Skip to content

2-Category

A strict higher category with objects, arrows between objects, and arrows between parallel arrows, composed by compatible vertical and horizontal laws.

Version
v1 · 2026-10-03 · History
Domain-specific #
12963
Domain group
Formal Sciences
Origin domain
Mathematics
Subdomains
Category Theory, Higher Category Theory → Mathematics
Aliases
Strict 2-category

Core Idea

A strict 2-category organizes relationships at two compositional levels. It has objects (0-cells), arrows between objects (1-cells), and arrows between parallel 1-cells (2-cells). For each pair of objects \(A,B\), the 1-cells \(A\to B\) and the 2-cells between them form a hom-category. Thus 2-cells compose vertically within a hom-category. Across \(A\to B\to C\), a composition functor takes a pair of hom-categories to the hom-category from \(A\) to \(C\), composing both 1-cells and compatible 2-cells horizontally. The functoriality of this operation gives the interchange law: performing two vertical compositions and then composing horizontally equals horizontal composition first and then vertical composition, whenever the cells are typed to make both sides meaningful.[1][2]

The adjective strict matters. Composition of 1-cells has identities and is associative as an equality, and horizontal composition of 2-cells has the corresponding strict laws. A bicategory or weak 2-category replaces the strict 1-cell associativity and unit equalities with specified coherent invertible 2-cells. It has the same broad dimensional vocabulary, but not the identical axiom system.[1][3] This entry uses “2-category” for the strict identity identified by the confirmed Strict 2-category candidate. Other requested titles that redirected to Wikipedia's 2-category page—Bicategory, (2,1)-category, 2-functor, Lax functor, and Duskin nerve—are not silently admitted as aliases.

Structural Signature

Sig role-phrases: typed 0/1-cells → hom-categories of 2-cells → vertical composition → horizontal composition functors → strict units, associativity and interchange.

  • 0-cells and typed 1-cells: Objects and source/target-typed arrows make the underlying ordinary category. Without them, a purported 2-cell has no parallel 1-cell boundaries to relate.
  • Hom-categories and 2-cells: Each hom \(\mathcal C(A,B)\) is a category whose objects are 1-cells \(A\to B\) and whose arrows are 2-cells between those 1-cells. Merely saying that two arrows are “related” does not supply a hom-category.[1]
  • Vertical composition and identity 2-cells: The category laws in each hom compose \(\alpha:f\Rightarrow g\) with \(\beta:g\Rightarrow h\) and give each 1-cell an identity 2-cell. Removing this operation removes the categorical structure of the homs.
  • Horizontal composition functors: For \(A,B,C\), a functor \(\mathcal C(B,C)\times\mathcal C(A,B)\to\mathcal C(A,C)\) composes adjacent 1-cells and 2-cells. Because it is a functor, it preserves vertical composition and identity 2-cells. That compatibility is the interchange law, not a decorative extra rule.[1][2]
  • Strict associativity and units: Reassociation of composable 1-cells, and corresponding horizontal 2-cell composites, yields equality on the nose; identity 1-cells act neutrally. If the relevant equalities are instead mediated by coherent associator and unitor isomorphisms, the structure is a bicategory, not a strict 2-category.[3]

These roles must be checked together. A category plus a collection of transformations is not necessarily a 2-category unless the transformations form hom-categories and horizontal composition acts functorially across them.

What It Is Not

It is not merely an ordinary category with an interesting collection of arrows. An ordinary category supplies objects, arrows, identities and associative composition; a strict 2-category additionally specifies categories of arrows between parallel arrows and how those higher arrows compose. An ordinary category can be embedded as a locally discrete 2-category by adding only identity 2-cells, but that is a construction, not evidence that the original ordinary-category description stated the full two-level structure.

It is not a bicategory by terminological fiat. Some mathematical writing uses “2-category” broadly, but a Morita construction with rings, bimodules and bimodule homomorphisms has tensor-product composition associative up to canonical isomorphism. That familiar weak case is not a strict instance without a separately specified strict presentation.[3] Nor is it a double category: a double category has distinct horizontal and vertical kinds of 1-arrow with squares between boundaries; the two compositions here are compositions of 2-cells, not two interchangeable 1-arrow directions.

