Representable Functors and the Yoneda Lemma

The Yoneda Lemma states that for any locally small category

C

, a functor

F : C \to Set

, and an object

A

, there is a natural bijection

Nat (C (A, -), F) ≅ F (A)

. In words: natural transformations out of a representable functor are completely determined by a single element. It applies throughout category theory, algebraic geometry, homotopy theory, and type theory. Proved/named after Nobuo Yoneda (c. 1954, via a famous discussion with Mac Lane).

A representable functor is one isomorphic to a hom-functor $C (A, -) : C \to Set$ (or contravariantly $C (-, A)$ ); the object $A$ is its representing object. The Yoneda Lemma computes all natural transformations from such a hom-functor into an arbitrary $Set$ -valued functor, showing they correspond exactly to elements of the functor evaluated at the representing object. This matters because it makes "an object is determined by its system of maps" into a precise theorem: the Yoneda embedding is fully faithful, so any category sits inside its category of presheaves without loss of information. The surprising fact is how much follows from such a one-line statement — uniqueness of universal properties, the structure of limits, and the entire functor-of-points philosophy in algebraic geometry all descend from it.

The Core Insight

To know an object $A$ , it suffices to know the set of arrows into $A$ from every other object, together with how those sets transform under composition. The hom-functor $C (-, A)$ packages exactly this data, and Yoneda says no information is lost: the functor remembers $A$ up to unique isomorphism. A natural transformation $C (A, -) \Rightarrow F$ is "as good as" a single element of $F (A)$ , because $id_{A} \in C (A, A)$ is a generic point that the transformation must send somewhere, and naturality forces everything else.

Formal Statement

Hom-Functor

Let $C$ be locally small (each $C (X, Y)$ is a set). For a fixed object $A \in C$ , the covariant hom-functor $h^{A} = C (A, -) : C \to Set$ is defined on objects and morphisms by

h^{A} (X) = C (A, X), h^{A} (f) = f \circ (-) : C (A, X) \to C (A, Y) for f : X \to Y .

⟳post-composition by

f

sends

g : A \to X

f \circ g : A \to Y

; functoriality

h^{A} (g \circ f) = h^{A} (g) \circ h^{A} (f)

follows from associativity of composition.

Dually, the contravariant hom-functor $h_{A} = C (-, A) : C^{op} \to Set$ sends $X \mapsto C (X, A)$ and $f : X \to Y$ to precomposition $(-) \circ f : C (Y, A) \to C (X, A)$ .

Representable Functor

A functor $F : C \to Set$ is representable if there exists an object $A \in C$ and a natural isomorphism

η : C (A, -) ≅ F .

⟳a natural isomorphism is a natural transformation with components

η_{X}

that are bijections for every

X

; the pair

(A, η)

is a

representation of $F$ . A contravariant functor $F : C^{op} \to Set$ is representable if $F ≅ C (-, A)$ .

The object $A$ is the representing object; by the universal element correspondence (below) a representation is equivalent to a universal element $u = η_{A} (id_{A}) \in F (A)$ .

The Yoneda Lemma

Let $C$ be locally small, $F : C \to Set$ a functor, and $A \in C$ . Then the map

Φ : Nat (C (A, -), F) ≅ F (A), Φ (α) = α_{A} (id_{A}),

is a bijection, natural in both $A$ and $F$ .

⟳

α_{A}

is the component of the natural transformation at the object

A

; evaluating it on the identity morphism

id_{A} \in C (A, A)

produces an element of

F (A)

. Naturality in

A

and

F

means

Φ

commutes with the actions of morphisms

A^{'} \to A

and natural transformations

F \to F^{'}

Yoneda Embedding

The assignment $A \mapsto C (-, A)$ defines a functor, the Yoneda embedding,

よ : C ⟶ C := [C^{op}, Set], A \mapsto C (-, A),

which is full and faithful: for all $A, B$ ,

C (A, B) ≅ Nat (C (-, A), C (-, B)) .

⟳apply [3] to the contravariant version with

F = C (-, B)

evaluated at the representing object

A

; the universal element corresponding to a natural transformation is a morphism

A \to B

, giving the bijection.

Why It Holds / Derivation

Bijectivity of $\Phi$ via an explicit inverse

We construct $Ψ : F (A) \to Nat (C (A, -), F)$ inverse to $Φ$ .

Step 1 — Define $Ψ$ . Given

a \in F (A)

, define a transformation

Ψ (a)

with components

Ψ (a)_{X} : C (A, X) \to F (X), Ψ (a)_{X} (g) = F (g) (a) (g : A \to X) .

⟳

g : A \to X

induces

F (g) : F (A) \to F (X)

; apply it to the chosen element

a \in F (A)

Step 2 — $Ψ (a)$ is natural. For

f : X \to Y

we must show

F (f) \circ Ψ (a)_{X} = Ψ (a)_{Y} \circ h^{A} (f)

. Evaluating on

g \in C (A, X)

F (f) (F (g) (a)) = F (f \circ g) (a) = Ψ (a)_{Y} (f \circ g) = Ψ (a)_{Y} (h^{A} (f) (g)) .

