Category Theory

Category theory is the mathematics of structure and of the maps that preserve it.

10 frameworks, from 1945 to today · Category Theory · Mathematics

still a working view ended · each leads to its entry below

Overview

Category theory is the mathematics of structure and of the maps that preserve it1. Where most of mathematics asks what a thing is, category theory asks how it relates to everything around it. A category is a collection of objects together with arrows (also called morphisms) between them, such that arrows compose and every object has an identity arrow123. That is nearly the whole definition, and its austerity is the point: the same small vocabulary turns out to describe groups and their homomorphisms, spaces and continuous maps, logical theories and their translations, data schemas and their migrations, and much else. What is at stake in the field's history is a shift in what counts as an explanation. Category theorists hold that a construction is understood when it is characterized by the role it plays—by the arrows into and out of it—rather than by the set-theoretic material it is built from. The story of the subject is the story of that idea being pushed further and further, from a bookkeeping device in algebraic topology to a proposed foundation for all of mathematics.

Inventing a language for natural transformations (1942–1945)

The field began with a problem of comparison. In the late 1930s and early 1940s, Samuel Eilenberg and Saunders Mac Lane were working in algebraic topology, studying how spaces can be built from simpler pieces and how the resulting algebraic invariants behave. To make precise what it means for one construction to depend naturally on another—for instance, for a homology group to vary with the space—they needed a notion of map between functors1. A functor is a structure-preserving passage from one category to another; a natural transformation is a consistent family of arrows relating two such functors1.

Their paper "General Theory of Natural Equivalences" (1945) introduced categories, functors, and natural transformations together1. The categories were at first a convenience, a way of saying what a natural transformation was a transformation between. The decisive move was to notice that the three notions form a hierarchy—objects and arrows, then functors between categories, then natural transformations between functors—and that this hierarchy is itself worth studying. The vocabulary made possible a style of argument that had been implicit in topology: instead of computing with elements, one shows that two constructions agree because they are related by a natural isomorphism, or because each satisfies the same universal property.

Universal constructions and adjoint functors (1940s–1950s)

The next turning point was the realization that many constructions in algebra and topology are best defined not by a formula but by a property1. A product of two objects, for example, can be described as an object with projections to each factor such that any other object mapping to both factors factors uniquely through it4. This is a universal property: a characterization by the arrows around an object15. Limits and colimits—general products, pullbacks, equalizers and their duals—were developed in this spirit, largely by Mac Lane and Eilenberg in the late 1940s, and given systematic treatment in Mac Lane's textbook Categories for the Working Mathematician (1971), which taught the subject to a generation.

The most consequential idea of this period was the adjoint functor. Two functors are adjoint when arrows out of one correspond naturally to arrows into the other—a kind of optimal translation between two categories6. Daniel Kan introduced adjoint functors in 195865. They turned out to unify an extraordinary range of constructions: free groups are left adjoint to forgetful functors, products are right adjoints to diagonal functors, and many "universal" objects in algebra are adjoints in disguise. The slogan that emerged, associated with Mac Lane, is that adjoints arise everywhere because they capture the general notion of a best approximation. This is the toolkit—limits, colimits, adjoints, the Yoneda lemma (which says an object is determined up to isomorphism by the arrows into it)—that became the working core of the subject.

Abelian categories and homological algebra (1950s–1960s)

Homological algebra supplied category theory's first large-scale application and, in return, one of its central concepts. Algebraic topologists had learned to associate to a space a sequence of abelian groups and connecting maps, and to extract invariants from the failure of certain sequences to be exact. In the 1950s, Alexander Grothendieck and others abstracted the setting: an abelian category is a category in which one can add arrows and form kernels and cokernels, so that the exact-sequence arguments of homological algebra go through7. Grothendieck's 1957 paper "Sur quelques points d'algèbre homologique" (the "Tôhoku paper") showed that the theory could be developed axiomatically, and introduced the technique of deriving functors in this general setting. The payoff was that the same machinery applied to modules, to sheaves, and to other structures at once, and that one could prove results about all of them with a single argument.

