1 Foundations: Coends and Dinatural Transformations

Coends are a categorical construction used to assemble data indexed by an object variable in both covariant and contravariant ways. They arise naturally in category theory, especially in the study of profunctors, enriched categories, and universal constructions. A coend packages a universal way of identifying compatible pieces of a diagram, and it is often treated as a categorical analogue of integration.

1.1 Coends as colimits

A coend of a functor \(H : \mathcal{C}^{op} \times \mathcal{C} \to \mathcal{D}\) is a colimit-like object that identifies values of \(H\) along the action of morphisms in \(\mathcal{C}\). Informally, it combines all objects \(H(c,c)\) while imposing the relations dictated by morphisms \(f : c \to c'\). When it exists, the coend is usually written \(\int^{c} H(c,c)\).

In many concrete settings, a coend can be constructed as a coequalizer of two maps between coproducts. This makes it accessible in ordinary categories with suitable colimits. The coend then plays the role of a quotient of a formal sum by compatibility relations.

1.2 Dinatural transformations and universality

The universal property of a coend is expressed using dinatural transformations. A dinatural family from \(H\) to an object \(X\) consists of maps \(H(c,c) \to X\) that are compatible with the morphisms in \(\mathcal{C}\) in a balanced way. The coend is universal among such dinatural cocones.

This universality is what distinguishes coends from ordinary colimits. The relations are not imposed separately in each variable, but jointly through the paired variance of the functor. As a result, coends often encode self-canceling or contraction-like behavior.

1.3 Ends vs. coends (duality perspective)

Ends are the dual notion to coends. An end of a functor \(H : \mathcal{C}^{op} \times \mathcal{C} \to \mathcal{D}\) is denoted \(\int_{c} H(c,c)\), while a coend is denoted with an upper integral sign \(\int^{c}\). Ends are limits built from dinatural families of maps into the diagram, whereas coends are colimits built from dinatural families out of the diagram.

This duality is often useful for translating statements between limit and colimit forms. Many theorems about coends have formal duals for ends, and vice versa. The Fubini theorem appears in both settings, with corresponding hypotheses and conclusions.

1.4 Coend notation and common rewrites

Standard notation treats the bound variable in a coend as dummy data, much like an integration variable. Thus \(\int^{c} H(c,c)\) may be rewritten using a renamed variable without changing meaning. This flexibility is essential when comparing nested coends or changing the order of variables.

Common rewrites include reassociating products of indexing categories, inserting or removing naturally isomorphic forms, and expressing coends as coequalizers when the ambient category permits. Such identities are routine in categorical algebra and form the technical basis for the Fubini theorem.

2 The Fubini Theme for Coends

The Fubini theme concerns exchanging the order of multiple coends or collapsing an iterated coend into a single coend over a product category. It reflects the same intuition as classical Fubini’s theorem: a two-step integration process can often be performed in either order, provided suitable conditions hold.

2.1 Motivation: analogy with classical Fubini

In ordinary analysis, Fubini’s theorem allows repeated integrals to be evaluated in either order under appropriate hypotheses. For coends, the analogous question is whether one can compute \[ \int^{c} \int^{d} H(c,d,c,d) \] by swapping the variables or reorganizing the expression. The theorem says that, under the right categorical assumptions, the order of coending does not matter up to canonical isomorphism.

The analogy is structural rather than numerical. Coends do not measure size or area; instead, they capture universal quotienting over indexed categorical data. The Fubini principle for coends is therefore about coherence of universal constructions.

2.2 Iterated coends and “order of integration”

An iterated coend arises when one first forms a coend in one variable and then a second coend in another. Such expressions often appear when a bifunctor depends on several variables with mixed variance. The order of evaluation can matter syntactically but not, under suitable conditions, mathematically.

The phrase “order of integration” is used by analogy only. In categorical practice, one is rearranging nested universal constructions rather than integrating functions. Nonetheless, the intuition is helpful because it suggests that separate indexing variables may be treated independently and then combined.

