 and 3.1 distinguish the notions of standard space and standard Borel space.

For the categorical treatment of probability-preserving kernels and Bayesian inversion:

  • Fredrik Dahlqvist, Alexandra Silva, Vincent Danos, and Ilias Garnier, Borel Kernels and their Approximation, Categorically, Electronic Notes in Theoretical Computer Science 341 (2018), 91–119, doi:10.1016/j.entcs.2018.11.006.

For a discussion of the Giry monad on standard Borel spaces:

  • Kirk Sturtz, Deriving the Giry algebras on standard Borel spaces using ℝ ∞\mathbb{R}_{\infty}-generalized points, arXiv:2409.14861v5 (2025).
category: probability

Last revised on October 23, 2026 at 07:51:01. See the history of this page for a list of all contributions to it.