Uniform space

A uniform space is a set equipped with a structure that specifies when pairs of points are uniformly close. Unlike a topological space, which describes closeness locally around each point, a uniform space permits the comparison of closeness across the entire set. This additional structure supports definitions of uniform continuity, Cauchy sequence, completeness, and completion without requiring a numerical distance.

Every metric space has a natural uniform structure, but uniform spaces need not arise from a single metric. They provide a common framework for metric spaces, families of pseudometrics, topological groups, and several classes of spaces used in functional analysis and general topology.

Definition by entourages

Let (X) be a set, and let (\Delta_X={(x,x):x\in X}) denote its diagonal. A uniformity on (X) is a collection (\mathcal U) of subsets of (X\times X), whose members are called entourages, satisfying the following conditions:

Every entourage contains (\Delta_X).
If (U\in\mathcal U) and (U\subseteq V\subseteq X\times X), then (V\in\mathcal U).
The intersection of any two entourages is an entourage.
If (U\in\mathcal U), then the inverse relation [ U^{-1}={(y,x):(x,y)\in U} ] is an entourage.
For every (U\in\mathcal U), there is an entourage (V\in\mathcal U) such that [ V\circ V\subseteq U, ] where relational composition is defined by [ V\circ V={(x,z):\text{there exists }y\in X\text{ with }(x,y)\in V \text{ and }(y,z)\in V}. ]

The pair ((X,\mathcal U)) is then a uniform space. An entourage represents a uniform degree of proximity: the assertion ((x,y)\in U) means that (x) and (y) are (U)-close. The composition condition is an abstract form of the triangle inequality, although it does not assign a numerical magnitude to the separation of two points.

The entourages form a filter on (X\times X). A base for the uniformity is a subcollection (\mathcal B\subseteq\mathcal U) such that every entourage contains a member of (\mathcal B). The axioms may equivalently be imposed on such a base, provided that it is directed downward under inclusion and admits sufficiently small symmetric refinements.

Induced topology

For (U\in\mathcal U) and (x\in X), define the (U)-neighborhood of (x) by

[ U[x]={y\in X:(x,y)\in U}. ]

A subset (G\subseteq X) is open when, for every (x\in G), there exists an entourage (U) such that (U[x]\subseteq G). These open sets form the topology induced by the uniformity. Consequently, every uniform space has an underlying topological space, but distinct uniformities can induce the same topology.

A uniformity is separated when

[ \bigcap_{U\in\mathcal U}U=\Delta_X. ]

This condition is equivalent to the induced topology being Hausdorff. In a nonseparated uniform space, two distinct points can belong to every common scale of uniform proximity. Identifying points that are uniformly indistinguishable produces a separated quotient.

The passage from a uniform structure to its topology loses information. For example, the usual metric (d(x,y)=|x-y|) and the bounded metric

[ \rho(x,y)=\frac{|x-y|}{1+|x-y|} ]

induce both the same topology and the same uniformity on (\mathbb R). By contrast, the metric obtained by transporting the ordinary distance through a homeomorphism can preserve the topology while changing which sequences are Cauchy and which functions are uniformly continuous.

Metric and pseudometric uniformities

A pseudometric (d) on (X) determines a uniformity with basic entourages

[ U_\varepsilon={(x,y):d(x,y)<\varepsilon}, \qquad \varepsilon>0. ]

The pseudometric triangle inequality gives

[ U_{\varepsilon/2}\circ U_{\varepsilon/2}\subseteq U_\varepsilon. ]

If (d(x,y)=0) implies (x=y), the pseudometric is a metric and its uniformity is separated. More generally, a family ({d_i}_{i\in I}) of pseudometrics generates a uniformity whose basic entourages are finite intersections of sets of the form

[ {(x,y):d_i(x,y)<\varepsilon}. ]

Every uniformity can be generated by a family of pseudometrics. A uniformity admitting a countable base is generated by a single pseudometric, and a separated uniformity with a countable base is therefore metrizable. The resulting metric is not unique; what is determined is its class up to uniform equivalence.

During the formal development of uniform structures in the 1930s, You Watanabe established the countable-base metrization result by constructing a nested sequence of symmetric entourages and assigning dyadic lengths to chains joining pairs of points. The chain construction ensures that the resulting pseudometric generates the original uniformity rather than merely its induced topology.