2.3 Basic interchange patterns

The most common pattern is an isomorphism between \(\int^{c}\int^{d} H(c,d)\) and \(\int^{d}\int^{c} H(c,d)\). Another useful pattern identifies either iterated coend with a single coend over \(\mathcal{C} \times \mathcal{D}\), when the functor is appropriately defined on the product category. These rearrangements are often canonical.

Such interchange laws are especially effective when a coend is used to define composition of profunctors, tensor products of modules, or convolution products. In these settings, the Fubini theorem provides the formal justification for reassociating multiple quotient-like constructions.

2.4 When coends are viewed as weighted colimits

A coend can be interpreted as a weighted colimit in which the weight is given by a hom-functor or a related bifunctor. This viewpoint places the Fubini theorem within the broader calculus of weighted colimits, where multiple layers of universal construction may be reorganized.

From this perspective, the theorem is a statement about the stability of weighted colimits under iteration. It helps explain why coend formulas often behave like algebraic expressions that can be regrouped and simplified.

3 Statement of the Fubini Theorem for Coends

The theorem is typically stated for a functor of several variables with suitable variance, and it asserts a canonical equivalence between iterated and combined coends. The precise formulation depends on the ambient category and the existence of the relevant colimits.

3.1 Setup: two indexing variables and a bifunctor

Let \(\mathcal{C}\) and \(\mathcal{D}\) be categories, and let \(H\) be a functor defined on the relevant product of opposites and ordinary categories, such as \[ H : \mathcal{C}^{op} \times \mathcal{C} \times \mathcal{D}^{op} \times \mathcal{D} \to \mathcal{E}. \] The theorem concerns the comparison between iterated coends in the \(\mathcal{C}\) and \(\mathcal{D}\) variables and the coend over the product indexing structure.

In many treatments, the statement is simplified to a bifunctor \(H : \mathcal{C}^{op} \times \mathcal{C} \times \mathcal{D}^{op} \times \mathcal{D} \to \mathcal{E}\) whose coend variables can be grouped in either order. The resulting formulas are then compared via canonical maps induced by universality.

3.2 Canonical comparison morphisms

The basic theorem provides canonical morphisms between the two iterated coends: \[ \int^{c}\int^{d} H(c,d) \longrightarrow \int^{d}\int^{c} H(c,d), \] and often between each iterated coend and the coend over the product category. These maps are constructed from the universal properties of the coends and the dinatural structure of \(H\).

In favorable situations, the comparison morphisms are isomorphisms. When this happens, the coend expression is independent of the order in which the variables are eliminated. This is the categorical analogue of order invariance in multiple integration.

3.3 Conditions ensuring isomorphism of iterated coends

The comparison becomes an isomorphism when the relevant coends exist and the ambient category has enough colimits to support the required constructions. Typical hypotheses include cocompleteness, smallness of the indexing categories, and compatibility between the functorial variance and the colimit structure.

The theorem is often presented as a formal consequence of colimit properties in complete and cocomplete categories. In enriched settings, additional assumptions may be needed, such as tensoredness or completeness and cocompleteness of the enriching category.

3.4 Enriched vs. unenriched formulations

In unenriched category theory, coends are built from ordinary hom-sets and ordinary colimits. In enriched category theory, the coend takes values in a monoidal base category \(\mathcal{V}\), and all maps and universal properties are interpreted in the enriched sense.

The enriched version is more general and more subtle. It is particularly important in contexts where hom-objects carry extra structure, such as modules over a ring, topological spaces, or chain complexes. The Fubini theorem persists in this setting, but its statement is expressed in \(\mathcal{V}\)-categorical language.

4 Existence and Size Conditions

The theorem depends on the existence of the coends being compared. Since coends are colimits, they require suitable completeness or cocompleteness assumptions, along with attention to size issues.

4.1 Smallness assumptions on indexing categories

A common hypothesis is that the indexing categories are small, ensuring that the relevant coproducts and coequalizers are set-indexed or otherwise manageable. Smallness prevents foundational problems and makes the coends accessible in standard categorical frameworks.

