Yoneda lemma

1 Statement of the Yoneda lemma

The Yoneda lemma is a recognition principle in category theory: it identifies natural transformations out of a representable Hom-functor with elements of the target functor evaluated at the representing object. This turns otherwise “global” data (a natural transformation) into “local” data (an element).

1.1 The locally small setting

A category \(\mathcal{C}\) is *locally small* if for all objects \(A,B\) the hom-collection \(\mathcal{C}(A,B)\) is a set. In this setting, the usual Yoneda statement concerns set-valued functors and set-based natural transformations.

Fix an object \(A\in \mathcal{C}\). The (covariant) representable functor associated to \(A\) is \[ \mathrm{Hom}(A,-):\mathcal{C}\to \mathbf{Set},\quad X\mapsto \mathcal{C}(A,X), \] and the (contravariant) representable functor is \[ \mathrm{Hom}(-,A):\mathcal{C}^{op}\to \mathbf{Set},\quad X\mapsto \mathcal{C}(X,A). \]

1.2 Covariant form

Let \(F:\mathcal{C}\to \mathbf{Set}\) be a covariant functor. The covariant Yoneda lemma states that for every object \(A\in\mathcal{C}\) there is a natural bijection \[ \mathrm{Nat}\big(\mathrm{Hom}(A,-),F\big)\;\cong\; F(A). \] Here \(\mathrm{Nat}(\_,\_)\) denotes the set of natural transformations between functors.

1.3 Contravariant form

Let \(F:\mathcal{C}^{op}\to \mathbf{Set}\) be a contravariant functor. The contravariant Yoneda lemma states that for every \(A\in\mathcal{C}\) there is a natural bijection \[ \mathrm{Nat}\big(\mathrm{Hom}(-,A),F\big)\;\cong\; F(A). \]

1.4 Natural transformations as elements

The key content of the lemma is that a natural transformation \(\eta:\mathrm{Hom}(A,-)\Rightarrow F\) is completely determined by (and corresponds to) a single element of \(F(A)\). Concretely, the transformation “evaluated at the representing object” yields that element, and conversely any element produces a transformation.

1.5 The explicit correspondence

In the covariant case, given \(\eta:\mathrm{Hom}(A,-)\Rightarrow F\), define an element \[ \eta_A(\mathrm{id}_A)\in F(A). \] Conversely, given an element \(x\in F(A)\), define a natural transformation \(\eta^x:\mathrm{Hom}(A,-)\Rightarrow F\) by specifying, for each object \(X\), a function \[ \eta^x_X:\mathcal{C}(A,X)\to F(X),\quad f\mapsto F(f)(x), \] where \(F(f):F(A)\to F(X)\) is the action of \(F\) on morphisms.

In the contravariant case, the roles are analogous: for \(\eta:\mathrm{Hom}(-,A)\Rightarrow F\), the corresponding element is \(\eta_A(\mathrm{id}_A)\in F(A)\); from \(x\in F(A)\) one sets \[ \eta^x_X:\mathcal{C}(X,A)\to F(X),\quad g\mapsto F(g)(x), \] with \(F(g):F(A)\to F(X)\) interpreted appropriately using contravariance.

1.6 Functoriality in the representing object

The correspondence is not merely a bijection of sets; it is *natural* in the representing object \(A\) as well as in the functor \(F\). Informally, if \(u:A'\to A\) is a morphism in \(\mathcal{C}\), then the element corresponding to a transformation changes compatibly with precomposition by \(u\) (or postcomposition, depending on the variance). This natural dependence is what enables Yoneda to function as an identification across the category, rather than as a one-off counting argument.

2 Yoneda embedding

The Yoneda lemma yields a canonical way to view each object of \(\mathcal{C}\) as a representable functor. This is formalized by the Yoneda embedding, which is fundamental for comparing abstract categories to concrete functor categories.

2.1 Definition of the Yoneda embedding

For a locally small category \(\mathcal{C}\), the covariant Yoneda embedding is the functor \[ Y:\mathcal{C}\to [\mathcal{C}^{op},\mathbf{Set}],\quad A\mapsto \mathrm{Hom}(-,A). \] Often one uses the contravariant representable functors to ensure the codomain is a functor category \([\mathcal{C}^{op},\mathbf{Set}]\). Alternatively, depending on conventions, one can embed using \(\mathrm{Hom}(A,-)\) into \([\mathcal{C},\mathbf{Set}]\). In either setup, representability is the organizing principle.