Sheaves, Grothendieck toposes, and geometry (1957–1964)

Grothendieck's school then pushed the categorical viewpoint into algebraic geometry. To handle the local-to-global passage in geometry, one uses sheaves: assignments of data to open regions of a space that glue consistently8. Grothendieck recast a space as the category of sheaves on it—the topos—and, crucially, showed that a topos can be described by a site, a category equipped with a notion of covering, without any underlying space. This made possible a relative, functorial geometry in which spaces are studied through the categories of sheaves they support. The concept of a Grothendieck topos, named in the early 1960s, is one of the field's great exports: it turned a topological tool into a general categorical object.

Categorical logic and the elementary topos (1963–1972)

A parallel development came from logic. In 1963, F. William Lawvere, working from the algebraic side, proposed that algebraic theories could be understood as categories with finite products, and that models of a theory are functors from such a category into the category of sets. This functorial semantics reframed logic as a branch of category theory: a theory is a category, a model is a functor, and a translation between theories is a functor between their categories. Around the same time, the Curry–Howard correspondence—that proofs and programs, and propositions and types, are two views of one structure—was given a categorical form by Lawvere and others, with the observation that the simply typed lambda calculus corresponds to cartesian closed categories.

In 1969–1970, Lawvere and Myles Tierney isolated the elementary topos: a category with enough structure to interpret higher-order intuitionistic logic, defined by axioms that do not mention sheaves or sites. Every Grothendieck topos is an elementary topos, but the elementary notion is first-order and can serve as a foundation. Lawvere's 1970 lectures and the resulting literature argued that set theory could be replaced by topos theory as a foundation for mathematics, with the category of sets as one topos among many. This remains a live position rather than a settled consensus, but it permanently changed how logicians and category theorists talk about theories and their models.

Monoidal and enriched categories (1960s–1970s)

Two generalizations widened the reach of the subject. A monoidal category is a category equipped with a notion of combining objects—a tensor product—together with coherence data making the combination associative and unital up to isomorphism9. The idea, developed by Mac Lane and others in the early 1960s (Mac Lane's coherence theorem dates from 1963), lets one do algebra inside a category rather than in sets. Enriched category theory, developed by Lawvere and others in the 1960s and 1970s, goes further: instead of asking whether there is an arrow between two objects, one assigns a value to the pair drawn from some monoidal category—a set, a truth value, a metric space, a chain complex. Lawvere's 1973 observation that metric spaces are categories enriched over the nonnegative reals is the classic illustration of how far the notion stretches.

Model categories and homotopical algebra (1967–1970s)

Topology had always been category theory's home ground, and in the late 1960s Daniel Quillen gave the subject a general theory of homotopy. A model category is a category with three distinguished classes of arrows—weak equivalences, fibrations, cofibrations—satisfying axioms that let one do homotopy theory abstractly10. Quillen's Homotopical Algebra (1967) showed that the homotopy category of a model category can be studied without choosing a particular space or chain complex, and that many different model structures present the same homotopy theory. This made possible a unified treatment of rational homotopy theory, derived categories, and later algebraic K-theory, and it set the stage for the higher-categorical developments that followed.

Higher categories (1960s–present)

The hierarchy of objects, arrows, and arrows between arrows does not have to stop at natural transformations. A 2-category has objects, arrows, and 2-arrows between arrows; a bicategory relaxes the composition laws to hold only up to coherent invertible 2-arrows1. These notions were introduced in the 1960s—bicategories by Jean Bénabou in 1967—and generalized upward to n-categories and weak ∞-categories. The motivating problem was that many structures in topology and algebra are naturally "weak": associativity holds only up to homotopy, and one needs a coherent system of higher homotopies to say so. The homotopy hypothesis, formulated by Grothendieck and others, proposes that weak ∞-groupoids are equivalent to homotopy types of spaces. In the 1990s and 2000s, André Joyal's quasi-categories (simplicial sets with a filling condition) and the work of Jacob Lurie and others gave a workable theory of ∞-categories, which now underpins derived algebraic geometry and much of modern homotopy theory. Higher category theory has grown into a field of its own, with several competing definitions of weak n-category and ongoing work on their equivalence.