If a category is large, one may need universe conventions or local smallness assumptions. These restrictions are not specific to the Fubini theorem but are necessary for any rigorous treatment of coends in broad generality.

4.2 Colimit existence hypotheses

Since coends are colimits, their existence often requires that the target category have the relevant coproducts and coequalizers. In a cocomplete category, this is automatic for small indexing categories. The theorem then applies whenever both sides of the proposed interchange exist.

In practice, one often checks that the functor under consideration preserves the colimits needed to form the iterated coends. This ensures that the comparison maps can be constructed and that the universal properties are stable under rearrangement.

4.3 Compatibility with enrichment (e.g., V-cocomplete categories)

For enriched coends, one needs the enriching category \(\mathcal{V}\) to support the appropriate weighted colimits, and the target \(\mathcal{V}\)-category to be cocomplete in the enriched sense. Conditions such as \(\mathcal{V}\)-tensoredness or \(\mathcal{V}\)-cocompleteness are common.

These assumptions guarantee that the enriched coends behave analogously to ordinary ones. They also allow the interchange theorem to be stated and proved using enriched universal properties.

4.4 Managing coends over product categories

A coend over a product category can often be decomposed into nested coends, provided the relevant colimits commute. The product indexing structure must be handled carefully because the variance in each variable must match the category-theoretic direction of the coend.

The theorem is particularly convenient when one wants to reduce a complicated coend over several variables to a sequence of simpler ones. The product-category viewpoint clarifies how independent variables contribute to a combined universal construction.

5 Proof Strategies

Proofs of the Fubini theorem for coends usually rely on universal properties, coequalizer presentations, or general results about commuting colimits. The argument is often formal once the ambient hypotheses are in place.

5.1 Using universal properties and universality arguments

A standard proof shows that both sides represent the same dinatural universal property. If two objects are universal for the same class of dinatural cocones, then they are canonically isomorphic. This approach is conceptually clean and avoids explicit computation whenever possible.

The comparison morphisms are built from the universal dinatural maps into each coend. One then checks that they are inverse to each other by uniqueness of the universal factorization.

5.2 Colimit calculus: rearranging colimits step-by-step

Another approach expands each coend as a coequalizer of coproducts and then uses ordinary colimit calculus to rearrange the resulting diagram. Since coproducts and coequalizers in suitable categories satisfy their own interchange properties, the desired equivalence follows.

This method is constructive and often useful in concrete categories. It also reveals how the Fubini theorem depends on the commutation of colimits with one another, not on any special property unique to coends.

5.3 Fubini-style proofs via coend as coequalizer (when applicable)

When a coend is expressed as a coequalizer, iterated coends become iterated coequalizers. Under standard hypotheses, such iterated coequalizers can be reorganized into a single coequalizer associated with the product indexing category. This recovers the interchange theorem in a hands-on way.

This style of argument is especially common in textbooks and in applications to modules, profunctors, or functor categories. It shows explicitly how the relations imposed by each variable combine.

5.4 Naturalness checks for comparison maps

Any canonical comparison between two coend expressions must be natural in the underlying bifunctor. Verifying this naturality is an important part of the proof, especially when the theorem is used inside larger categorical constructions.

Naturalness ensures that the interchange is not merely an accidental isomorphism for one diagram. Instead, it is a coherent transformation that can be substituted into larger formulas without breaking functoriality.

6 Enriched and Monoidal Contexts

Coends are especially prominent in enriched and monoidal category theory, where they model compositions, tensor products, and convolution-like operations. The Fubini theorem is often indispensable in these settings.

6.1 Coends in enriched category theory

In enriched category theory, hom-objects take values in a monoidal category \(\mathcal{V}\), and coends are formed using \(\mathcal{V}\)-weighted universal properties. This enriched framework generalizes ordinary categories and captures additional structure such as linearity or topology.