It is not a 2-functor or a lax functor, both of which are maps between higher categories under different preservation conditions. Nor is a Duskin nerve the structure itself: it is a construction encoding higher-categorical data. A (2,1)-category is a narrower case in which the 2-cells are invertible; invertibility is not required for every strict 2-category.[1]

Scope of Application

Strict 2-categories occur literally in higher category theory whenever objects, 1-cells, and transformations between parallel 1-cells can be equipped with the specified strict operations. Cat, under a declared size convention, uses categories as objects, functors as 1-cells, and natural transformations as 2-cells. Functor categories provide the hom-categories, and ordinary functor composition is the horizontal composition functor.[2][1]

The 2-category of groupoids is another literal habitat. Groupoids, functors, and natural transformations retain the same strict functor-composition laws; here every component of a natural transformation is invertible because it is an arrow in a groupoid, so this example additionally has the (2,1)-property.[1] More specialized categories of fibered categories or stacks over a fixed base can likewise be arranged in strict 2-categorical settings, but their size and admissible morphisms must be fixed before an instance is claimed.[1]

The name does not license applying these axioms to any three-layered organizational chart. One must actually specify higher morphisms, composable hom-categories and strict functorial laws.

Clarity

The structure separates three often-confused questions. First, is a map \(f:A\to B\) a 1-cell, or is a comparison \(\alpha:f\Rightarrow g\) a 2-cell between parallel maps? Second, is a displayed composition taking place within a single hom-category (vertical), or across composable object boundaries (horizontal)? Third, does an associativity claim mean literal equality or coherent isomorphism? The strict 2-category answers all three through typed hom-categories and composition functors. In Cat, for example, a natural transformation is not another functor between the original categories; it is a morphism between functors in \(\mathrm{Fun}(A,B)\).[2]

Manages Complexity

Many equations among natural transformations, whiskerings, functor compositions and identities can be organized as two categorical operations with one compatibility condition. Rather than proving every pasted square by a bespoke component calculation, one checks that each hom is a category, horizontal composition is a functor, and identities/associativity are strict. Interchange then follows from functoriality. This compression does not erase typing: a 2-cell can only relate 1-cells with the same source and target, and only endpoint-compatible homs can compose horizontally. It also does not solve size questions for a large collection such as Cat.[1][2]

Abstract Reasoning

To test a proposed model, first identify the 0-cells and the 1-cells with their endpoints. For each \(A,B\), construct \(\mathcal C(A,B)\) and check that the proposed 2-cells have identities and associative vertical composition. Next check that composing across \(A\to B\to C\) is a functor on the product of hom-categories; this tests horizontal composition and interchange together. Finally test strict associativity and unit equalities, not just natural isomorphisms. Failure of the last test can point toward a bicategory; failure of the hom or interchange tests means even that diagnosis requires further work. This order is useful because it localizes the broken axiom instead of treating every “higher arrow” diagram as a single undifferentiated proof obligation.[1][3]

Knowledge Transfer

Within category theory the same test travels from Cat to sub-2-categories such as Groupoids: the objects and admissible 1-cells change, but hom-categories, vertical composition, horizontal functoriality and strict laws remain. One-object strict 2-categories give strict monoidal categories after relabeling 1-cells as monoidal objects and 2-cells as monoidal arrows; this is a literal mathematical translation, not a general claim that every monoidal category is already strict.[4]

Outside mathematical settings that actually define these operations, “relations between relations” is at most an analogy. The portable underlying Category prime retains objects, arrows, identities and composition. The extra 2-cell typing, hom-categories and interchange are precisely what keep this named entry in higher category theory.