On morphisms: given \(u:A\to A'\) in \(\mathcal{C}\), the induced natural transformation \[ Y(u):\mathrm{Hom}(-,A)\Rightarrow \mathrm{Hom}(-,A') \] is defined componentwise by postcomposition: \[ Y(u)_X:\mathcal{C}(X,A)\to \mathcal{C}(X,A'),\quad f\mapsto u\circ f. \]

2.2 Full faithfulness

The Yoneda lemma implies that \(Y\) is fully faithful. Concretely, \[ \mathrm{Hom}_{\mathcal{C}}(A,A')\;\cong\; \mathrm{Nat}\big(\mathrm{Hom}(-,A),\mathrm{Hom}(-,A')\big), \] and this bijection is natural in both variables. Thus, morphisms in \(\mathcal{C}\) can be recovered exactly as natural transformations between the corresponding representable functors.

2.3 Interpretation of objects via representable functors

Through Yoneda embedding, every object \(A\) is represented by the functor \(\mathrm{Hom}(-,A)\) (or \(\mathrm{Hom}(A,-)\)). This perspective treats an object as “the thing that \(A\) sees” via all incoming or outgoing maps. Two objects are isomorphic precisely when their representable functors are naturally isomorphic, since full faithfulness turns that natural equivalence back into an isomorphism in the original category.

2.4 Recovering morphisms from natural transformations

A practical interpretation of full faithfulness is that a natural transformation between representable functors corresponds to a unique morphism of \(\mathcal{C}\). In the standard contravariant embedding, a natural transformation \(\theta:\mathrm{Hom}(-,A)\Rightarrow \mathrm{Hom}(-,A')\) is determined by the image of \(\mathrm{id}_A\) under \(\theta_A\), and that image yields the morphism \(A\to A'\). This makes the Yoneda embedding a bridge between categorical morphisms and functorial behavior.

3 Consequences and applications

The lemma has numerous structural consequences. Many results in category theory use Yoneda as a reduction tool: instead of reasoning about arbitrary functors directly, one reasons about natural transformations out of representables.

3.1 Representables determine functors

Representable functors act as test objects for naturality. If two functors \(F,G:\mathcal{C}\to \mathbf{Set}\) satisfy \[ \mathrm{Nat}(\mathrm{Hom}(A,-),F)\cong \mathrm{Nat}(\mathrm{Hom}(A,-),G) \] naturally in \(A\), then the Yoneda lemma forces \(F\) and \(G\) to be isomorphic. More generally, knowing how \(F\) behaves on elements is determined by its behavior on morphisms into \(F\) from representables.

3.2 Characterization of natural transformations

Yoneda provides a method to construct or characterize natural transformations: to define a natural transformation \(\eta:\mathrm{Hom}(A,-)\Rightarrow F\), it suffices to choose an element of \(F(A)\). Conversely, if \(\eta\) is given, one computes the corresponding element as \(\eta_A(\mathrm{id}_A)\). This “elementwise generation” is frequently used in proofs.

3.3 Limits, colimits, and representability

Limits and colimits can often be detected or expressed via representability. For example, if an object \(L\) represents a functor encoding a universal property, then that representability translates the universal property into a natural isomorphism. Yoneda is central in showing that certain universal constructions correspond to representable functors, and that representability yields uniqueness up to isomorphism.

3.4 Adjunctions through representability

Adjunctions are frequently characterized by representable functors. A common pattern is that an adjunction \(L\dashv R\) yields natural bijections \[ \mathcal{D}(L A, B)\cong \mathcal{C}(A, R B), \] which can be viewed as an isomorphism between representable functors. Yoneda can then convert these bijections into statements about the corresponding unit and counit morphisms, as well as about how morphisms transpose across the adjunction.

3.5 (Co)representing objects and universal properties

If a functor \(F:\mathcal{C}\to \mathbf{Set}\) is representable, meaning \(F\cong \mathrm{Hom}(A,-)\) for some \(A\), then \(A\) is a representing object for \(F\). The universal property embedded in representability often yields an efficient way to prove uniqueness: any other candidate representing object must be uniquely isomorphic to \(A\). Dually, corepresentable functors use \(\mathrm{Hom}(-,A)\) and lead to dual universal properties.