Graphical calculi and computation (1980s–present)

As category theory moved into computer science, a practical need arose: reasoning about morphisms in monoidal and higher categories by drawing them. String diagrams—a graphical notation in which objects are wires and morphisms are nodes, with composition as wiring and tensor as juxtaposition—were developed in the 1980s and 1990s, notably in the work of Joyal and Ross Street on braided monoidal categories and in the categorical quantum mechanics program of Samson Abramsky and Bob Coecke in the 2000s. The graphical calculus is not merely a convenience: coherence theorems say that a diagrammatic equation holds in a monoidal category exactly when it holds for the diagrams, so the pictures are a rigorous proof method. This made monoidal categories usable in quantum information, where the tensor product models composite systems, and in the semantics of programming languages.

Applied category theory (2000s–present)

The most recent turn carries the vocabulary outside pure mathematics. Categorical semantics of computation, building on the Curry–Howard–Lambek correspondence, gives types and programs a categorical meaning and underlies the design of proof assistants and typed programming languages. Functorial data migration, developed by David Spivak and others in the 2010s, treats a database schema as a category and a migration as a functor, so that schema changes become compositional. Categorical systems theory models dynamical systems and their interconnection with monoidal and enriched categories. The Applied Category Theory conference series, begun in 2018, and the textbook Seven Sketches in Compositionality (2018) by Brendan Fong and Spivak mark the emergence of a self-conscious applied community. The field's open questions are correspondingly broad: whether a single workable definition of weak n-category will win out, how far topos-theoretic foundations can replace set theory in practice, and which of the many applied models will prove durable rather than merely suggestive.

References

  1. Category theory (introduction, Historical notes, Functors, Natural transformations, Universal constructions, limits, and colimits, Higher-dimensional categories), Wikipedia
  2. Category (mathematics), Wikipedia
  3. Morphism, Wikipedia
  4. Diagonal functor, Wikipedia
  5. Universal property (introduction, History), Wikipedia
  6. Adjoint functors (introduction, History), Wikipedia
  7. Abelian category (introduction, History), Wikipedia
  8. Sheaf (mathematics), Wikipedia
  9. Monoidal category, Wikipedia
  10. Model category, Wikipedia

The frameworks of Category Theory

Each framework is a way of working the field has used: when it held, and the question it answers.

  1. 1945-Present

    Universal Algebra and Adjoint Functors

    What makes a construction universal?

    About Universal Algebra and Adjoint Functors
    Universal algebra asks what all algebraic structures—groups, rings, lattices, modules—have in common once you strip away their particular operations and axioms. Category theory asks what all mathematical structures have in common once you strip away their elements and keep only the arrows. The framework gathered under "universal algebra and adjoint functors" is the meeting point of these two questions: the discovery that the constructions universal algebra cares about (free algebras, quotients, products, tensor products) are all instances of a single categorical pattern, adjunction, and that the theorems universal algebra proves about varieties are theorems about adjoint functors.

    The problem: the same construction keeps reappearing

    Anyone who learns algebra meets the same move over and over. To build a free group on a set XX, you take all formal words in the elements of XX and their inverses, then impose the group axioms by taking a quotient. To build a free module, you take formal linear combinations. To build a polynomial ring, formal expressions. To build a group from a monoid by adjoining inverses, a similar formal construction. Each time, the recipe is: start with the bare data, generate freely, then impose relations.
    Meanwhile, in topology and algebra, another move recurs: given a map that is not a homomorphism, find the "best" homomorphism that approximates it. The abelianization of a group, the completion of a metric space, the sheafification of a presheaf, the Stone–Čech compactification—each is the universal solution to a problem of extending or correcting a map.
    Universal algebra and adjoint functors exist to answer: why do these constructions keep working the same way, and what can be proved about all of them at once? The payoff is not elegance for its own sake. Once you know a construction is an adjoint, you get its existence, its uniqueness up to isomorphism, and its interaction with limits and colimits for free, and you can transfer theorems between algebra, topology, and logic.