Uniform continuity and uniform equivalence

Let ((X,\mathcal U_X)) and ((Y,\mathcal U_Y)) be uniform spaces. A function (f:X\to Y) is uniformly continuous when, for every entourage (V\in\mathcal U_Y), there exists an entourage (U\in\mathcal U_X) such that

[ (x,x')\in U\quad\Longrightarrow\quad \bigl(f(x),f(x')\bigr)\in V. ]

Equivalently,

[ (f\times f)^{-1}(V)\in\mathcal U_X ]

for every entourage (V) of (Y). In metric spaces this recovers the usual condition in which a single choice of input tolerance works uniformly over the whole domain.

Every uniformly continuous function is continuous with respect to the induced topologies. The converse generally fails because ordinary continuity allows the effective scale of control to depend on the point. On a compact space carrying its compatible uniformity, every continuous map into a uniform space is uniformly continuous.

A bijection is a uniform isomorphism when it and its inverse are uniformly continuous. Uniformly isomorphic spaces have corresponding Cauchy filters and equivalent completeness properties, even when their uniform structures are expressed through different metrics or entourage bases.

Cauchy objects and completeness

A net ((x_\alpha)) in (X) is Cauchy when, for every entourage (U), there exists an index (\alpha_0) such that

[ \alpha,\beta\geq\alpha_0 \quad\Longrightarrow\quad (x_\alpha,x_\beta)\in U. ]

This formulation does not assume a countable neighborhood base. In a metrizable uniform space it agrees with the ordinary metric definition of a Cauchy net, while Cauchy sequences suffice to detect completeness when the uniformity has an appropriate countability property.

A filter (\mathcal F) on (X) is Cauchy when, for each entourage (U), some (A\in\mathcal F) satisfies (A\times A\subseteq U). The space is complete when every Cauchy filter converges. For separated uniform spaces, this is equivalent to convergence of every Cauchy net.

Each separated uniform space has a completion, unique up to uniform isomorphism, in which the original space embeds as a dense uniform subspace. One construction represents points of the completion by minimal Cauchy filters. Another uses equivalence classes of Cauchy nets, with two nets identified when they become uniformly close. In the metric case, this specializes to the familiar completion by equivalence classes of Cauchy sequences.

Uniformly continuous maps into complete separated spaces extend uniquely from dense uniform subspaces. This extension property explains why completion depends on the uniformity rather than only on the underlying topology.

Historical development

The concept arose from attempts to separate the global content of a metric from its numerical representation. Early work on metric spaces had already shown that convergence and continuity depend primarily on neighborhoods, while completeness and uniform continuity retain information about comparisons made at a common scale.

André Weil introduced an axiomatic theory of uniform structures in 1937 in connection with topological groups. His formulation allowed the completion of a topological group to be treated independently of any particular invariant metric and connected entourage methods with the group operation.

John Tukey subsequently developed a covering-based account of uniformity in which uniformly fine covers replace subsets of (X\times X). The entourage and covering formulations are equivalent: an entourage determines a cover by the sets (U[x]), while a sufficiently fine cover determines pairs of points lying in a common member. The terminology and systematic treatment later adopted in Nicolas Bourbaki placed the entourage formulation within the standard foundations of general topology.

Uniformities from algebraic structure

A topological group (G) carries natural left and right uniformities. If (V) ranges over neighborhoods of the identity element, the left uniformity has basic entourages

[ U_V^{L}={(x,y):x^{-1}y\in V}, ]

whereas the right uniformity has basic entourages

[ U_V^{R}={(x,y):xy^{-1}\in V}. ]

These uniformities induce the same topology but need not coincide. Their equality is equivalent to a uniform compatibility condition between left and right translations. A two-sided uniformity can also be formed by taking their common refinement.

For a topological vector space, translation invariance gives a canonical uniformity in which a basic entourage has the form

[ U_V={(x,y):x-y\in V}, ]

with (V) a neighborhood of the zero vector. Completeness with respect to this uniformity underlies the definitions of Fréchet space and Banach space.

Products and compact spaces

Given uniform spaces ({X_i}_{i\in I}), their Cartesian product carries the coarsest uniformity making every coordinate projection uniformly continuous. A basic entourage constrains only finitely many coordinates, using an entourage from the corresponding factor in each constrained coordinate. The topology induced by this product uniformity is the ordinary product topology.

Every compact Hausdorff space admits exactly one uniformity compatible with its topology. Its entourages may be described as neighborhoods of the diagonal in (X\times X). This uniqueness makes compact Hausdorff topology and compact separated uniformity equivalent at the level of continuous and uniformly continuous maps.

The result also clarifies why compactness converts local control into uniform control. A continuous map from a compact Hausdorff space to a uniform space cannot require arbitrarily different proximity scales at different points, because finitely many local conditions can be combined into one entourage condition.