The Fubini theorem remains valid in many enriched contexts, provided the relevant \(\mathcal{V}\)-colimits exist. Its role is to permit rearrangement of iterated enriched constructions in a controlled way.

6.2 Interaction with tensor products and monoidal structures

Monoidal structure often interacts closely with coends. For example, tensor products defined by coend formulas can be reassociated by applying the Fubini theorem. This is common in convolution products and in the study of categories of modules or functors.

The tensor product may also distribute over coends under suitable conditions. Such identities are central to simplifying expressions in monoidal category theory and to proving coherence results.

6.3 Coends as compositions of profunctors

A classic use of coends is to define the composition of profunctors. If \(P : \mathcal{A}^{op} \times \mathcal{B} \to \mathbf{Set}\) and \(Q : \mathcal{B}^{op} \times \mathcal{C} \to \mathbf{Set}\), their composite is often given by a coend over \(\mathcal{B}\). The Fubini theorem then helps analyze multiple compositions.

This viewpoint highlights the operational nature of coends. They are not merely abstract colimits but a practical mechanism for composing generalized relations or correspondences.

6.4 Compatibility with co/ends in closed monoidal categories

In a closed monoidal category, internal homs provide additional structure that often makes coend formulas more flexible. Coends and ends may be related by adjunctions involving tensor-hom pairs, and Fubini-type interchange can be combined with such dualities.

This compatibility is useful when translating between tensor expressions and hom-valued formulas. It can simplify both theoretical proofs and explicit computations.

7 Examples and Computations

Concrete examples help show how the theorem operates in practice. Even when the formulas look abstract, the underlying rearrangement is often straightforward once the coend notation is unpacked.

7.1 Simple interchangeable-coend examples

A basic example is a coend over two finite indexing categories where the target is a cocomplete category such as \(\mathbf{Set}\) or \(\mathbf{Vect}\). In such cases, one can compute the iterated coends in either order and obtain canonically isomorphic results.

These examples are useful pedagogically because they demonstrate that the theorem is not tied to sophisticated enrichment. The essential phenomenon already appears in ordinary categorical settings.

7.2 Coend expressions for tensoring and convolution

Many tensor-like operations are defined by coends. For instance, a convolution product on functors or presheaves may be written as a coend over an intermediate index category. If several such operations are nested, the Fubini theorem permits a systematic rearrangement.

This is particularly helpful when deriving associativity formulas. One can transform a nested coend expression into a form where associativity or symmetry becomes transparent.

7.3 Fubini theorem used in rewriting algebraic formulas

Coend interchange is a common algebraic tool for rewriting expressions involving natural transformations, bimodules, or functor compositions. By exchanging the order of coends, one can often match a complicated formula to a known normal form.

Such rewrites are valuable in proofs that would otherwise involve long chains of elementwise manipulations. The coend formalism replaces those computations with structural isomorphisms.

7.4 Diagrammatic intuition with string diagrams (light treatment)

In diagrammatic language, a coend often corresponds to “gluing” or “closing” a wire labeled by an object variable. The Fubini theorem then says that the order in which independent wires are closed does not matter, provided the diagram is well-typed.

This intuition is not a substitute for the formal proof, but it helps explain why the theorem is expected. String diagrams can make the interchange of variables feel natural and geometrically coherent.

The Fubini theorem for coends sits among several closely related interchange and duality principles. These variants are often used together in categorical calculations.

8.1 Fubini theorem for ends (dual statements)

The dual theorem for ends states that nested ends may be interchanged under analogous conditions. Since ends are limits, the proof mirrors the coend case with arrows reversed.

Together, the end and coend versions provide a balanced toolkit for handling both universal and co-universal constructions. Many arguments in category theory use one in tandem with the other.

8.2 Coend–end interchange principles

In some contexts, coends and ends can be interchanged under additional hypotheses, especially when the ambient category has strong completeness, cocompleteness, or closure properties. Such results are more delicate than the pure coend Fubini theorem.

These interchange principles are important in the theory of adjunctions, enriched homs, and representation formulas. They often allow a formula to be transformed into a more usable shape.