    The organizing idea, and what it is set against

    The organizing idea is universality: an object is characterized not by what it is made of but by the unique maps out of it (or into it) that it admits. A free group on XX is the group F(X)F(X) with a map η:X→F(X)\eta : X \to F(X) such that every function from XX into a group GG extends uniquely to a homomorphism F(X)→GF(X) \to G. Nothing about words or quotients appears in that description; the object is pinned down entirely by its arrows.
    Set this against three neighbouring approaches:
    • Element-based universal algebra (Birkhoff, Tarski, and the Polish and American schools from the 1930s) studies varieties—classes of algebras defined by equations—by manipulating terms, subalgebras, homomorphic images, and direct products. It assumes you work inside a signature of operations and prove things by induction on terms. Adjoint functors replace term manipulation with arrow-theoretic characterization.
    • Set-theoretic construction assumes each construction is built and then verified. The categorical approach assumes the universal property is the definition and the construction is one way (not the only way) to realize it.
    • Homological algebra, which grew alongside and partly out of this work from the 1940s onward, also uses arrows and universal properties, but its centre of gravity is chain complexes and exact sequences. The universal-algebra-and-adjunction tradition stays with general algebraic structures and the logic of equations.

    Who shaped it, and when

    The categorical half begins with Samuel Eilenberg and Saunders Mac Lane's "General Theory of Natural Equivalences" (1945), which introduced categories, functors, and natural transformations. Daniel Kan introduced adjoint functors in "Adjoint Functors" (1958), the paper that gave the notion its name and its first theorems. Saunders Mac Lane's 1948 paper on duality for modules and his later work with Garrett Birkhoff connected the categorical machinery to universal algebra; Birkhoff's 1935 characterization of varieties as classes closed under homomorphic images, subalgebras, and products is the algebraic backbone. The synthesis into a textbook tradition came with Mac Lane's Categories for the Working Mathematician (1971) and, on the algebra side, with the school around George Grätzer and later universal-algebra texts. The framework is thus a genuine research tradition with identifiable authors, not a retrospective label—though "universal algebra and adjoint functors" as a single phrase is a teaching synthesis of two literatures that developed in parallel and interlocked.