4 Generalizations

Yoneda’s basic idea persists beyond ordinary set-valued functors. It adapts to enriched category theory and to higher-categorical contexts, where hom-objects are not merely sets.

4.1 Enriched Yoneda lemma

In an enriched category \(\mathcal{C}\) enriched over a monoidal category \(\mathcal{V}\), hom-objects \(\mathcal{C}(A,B)\) live in \(\mathcal{V}\) rather than in \(\mathbf{Set}\). Representable enriched functors and natural transformations must be interpreted in the enriched sense (e.g., using \(\mathcal{V}\)-natural transformations). The enriched Yoneda lemma then identifies the enriched natural transformations out of a representable with the corresponding enriched hom-object evaluated at the representing data. The precise statement depends on the enrichment framework, but the conceptual mechanism remains the same: naturality is controlled by the representing morphism \(\mathrm{id}_A\).

4.2 Yoneda for higher-categorical settings (overview)

Higher categories replace sets of morphisms with higher-dimensional structure (2-morphisms, 3-morphisms, and so on). A higher-categorical version of Yoneda often treats representability up to equivalence rather than up to strict equality, and naturality may hold up to specified coherent isomorphisms or homotopies. The general theme is that “mapping objects” represent functorial behavior, and that the universal properties governing those mapping objects still encode higher natural transformations.

4.3 Variants using ends and coends

Ends and coends provide a formal calculus for expressing universal constructions and natural transformations. Variants of the Yoneda lemma can be phrased using ends: for instance, the set (or enriched object) of natural transformations between two functors can be expressed as an end. In that language, the representability of one functor collapses the end to a simpler expression, yielding the Yoneda correspondence. Similarly, coends appear in dual formulations and in contexts where one integrates over varying variables with colimit-like behavior.

4.4 Yoneda lemma in 2-category contexts (overview)

In a 2-category, there are objects, 1-morphisms, and 2-morphisms. A Yoneda-style result there relates 2-naturality (and possibly modifications between 2-natural transformations) to data determined by the identity 1-morphism. While the bookkeeping is more intricate, the central principle continues: representable 2-functors control higher natural transformations, enabling embeddings of objects into functor 2-categories and allowing reconstruction of morphisms from transformation data.

5 Computational viewpoints

The Yoneda lemma also supports concrete computation with natural transformations. Many workflows in practice involve turning element-level choices into global naturality conditions and then checking coherence mechanically.

5.1 Using the lemma to build natural transformations

To construct a natural transformation \(\eta:\mathrm{Hom}(A,-)\Rightarrow F\), one chooses an element \(x\in F(A)\). The lemma then dictates the value at every object \(X\) and every morphism \(f:A\to X\) via \[ \eta_X(f)=F(f)(x). \] This reduces the problem of defining a transformation into the simpler task of selecting a single element and letting functoriality propagate the rest.

5.2 Verifying naturality via elementwise checks

The definition above makes naturality largely automatic: because \(\eta\) is expressed through functor action \(F(-)\), commutativity of the naturality squares follows from the functorial identity \[ F(g\circ f)=F(g)\circ F(f) \] together with \(\mathrm{id}\)-behavior. In practical proofs, one often checks naturality by chasing an element \(f\in \mathcal{C}(A,X)\) through the relevant square and observing that both routes compute to \(F(\text{composition})(x)\).

5.3 Examples with Set-valued functors

Let \(F:\mathcal{C}\to \mathbf{Set}\) be a set-valued functor. Choosing \(x\in F(A)\) defines \(\eta^x\). The lemma guarantees that every natural transformation arises this way, so the construction is exhaustive. This viewpoint is especially helpful when \(F(A)\) has a concrete interpretation (e.g., structured maps, states, or labeling schemes) and one wants to describe transformations in terms of those elements.

5.4 Examples with presheaves and sheaves (basic outline)

Presheaves are functors \(\mathcal{C}^{op}\to \mathbf{Set}\). For a presheaf \(P\), a natural transformation \(\mathrm{Hom}(-,A)\Rightarrow P\) corresponds to an element of \(P(A)\). When one introduces sheaves, additional gluing axioms constrain which presheaves come from geometric or topological data, but the Yoneda correspondence at the presheaf level still provides the foundational elementwise description of natural maps from representables. In that setting, representables serve as local probes for the presheaf’s behavior.

