Cartesian closed category
In category theory, a category is cartesian closed if, roughly speaking, any morphism defined on a product of two objects can be naturally identified with a morphism defined on one of the factors. These categories are particularly important in mathematical logic and the theory of programming, in that their internal language is the simply typed lambda calculus. They are generalized by closed monoidal categories, whose internal language, linear type systems, are suitable for both quantum and classical computation.
- It has a terminal object.
- Any two objects X and Y of C have a product X×Y in C.
- Any two objects Y and Z of C have an exponential ZY in C.
The first two conditions can be combined to the single requirement that any finite (possibly empty) family of objects of C admit a product in C, because of the natural associativity of the categorical product and because the empty product in a category is the terminal object of that category.
The third condition is equivalent to the requirement that the functor –×Y (i.e. the functor from C to C that maps objects X to X×Y and morphisms φ to φ×idY) has a right adjoint, usually denoted –Y, for all objects Y in C. For locally small categories, this can be expressed by the existence of a bijection between the hom-sets
which is natural in both X and Z.
If a category is such that all its slice categories are cartesian closed, then it is called locally cartesian closed.
Examples of cartesian closed categories include:
- The category Set of all sets, with functions as morphisms, is cartesian closed. The product X×Y is the cartesian product of X and Y, and ZY is the set of all functions from Y to Z. The adjointness is expressed by the following fact: the function f : X×Y → Z is naturally identified with the curried function g : X → ZY defined by g(x)(y) = f(x,y) for all x in X and y in Y.
- The category of finite sets, with functions as morphisms, is cartesian closed for the same reason.
- If G is a group, then the category of all G-sets is cartesian closed. If Y and Z are two G-sets, then ZY is the set of all functions from Y to Z with G action defined by (g.F)(y) = g.(F(g-1.y)) for all g in G, F:Y → Z and y in Y.
- The category of finite G-sets is also cartesian closed.
- The category Cat of all small categories (with functors as morphisms) is cartesian closed; the exponential CD is given by the functor category consisting of all functors from D to C, with natural transformations as morphisms.
- If C is a small category, then the functor category SetC consisting of all covariant functors from C into the category of sets, with natural transformations as morphisms, is cartesian closed. If F and G are two functors from C to Set, then the exponential FG is the functor whose value on the object X of C is given by the set of all natural transformations from (X,−) × G to F.
- The earlier example of G-sets can be seen as a special case of functor categories: every group can be considered as a one-object category, and G-sets are nothing but functors from this category to Set
- The category of all directed graphs is cartesian closed; this is a functor category as explained under functor category.
- In algebraic topology, cartesian closed categories are particularly easy to work with. Neither the category of topological spaces with continuous maps nor the category of smooth manifolds with smooth maps is cartesian closed. Substitute categories have therefore been considered: the category of compactly generated Hausdorff spaces is cartesian closed, as is the category of Frölicher spaces.
- In order theory, complete partial orders (cpos) have a natural topology, the Scott topology, whose continuous maps do form a cartesian closed category (that is, the objects are the cpos, and the morphisms are the Scott continuous maps). Both currying and apply are continuous functions in the Scott topology, and currying, together with apply, provide the adjoint.
- A Heyting algebra is a Cartesian closed (bounded) lattice. An important example arises from topological spaces. If X is a topological space, then the open sets in X form the objects of a category O(X) for which there is a unique morphism from U to V if U is a subset of V and no morphism otherwise. This poset is a cartesian closed category: the "product" of U and V is the intersection of U and V and the exponential UV is the interior of U∪(X\V).
The following categories are not cartesian closed:
- The category of all vector spaces over some fixed field is not cartesian closed; neither is the category of all finite-dimensional vector spaces. This is because the tensor product is not a category-theoretic product; in particular, the act of tensoring destroys the projection morphisms. They are, however, symmetric monoidal closed categories: the set of linear transformations between two vector spaces forms another vector space, so they are closed. The tensor product does have a right adjoint, a mapping object, which is the set of linear maps between vector spaces; however, it is not properly called the exponential object.
- The category of abelian groups is not cartesian closed, for the same reason.
In cartesian closed categories, a "function of two variables" (a morphism f:X×Y → Z) can always be represented as a "function of one variable" (the morphism λf:X → ZY). In computer science applications, this is known as currying; it has led to the realization that simply-typed lambda calculus can be interpreted in any cartesian closed category.
The Curry-Howard-Lambek correspondence provides a deep isomorphism between intuitionistic logic, simply-typed lambda calculus and cartesian closed categories.
The renowned computer scientist John Backus has advocated a variable-free notation, or Function-level programming, which in retrospect bears some similarity to the internal language of cartesian closed categories. CAML is more consciously modelled on cartesian closed categories.
In every cartesian closed category (using exponential notation), (XY)Z and (XZ)Y are isomorphic for all objects X, Y and Z. We write this as the "equation"
- (xy)z = (xz)y.
One may ask what other such equations are valid in all cartesian closed categories. It turns out that all of them follow logically from the following axioms:
- x×(y×z) = (x×y)×z
- x×y = y×x
- x×1 = x (here 1 denotes the terminal object of C)
- 1x = 1
- x1 = x
- (x×y)z = xz×yz
- (xy)z = x(y×z)
Bicartesian closed categories extend cartesian closed categories with binary coproducts and an initial object, with products distributing over coproducts. Their equational theory is extended with the following axioms:
- x + y = y + x
- (x + y) + z = x + (y + z)
- x(y + z) = xy + xz
- x(y + z) = xyxz
- 0 + x = x
- x×0 = 0
- x0 = 1
Note however that the above list is not complete; type isomorphism in the free BCCC is not finitely axiomatizable, and its decidability is still an open problem.
- John C. Baez and Mike Stay, "Physics, Topology, Logic and Computation: A Rosetta Stone", (2009) ArXiv 0903.0340 in New Structures for Physics, ed. Bob Coecke, Lecture Notes in Physics vol. 813, Springer, Berlin, 2011, pp. 95-174.
- S. Mac Lane, "Categories for the Working Mathematician"
- H.P. Barendregt, The Lambda Calculus, (1984) North-Holland ISBN 0-444-87508-5 (See theorem 1.2.16)
- S. Soloviev. "Category of Finite Sets and Cartesian Closed Categories", Journal of Soviet Mathematics, 22, 3 (1983)
- Fiore, Cosmo, and Balat. Remarks on Isomorphisms in Typed Lambda Calculi with Empty and Sum Types, ""