    Core concepts, in learning order

    1. Signature, algebra, homomorphism. A signature Σ\Sigma is a set of operation symbols, each with an arity n≥0n \ge 0. A *Σ\Sigma-algebra* is a set AA together with, for each nn-ary symbol ff, a function fA:An→Af_A : A^n \to A. A homomorphism h:A→Bh : A \to B is a function commuting with every operation: h(fA(a1,…,an))=fB(h(a1),…,h(an))h(f_A(a_1,\dots,a_n)) = f_B(h(a_1),\dots,h(a_n)). Groups, rings, lattices, and monoids are all algebras for suitable signatures. This is the object language; everything else is built on it.
    2. Terms, equations, and the term algebra. From a set XX of variables and a signature, terms are built by applying operation symbols to variables and to other terms. The term algebra TΣ(X)T_\Sigma(X) has the terms as its elements, with operations acting by formal application. It is the syntactic object every free algebra is a quotient of.
    3. Varieties and Birkhoff's theorem. A variety (equational class) is the class of all algebras satisfying a set of equations between terms. Birkhoff's theorem (1935) says a class of algebras is a variety exactly when it is closed under homomorphic images, subalgebras, and arbitrary direct products. This is the first great structural theorem: it converts a syntactic condition (equations) into a closure condition (operations on classes).
    4. Categories, functors, natural transformations. A category has objects and arrows with associative composition and identities. A functor F:C→DF : \mathcal{C} \to \mathcal{D} sends objects to objects and arrows to arrows, preserving identities and composition. A natural transformation α:F⇒G\alpha : F \Rightarrow G assigns to each object CC an arrow αC:F(C)→G(C)\alpha_C : F(C) \to G(C) such that for every f:C→C′f : C \to C', αC′∘F(f)=G(f)∘αC\alpha_{C'} \circ F(f) = G(f) \circ \alpha_C. The category Alg(Σ)\mathbf{Alg}(\Sigma) of Σ\Sigma-algebras and homomorphisms is the setting; the forgetful functor U:Alg(Σ)→SetU : \mathbf{Alg}(\Sigma) \to \mathbf{Set} sends an algebra to its underlying set.
    5. Adjoint functors. Functors F:C→DF : \mathcal{C} \to \mathcal{D} and G:D→CG : \mathcal{D} \to \mathcal{C} are adjoint, written F⊣GF \dashv G, when there is a natural bijection
    D(F(C),D)≅C(C,G(D))\mathcal{D}(F(C), D) \cong \mathcal{C}(C, G(D))
    for all CC in C\mathcal{C} and DD in D\mathcal{D}. FF is the left adjoint, $G the right adjoint. The bijection says: a map out of F(C)$ corresponds exactly to a map into $G(D). Equivalently, there are a unit \eta : \mathrm{id}_{\mathcal{C}} \Rightarrow G F$ and a counit $\varepsilon : F G ⇒\Rightarrow \mathrm{id}_{\mathcal{D}} satisfying the triangle identities.
    The worked example the tradition uses: the free group functor F : \mathbf{Set} \to \mathbf{Grp}isleftadjointtotheforgetfulfunctoris left adjoint to the forgetful functorU : \mathbf{Grp} \to \mathbf{Set}. The adjunction bijection says a group homomorphism F(X)F(X) \to Gisthesamedataasafunctionis the same data as a functionX \to U(G)U(G). Concretely, to define a homomorphism out of a free group you need only say where the generators go—no need \to check relations, because there are none. The unit \eta_X : X \to U F(X)F(X) is the insertion of generators. The same pattern gives free monoids, free modules, free rings, and polynomial rings, each as a left adjoint \to a forgetful functor.
    6. Limits and colimits. A limit of a diagram is a universal cone over it; a colimit is a universal cocone. Products and equalizers are limits; coproducts and coequalizers are colimits. The key theorem: right adjoints preserve limits, left adjoints preserve colimits. Since the forgetful functor U : \mathbf{Grp} \to \mathbf{Set} is a right adjoint, it preserves products—the underlying set of a product of groups is the product of the underlying sets. Since F is a left adjoint, it preserves coproducts: the free group on a disjoint union is the free product of the free groups.
    7. The adjoint functor theorem. Not every functor has an adjoint. Freyd's adjoint functor theorem (1964) gives conditions: a functor G : \mathcal{D} \to \mathcal{C}withwith\mathcal{D} complete has a left adjoint if and only if it preserves all limits and satisfies a "solution set" condition. This is the categorical counterpart of the universal-algebra fact that free algebras exist for any signature: the forgetful functor from algebras \to sets preserves limits and has a solution set (the term algebra on the generating set), so it has a left adjoint.
    8. Monads and algebras over a monad. Every adjunction F \dashv Ginducesa∗monad∗induces a *monad*T = G Fonon\mathcal{C},withunit, with unit\etaandmultiplicationand multiplication\mu = G \varepsilon F. Algebras for the monad are objects Cwithamapwith a mapT(C)T(C) \to C$ satisfying unit and associativity laws. For the free-group adjunction, the monad sends a set to the underlying set of the free group on it—the "word" monad. Beck's monadicity theorem (1967) characterizes when a category is equivalent to the category of algebras over a monad, giving a precise categorical version of "this category is exactly the algebras for this signature." This is where universal algebra and adjunction theory fuse most tightly.