6 Relationship to other core ideas

Yoneda sits at the crossroads of several standard concepts: functor categories, representability, and adjunctions. Its role is to connect these notions through natural transformations.

6.1 Functor categories and naturality

A functor category \([\mathcal{C},\mathbf{Set}]\) (or its variants) organizes functors as objects and natural transformations as morphisms. Yoneda explains why representable functors are particularly informative inside such categories: natural transformations from representables correspond to elements of target functors, allowing one to interpret morphisms in the functor category in terms of element-level data.

6.2 Representable functors

Representable functors are those isomorphic to \(\mathrm{Hom}(A,-)\) or \(\mathrm{Hom}(-,A)\). The Yoneda lemma characterizes how representable functors interact with arbitrary functors: they act like canonical generators for the functor category’s mapping behavior. This also implies that representable functors are rigid: if two representables are naturally isomorphic, then the representing objects are isomorphic.

6.3 Adjoint functors and hom-set adjunctions

Adjoint functors often arise from representability of hom-functors. Since \(\mathrm{Hom}(-,A)\) and \(\mathrm{Hom}(A,-)\) are themselves representable, Yoneda helps translate between hom-set bijections (or natural isomorphisms between representables) and units/counits of adjunctions. Thus, adjunctions can be studied by analyzing how mapping properties correspond to naturality constraints.

6.4 Density and comparison theorems (high-level)

In broader settings, representable functors can form dense subcategories, enabling comparison theorems: one can reconstruct a functor or a natural transformation from its behavior on representables. While the detailed statements depend on additional hypotheses (such as smallness conditions and the ambient category), Yoneda’s insight is the conceptual foundation behind these reconstruction principles.

7 Examples

The following examples illustrate how the Yoneda lemma works concretely, emphasizing how element choices correspond to natural transformations.

7.1 Representable functors in Set

Take \(\mathcal{C}=\mathbf{Set}\). For a set \(A\), the covariant representable functor \(\mathrm{Hom}(A,-)\) assigns to each set \(X\) the set of functions \(A\to X\). If \(F:\mathbf{Set}\to\mathbf{Set}\) is any functor, then natural transformations \(\mathrm{Hom}(A,-)\Rightarrow F\) correspond bijectively to elements of \(F(A)\). In practice, an element \(x\in F(A)\) determines a transformation by sending each function \(f:A\to X\) to \(F(f)(x)\).

7.2 Natural transformations between Hom-functors

Let \(A,A'\) be objects of \(\mathcal{C}\). Consider natural transformations \[ \theta:\mathrm{Hom}(-,A)\Rightarrow \mathrm{Hom}(-,A'). \] By Yoneda, such \(\theta\) corresponds to an element of \(\mathrm{Hom}(A,A')\), i.e., to a specific morphism \(u:A\to A'\). Under this correspondence, \(\theta_X\) sends a map \(f:X\to A\) to the composite \(u\circ f:X\to A'\). This example shows how Yoneda recovers morphisms as transformation data.

7.3 Presheaf perspective

Let \(P:\mathcal{C}^{op}\to\mathbf{Set}\) be a presheaf. A natural transformation \(\mathrm{Hom}(-,A)\Rightarrow P\) is the same as an element of \(P(A)\). The element can be viewed as a “generalized element” of the presheaf at stage \(A\). Naturality ensures that the chosen element propagates along morphisms into \(A\), exactly matching the presheaf’s restriction maps.

7.4 Metric-free examples in abstract categories (setup-focused)

In an abstract category \(\mathcal{C}\) without additional structure such as topology or metrics, one can still apply Yoneda purely formally. For instance, suppose \(F:\mathcal{C}\to\mathbf{Set}\) interprets each object \(X\) as a set of “configurations” of type \(X\). Then an element \(x\in F(A)\) encodes a rule for assigning to each morphism \(f:A\to X\) a configuration in \(F(X)\), namely \(F(f)(x)\). The naturality condition guarantees compatibility with composition in \(\mathcal{C}\), so the rule depends only on the categorical wiring of maps rather than on any metric or analytic structure.