Examples

Canonical: Cat

Fix a size convention and take categories as 0-cells, functors as 1-cells, and natural transformations as 2-cells. If \(F,G:A\to B\), then \(\alpha:F\Rightarrow G\) has a component \(\alpha_a:F(a)\to G(a)\) for each \(a\in A\), subject to naturality. Natural transformations compose componentwise vertically. Functor composition and whiskering compose them horizontally. The formal-properties derivation shows the horizontal operation is functorial, while composition of functors and identity functors are strictly associative and unital.[2][1]

Mapped back: 0-cells and typed 1-cells → categories and functors; Hom-categories and 2-cells → \(\mathrm{Fun}(A,B)\) and its natural transformations; Vertical composition and identity 2-cells → componentwise transformation composition and identity transformations; Horizontal composition functors → \(\mathrm{Fun}(B,C)\times\mathrm{Fun}(A,B)\to\mathrm{Fun}(A,C)\), with interchange; Strict associativity and units → literal associativity of functor composition and identity functors.

Applied mathematical setting: Groupoids

Restrict Cat to groupoids, with functors between groupoids and natural transformations between parallel functors. The hom-category \(\mathrm{Fun}(G,H)\) still exists; its natural transformations compose componentwise. Every component is an arrow of \(H\), hence invertible. Componentwise inverses are natural, so each 2-cell is invertible. This gives a strict 2-category that also satisfies the narrower (2,1)-condition; the latter does not define every strict 2-category.[1]

Mapped back: 0-cells and typed 1-cells → groupoids and functors; Hom-categories and 2-cells → groupoid-functor categories with natural transformations; Vertical composition and identity 2-cells → componentwise compositions/identities, here invertible; Horizontal composition functors → inherited functor/whiskering composition and interchange; Strict associativity and units → the same exact functor-composition equalities as in Cat.

Structural Tensions

T1: Strict equality vs coherent isomorphism. Strict equations make reassociation and unit insertion unambiguous and support direct equational proofs. They also reject a naturally presented bicategory when composition is only associative through chosen invertible comparison 2-cells. Moving to a weak presentation gains faithful coverage but adds coherence data and proof obligations; calling that weak presentation strict obscures the difference. Diagnostic: In the proposed model, are the two bracketings literally the same 1-cell or merely isomorphic by a specified associator?[3]

T2: One-dimensional compression vs transformation-level distinctions. Forgetting 2-cells leaves an ordinary category and simplifies notation. It loses the ability to say how two parallel 1-cells compare, such as via a natural transformation. Keeping hom-categories preserves that comparison but requires vertical/horizontal typing. Diagnostic: Does the theorem need a transformation between maps, or only a composite of maps?

T3: Natural large examples vs controlled size. Cat and Groupoids offer direct mathematical examples but may have a proper class of objects under ordinary set conventions. Restricting to a universe or specified collection tames foundational size, yet may exclude categories needed for a particular construction. Diagnostic: Which categories are allowed as objects, and in which universe are the hom-categories small?[1]

Structural–Framed Character

Evaluative weight: Low. “Strict” states whether named composition equations hold, not whether the object is better than a weak presentation; the appropriate choice depends on the theorem.

Human-practice dependence: Low. Mathematicians choose conventions and size universes, but after those choices the hom-category and functoriality axioms are formal conditions, not social practice.

Institutional origin: Low. The structure developed within mathematical research, yet no institutional rule or policy constitutes an instance; proof of the axioms does.

Vocabulary travel: Narrow for the full identity. “Objects,” “arrows” and “layers” travel easily; hom-category, 2-cell, horizontal functor, interchange and strict equality do not automatically travel with them.

Import vs recognition: Genuine reuse requires importing or discovering actual typed compositions and proving their laws. A diagram that merely looks tiered does not qualify through analogy.

Its character: Predominantly structural within higher mathematics, but domain-specific in the Encyclopedia: its rigor is formal and minimally evaluative, while the full named mechanism does not have demonstrated literal reach outside settings with category-theoretic 2-cells. It is not a prime merely because its arrows are abstract.

Structural Core vs. Domain Accent

Portable skeleton: The live parent Category supplies objects, typed arrows, identities and associative composition. That skeleton genuinely travels into other fields because it does not require maps between arrows. Within higher mathematics, the idea of adding a comparison level to maps is also recognizable, but treating “relations between relations” as a new cross-domain prime would be a future-prime question, not an established edge inferred from a slogan.