    How it developed

    The 1945 Eilenberg–Mac Lane paper supplied the language; Kan's 1958 paper supplied adjoints. Through the 1960s the two streams converged. Lawvere's 1963 thesis recast universal algebra as the study of categories with finite products and their functors, making algebraic theories themselves objects of a category. Beck's monadicity theorem and the "tripleability" literature of the mid-1960s gave the exact correspondence between varieties and monadic categories. Freyd's adjoint functor theorem and the general theory of limits made existence questions routine.
    The main debates were about scope and about foundations. One debate concerned whether the categorical reformulation added theorems or only reorganized proofs; the answer that emerged is that it adds theorems of a new kind—transfer principles, existence results, and uniform constructions—even when individual algebraic facts were known. Another concerned size: the category of all groups is not small, and the early theory needed careful handling of large categories, resolved by Grothendieck's universe conventions and by the distinction between locally small and large categories. A third, more philosophical debate asked whether "universal property" is a definition or a characterization; the working answer is that it is a definition, and any particular construction is a model of it.
    What changed in the field: universal algebra stopped being a collection of case-by-case theorems and became a branch of category theory, while category theory gained its most important source of examples and its most important application outside topology. The language of adjoints and monads spread into theoretical computer science (algebraic specifications, monads in programming semantics) and into logic (Lawvere's functorial semantics).

    Standing today

    Adjoint functors are taught in every first course in category theory, and the free–forgetful adjunction is the standard first example. Universal algebra survives as a subject in its own right—Grätzer's Universal Algebra and Burris and Sankappanavar's text remain in use—and its categorical reformulation is standard in graduate algebra. The monad–algebra correspondence is central to categorical logic and to the semantics of computation, where monads model effects. Open questions remain about the right general setting for varieties of algebras over enriched or higher-dimensional bases, about the precise relationship between equational logic and monadic categories in settings without finite limits, and about how far the adjunction framework extends to structures—such as those with binding or variable-binding operations—that are not captured by plain signatures. The framework's core claim, that a construction is understood when its universal property is stated, has proved durable enough that it is now simply how algebra is done.
    The 11 concepts its tower teaches
    • Signature, Algebra, and Homomorphism
    • Terms and the Term Algebra
    • Varieties and Birkhoff's Theorem
    • Categories, Functors, and Natural Transformations
    • Lawvere Theories
    • Adjoint Functors
    • Limits and Colimits
    • Adjoint Functor Theorem
    • Monads
    • Algebras over a Monad
    • Beck's Monadicity Theorem

    Fly to its tower in Nousfarer

  2. 1957-Present
  3. 1957-Present

    Topos

    When can a category of sheaves replace a space?

    Fly to its tower in Nousfarer

  4. 1963-Present

    Categorical logic

    How do theories and models relate categorically?

    Fly to its tower in Nousfarer

  5. 1963-Present
  6. 1967-Present

    Higher Category Theory

    How do arrows between arrows generalize?

    Fly to its tower in Nousfarer

  7. 1967-Present

    Model category

    How can homotopy theory be done abstractly?

    Fly to its tower in Nousfarer

  8. 1980-Present

    Categorical Semantics of Computation

    What categorical structure models computation?

    Fly to its tower in Nousfarer

  9. 1985-Present

    String diagram

    How can categorical composition be drawn?

    Fly to its tower in Nousfarer

  10. 2010-Present

    Applied Category Theory

    How can categorical tools solve real-world problems?

    Fly to its tower in Nousfarer

How the ideas connect

Each arrow runs from an earlier idea to the later one it shaped. The list below says what passed between them.

Universal Algebra and Adjoint FunctorsAbelian Categories and Homological Al…ToposCategorical logicMonoidal and Enriched Category TheoryHigher Category TheoryModel categoryCategorical Semantics of ComputationString diagramApplied Category Theory

Beyond the field

Where Category Theory's ideas came from in other fields, and what they led to there: the strongest 12 of 74 links.

A city in Nousfarer: a red-carpeted stair climbing terraces to a crowned tower on an acropolis, more towers round it, forest and snowy mountains beyond.
A city in Nousfarer, the game played in the browser.

Category Theory is a planet in Nousfarer

Its city has a tower for each of these 10 frameworks. Land, walk in, and climb one concept at a time.

Land on Category Theory

Related fields

Fields its ideas are linked to

More in Category Theory

See where Category Theory sits on the Map of Human Knowledge