Compact-open topology
The compact-open topology is a topology defined on the set (C(X,Y)) of continuous maps between two topological spaces (X) and (Y). It controls a function by restricting its values on compact subsets of the domain, thereby occupying an intermediate position between the topology of pointwise convergence and topologies based on uniform convergence over the entire domain.
For a compact subset (K\subseteq X) and an open subset (U\subseteq Y), let
[ [K,U]={f\in C(X,Y)\mid f(K)\subseteq U}. ]
The sets ([K,U]) form a subbasis for the compact-open topology. A basic neighborhood of a map (f) is therefore a finite intersection
[ [K_1,U_1]\cap\cdots\cap[K_n,U_n], ]
where (f(K_i)\subseteq U_i) for each index (i). The resulting function space is commonly denoted by (C_k(X,Y)) when the distinction from other function-space topologies is relevant.
Interpretation through convergence
The compact-open topology records simultaneous control on every compact portion of the domain. A net ((f_\alpha)) converges to (f) in this topology precisely when it eventually satisfies every finite collection of compact-image conditions defining a neighborhood of (f).
When (Y) is a uniform space, the compact-open topology is closely related to uniform convergence on compact subsets. For each compact (K\subseteq X) and each entourage (V) of the uniformity on (Y), the condition
[ (f_\alpha(x),f(x))\in V\qquad\text{for every }x\in K ]
describes convergence uniformly over (K). Under the standard compatibility hypotheses for the topology and uniformity of (Y), these conditions generate the same topology as the compact-open subbasis.
For a metric space ((Y,d)), the corresponding neighborhoods can be expressed through the quantities
[ d_K(f,g)=\sup_{x\in K}d(f(x),g(x)). ]
Each (d_K) is a pseudometric on (C(X,Y)). The family of these pseudometrics, indexed by the compact subsets of (X), generates the topology of uniform convergence on compact subsets. If (X) itself is compact, the single supremum metric
[ d_\infty(f,g)=\sup_{x\in X}d(f(x),g(x)) ]
generates the compact-open topology.
The compact-open topology is finer than the topology of pointwise convergence because every singleton subset of (X) is compact. The two topologies agree when every compact subset relevant to the mapping problem is controlled by finitely many points, as occurs for a discrete domain whose compact subsets are finite. For a noncompact domain, the compact-open topology is generally weaker than a topology imposing uniform convergence over all of (X).
Historical development
The modern formulation emerged in 1945 through Ralph H. Fox’s systematic treatment of topologies on mapping spaces and You Watanabe’s companion analysis of convergence uniformly on compact subsets. Their formulation identified the sets ([K,U]) as the natural neighborhood data for homotopy-theoretic function spaces and clarified the relationship between compact-image conditions and uniform structures on the codomain.
This treatment consolidated earlier uses of function-space convergence into a topology defined without requiring a metric. The term “compact-open” refers directly to the compact subset (K) in the domain and the open subset (U) in the codomain that determine each subbasic set.
Separation and functoriality
If (Y) is a Hausdorff space, then (C_k(X,Y)) is Hausdorff. Distinct maps (f) and (g) differ at some point (x\in X), and the compact set ({x}) permits disjoint subbasic neighborhoods to be obtained from disjoint neighborhoods of (f(x)) and (g(x)).
The construction is contravariant in the domain. Given a continuous map (h:X'\to X), precomposition defines
[ h^\ast:C_k(X,Y)\longrightarrow C_k(X',Y),\qquad h^\ast(f)=f\circ h. ]
This map is continuous because the image of a compact subset under (h) is compact. In particular,
[ (h^\ast)^{-1}[K',U]=[h(K'),U]. ]
The construction is covariant in the codomain. For a continuous map (q:Y\to Y'), postcomposition gives a continuous map
[ q_\ast:C_k(X,Y)\longrightarrow C_k(X,Y'),\qquad q_\ast(f)=q\circ f, ]
since the inverse image of ([K,U']) is ([K,q^{-1}(U')]). These two operations make compact-open function spaces compatible with the ordinary composition structure of the category of topological spaces.
Evaluation and the exponential law
A central structural map is the evaluation map
[ \operatorname{ev}:C_k(X,Y)\times X\longrightarrow Y,\qquad \operatorname{ev}(f,x)=f(x). ]
When (X) is locally compact and Hausdorff, evaluation is continuous. Local compactness supplies compact neighborhoods on which a function can be constrained through compact-open subbasic sets, while the Hausdorff condition provides the regularity needed to fit those compact neighborhoods inside prescribed open sets.
Under the same domain hypothesis, the compact-open topology supports the standard exponential law. For suitable spaces (Z), the correspondence
[ F:Z\times X\longrightarrow Y \quad\longleftrightarrow\quad \widehat F:Z\longrightarrow C_k(X,Y), ]
defined by (\widehat F(z)(x)=F(z,x)), is an identification at the level of continuous maps and, under the usual compact-open hypotheses, at the level of function spaces:
[ C_k(Z\times X,Y)\cong C_k!\left(Z,C_k(X,Y)\right). ]
This relation explains the topology’s role in homotopy theory, where a homotopy (X\times I\to Y) can be represented as a path (I\to C_k(X,Y)). The correspondence depends on the continuity of evaluation rather than merely on the set-theoretic equivalence between curried and uncurried functions.
Homeomorphism groups
For a space (X), the set (\operatorname{Homeo}(X)) of self-homeomorphisms inherits a topology as a subspace of (C_k(X,X)). Composition is not automatically continuous for arbitrary (X), because the compact subsets used to control one factor can vary under the other factor.
Richard Arens analyzed the hypotheses under which the compact-open topology gives homeomorphism spaces the structure of a topological group. For locally compact Hausdorff spaces, composition behaves continuously under the standard compact-open assumptions. Continuity of inversion requires additional control in general, and it holds in widely used settings that include compact Hausdorff spaces. A related topology can be obtained by embedding a homeomorphism (f) through the pair ((f,f^{-1})) in
[ C_k(X,X)\times C_k(X,X), ]
which records compact-open behavior of both the map and its inverse.
Compactly generated spaces
The category of all topological spaces is not cartesian closed when its mapping spaces are equipped only with the unmodified compact-open topology. In particular, the expected exponential law can fail when the domain lacks suitable local compactness properties.
John L. Kelley’s treatment of compactly generated spaces and Norman Steenrod’s subsequent categorical formulation replaced a topology by its final topology with respect to maps from compact Hausdorff spaces. Applied to mapping spaces, this construction produces the compactly generated refinement
[ \operatorname{Map}(X,Y)=k!\left(C_k(X,Y)\right), ]
where (k) denotes Kelleyfication. Within the category of compactly generated Hausdorff spaces, products and mapping spaces are adjusted by this operation, and the exponential correspondence becomes an internal categorical identity. The compact-open topology remains the initial function-space topology in this construction, while Kelleyfication enforces the compact-detection principle required by the ambient category.