Domain-bound mechanism: This identity needs hom-categories, 2-cells with parallel 1-cell boundaries, vertical composition, a horizontal composition functor, interchange, and strict equalities. Cat and Groupoids fill these roles differently while preserving them exactly. A bicategory changes the law, rather than merely the vocabulary; the Morita bicategory cannot become a strict example by relabeling its associator.[1][3]

Why not a prime: The underlying Category prime is portable. The specific two-dimensional categorical grammar here is a specialist formal model; no demonstrated cross-domain use keeps its full validity test intact without importing category theory itself. Analogies to organizations, software layers, or relationships among relationships can motivate inquiry but do not satisfy the inclusion test.

This entry is a kind of Category.

DAG parent — Category: Forget the 2-cells; the 0-cells and 1-cells with strict composition and identities form an ordinary category. The 2-category is a specialized enrichment of that framework, not an alias for it. The structured workspace edge records this single upward relation.

Related, not parents: Category of Small Categories is Cat viewed at the 1-level and is a canonical instance after adding its natural transformations. 3-Category adds a 3-morphism level; it is not needed as a parent of a strict 2-category. Double category organizes two kinds of 1-arrow and squares, a different two-dimensional shape. 2-group has one-object/invertibility restrictions, not the general identity. These comparisons are structural; shared terminology alone makes no DAG edge.

Relationships to Other Abstractions

Local relationship map for 2-CategoryParents appear above the current abstraction, mutual partners to the right, and children below. Node labels state whether each abstraction is prime or domain-specific; colors identify relation types.2-CategoryDOMAINPrime abstraction: Category — is a kind ofCategoryPRIME

Current abstraction 2-Category Domain-specific

Parents (1) — more general patterns this builds on

  • 2-Category is a kind of Category Prime

    A strict 2-category extends its underlying ordinary category with hom-categories and compatible 2-cell composition.

Hierarchy paths (3) — routes to 3 parentless roots

Neighborhood in Abstraction Space

2-Category sits in a sparse region of the domain-specific corpus (78th percentile for distinctiveness): few abstractions share its structure, so a faithful description tends to retrieve it precisely.

Family — Categories, Sheaves & Homotopy (21 abstractions)

Nearest neighbors

Computed from structural-signature embeddings · 2026-10-08

Not to Be Confused With

  • Bicategory (weak 2-category). Tell: coherent associator/unitor 2-isomorphisms replace strict 1-cell equations. The seed's Morita bicategory belongs here unless a strict presentation is explicitly chosen.[3]
  • (2,1)-category. Tell: every 2-cell is invertible. Groupoids meet this extra condition; Cat in general does not.[1]
  • 2-functor or lax functor. Tell: each maps one higher category to another and specifies how composition is preserved; neither is the source higher category itself.
  • Duskin nerve. Tell: a nerve encodes a higher-categorical structure into a simplicial presentation; it is not the source structure.
  • Double category. Tell: horizontal and vertical 1-arrow classes and square boundaries differ from the single 1-arrow level plus vertically/horizontally composable 2-cells here.
  • 3-category. Tell: requires a level of 3-cells between parallel 2-cells, absent from a strict 2-category.

References

[1] The Stacks Project, Categories, §4.29 “2-categories,” Definition 4.29.1 and Remark 4.29.3. Defines the strict structure, composition functor and examples including Cat and Groupoids. registry ↩a ↩b ↩c ↩d ↩e ↩f ↩g ↩h ↩i ↩j ↩k ↩l ↩m ↩n ↩o

[2] The Stacks Project, Categories, §4.28 “Formal properties”. Derives vertical/horizontal natural-transformation composition and its functoriality in Cat. registry ↩a ↩b ↩c ↩d ↩e ↩f

[3] John C. Baez, This Week's Finds in Mathematical Physics, Week 35, strict versus weak 2-category and associator/coherence discussion. registry ↩a ↩b ↩c ↩d ↩e ↩f ↩g

[4] John C. Baez, This Week's Finds in Mathematical Physics, Week 83, strict monoidal categories as one-object 2-categories. registry ↩