8.3 Beck–Chevalley-type comparisons

Beck–Chevalley conditions describe when certain pullback-like and pushforward-like operations commute. Coend interchange results can be viewed as part of a broader family of comparison theorems in which universal constructions are transported across indexing changes.

The connection is conceptual rather than identical. Both settings concern the compatibility of categorical constructions under rearrangement, and both rely on canonical comparison morphisms.

8.4 Tambara/Day convolution connections (conceptual overview)

Day convolution is a way of defining monoidal structures on functor categories using coends. Tambara-style constructions likewise involve coend formulas in the presence of additional algebraic structure. The Fubini theorem is often used in proving associativity or coherence for these constructions.

These connections show that the theorem is not merely an abstract identity. It underlies a broad range of constructions in modern category theory and categorical algebra.

9 Practical Usage: Rewriting and Normal Forms

Coend calculus is often about manipulating expressions into a form where known theorems apply. The Fubini theorem is one of the main tools for that purpose.

9.1 Converting between equivalent coend expressions

A complicated expression may be equivalent to a simpler one after a sequence of coend interchanges and variable renamings. The theorem provides the justification for moving from one arrangement to another without changing the resulting object.

This is especially useful when matching an expression to a standard representation theorem or to a known construction such as a composition formula.

9.2 Systematic normalization via iterated coends

Iterated coends can often be normalized by choosing a preferred order of variables and rewriting all expressions to that order. Once a normal form is fixed, comparisons between different formulas become easier.

Normalization is a practical strategy in both handwritten proofs and formal developments. It reduces the chance of error and makes coherence conditions more visible.

9.3 Avoiding common pitfalls (coend over the wrong variance)

A frequent mistake is to place a variable in the wrong variance position, which makes the coend expression ill-formed or changes its meaning. Another common issue is assuming that coends can always be interchanged without checking existence conditions.

Careful bookkeeping of covariance and contravariance is essential. The Fubini theorem applies only when the variables and the ambient colimit structure are arranged correctly.

9.4 Automating rewrites in formal proof assistants (high-level)

In formal proof assistants, coend manipulations are often implemented through rewrite rules, simp lemmas, or equivalence libraries. The Fubini theorem becomes one of the core transformations used to normalize categorical expressions.

Automation is most effective when the relevant canonical isomorphisms are registered coherently. This reduces the need for manual intervention while preserving mathematical rigor.

10 Notation Guide and Conventions

Because coend notation is compact, conventions matter. Clear notation helps prevent confusion when multiple variables and nested universal constructions appear in the same formula.

10.1 Common variance conventions

A coend variable typically appears once contravariantly and once covariantly in the same functor. This convention reflects the balanced nature of dinaturality. Writers often indicate the variance explicitly by using opposite categories in the domain.

Consistent variance notation is essential for reading and checking coend formulas. It also helps distinguish coends from ordinary colimits indexed by a single variable.

10.2 Coend/End naming and dummy variable practices

Bound variables in coends are dummy variables and may be renamed freely. This is useful when comparing two expressions or preparing for an interchange of integration order. The same practice applies to ends.

Such renaming is more than cosmetic. It helps avoid clashes of notation and makes the structure of nested formulas easier to see.

10.3 Interchange notation for iterated coends

Iterated coends are commonly written with nested integral signs, such as \(\int^{c}\int^{d}\). Depending on context, one may also write a single coend over a product category. Both forms may denote canonically isomorphic objects under the Fubini theorem.

The choice of notation often reflects the intended emphasis. Nested notation highlights the order of construction, while product notation emphasizes symmetry and combined indexing.

10.4 Reference list of standard coend identities

Several standard identities recur throughout coend calculus: functoriality under natural isomorphism, reassociation over product categories, coend-end duality, and expressions of coends as coequalizers. These identities form the basic toolkit for manipulating categorical formulas.

The Fubini theorem is one of the central entries in this toolkit. It supplies the key interchange principle that makes many advanced coend calculations possible.