⟳functoriality

F (f) \circ F (g) = F (f \circ g)

supplies the first equality; the definition of

h^{A} (f) = f \circ (-)

gives the last.

Step 3 — $Φ \circ Ψ = id$ . Compute

Φ (Ψ (a)) = Ψ (a)_{A} (id_{A}) = F (id_{A}) (a) = id_{F (A)} (a) = a

⟳

F

preserves identities, so

F (id_{A}) = id_{F (A)}

Step 4 — $Ψ \circ Φ = id$ . Let

α \in Nat (C (A, -), F)

with

a = Φ (α) = α_{A} (id_{A})

. For any

g : A \to X

, naturality of

α

g

gives the commuting square

α_{X} (g) = α_{X} (h^{A} (g) (id_{A})) = F (g) (α_{A} (id_{A})) = F (g) (a) = Ψ (a)_{X} (g) .

⟳write

g = g \circ id_{A} = h^{A} (g) (id_{A})

, then apply the naturality square

α_{X} \circ h^{A} (g) = F (g) \circ α_{A}

id_{A}

. Hence

α = Ψ (a)

Steps 3–4 show $Φ$ and $Ψ$ are mutually inverse, proving the bijection. Naturality of $Φ$ in $F$ and $A$ is checked by chasing the analogous squares. $■$

The Yoneda naturality square: $\mathrm{id}_A$ generates everything

Nobuo Yoneda | 1930–1996

Japanese mathematician working in homological algebra (notably Yoneda Ext groups and the splice product of extensions). The lemma bearing his name emerged not from a published paper but from a 1954 conversation with Saunders Mac Lane in a Paris train station (the Gare du Nord), after which Mac Lane carried the idea back and popularized it. Yoneda's own contributions to the cohomology of groups and the long exact sequence machinery were substantial, but the lemma's deceptively simple statement made it the lens through which generations learned that objects live through their relationships.

Building Intuition

Think of $C (A, -)$ as the probe "what can map out of $A$ ?". A natural transformation $α : C (A, -) \Rightarrow F$ is a coherent rule assigning to every arrow out of $A$ an element of $F$ . But every arrow $g : A \to X$ is just $g \circ id_{A}$ — it is the image of the identity under post-composition. So once you decide where $id_{A}$ goes (an element $a \in F (A)$ ), naturality forces $g \mapsto F (g) (a)$ everywhere. The identity is a generic element: a single seed from which the whole transformation grows. This is why "to specify a map of structured objects, give the image of one universal element" appears all over mathematics — the determinant as a map out of free modules, a group homomorphism out of a free group determined on generators, a continuous map out of a representable space. Representability turns abstract universal properties into concrete elements you can point at.

Worked Examples

Representing the forgetful functor on groups

Given: Let

U : Grp \to Set

be the forgetful functor sending a group

G

to its underlying set

∣ G ∣

. We ask: is

U

representable, and by which group? Steps:

A representation needs $G_{0}$ with $Grp (G_{0}, G) ≅ ∣ G ∣$ naturally. The free group on one generator is $Z$ , and a homomorphism $ϕ : Z \to G$ is uniquely determined by $ϕ (1) \in ∣ G ∣$ , with every element achievable.
Thus $Grp (Z, G) ≅ ∣ G ∣$ via $ϕ \mapsto ϕ (1)$ ; check naturality: for $ψ : G \to H$ , $ψ \circ ϕ \mapsto ψ (ϕ (1))$ matches $U (ψ)$ applied to $ϕ (1)$ . ✓
By Yoneda, natural transformations $U \Rightarrow U$ correspond to elements of $U (Z) = Z$ . The element $n \in Z$ gives the transformation $g \mapsto g^{n}$ ("raise to the $n$ th power"), which is indeed natural.

Answer:

U ≅ Grp (Z, -), representing object Z; Nat (U, U) ≅ Z (the power operations) .

Counting natural transformations between hom-functors on a poset

Given: Let

C

be the poset category

{0 \leq 1 \leq 2}

, where

C (i, j)

has exactly one element if

i \leq j

and is empty otherwise. Compute

Nat (C (1, -), C (0, -))

. Steps:

By the (covariant) Yoneda Lemma applied with $F = C (0, -)$ and $A = 1$ :

Nat (C (1, -), C (0, -)) ≅ C (0, -) (1) = C (0, 1) .

Since $0 \leq 1$ , the set $C (0, 1)$ has exactly one element.
Sanity check directly: a transformation must send $id_{1} \in C (1, 1)$ to an element of $C (0, 1)$ — only one choice — and then naturality fixes the rest. (Contrast: $Nat (C (0, -), C (1, -)) ≅ C (1, 0) = \emptyset$ , since $1 \neq \leq 0$ .)

Answer:

∣ Nat (C (1, -), C (0, -)) ∣ = 1, ∣ Nat (C (0, -), C (1, -)) ∣ = 0.

Common Misconceptions

Local smallness is not optional. The bijection

Φ

presumes

C (A, X)

is a genuine set so that hom-functors land in

Set

. In a category that is not locally small,

C (A, -)

may not be definable as a

Set

-valued functor and the Yoneda statement fails to even type-check. Also note: the embedding is fully faithful but not surjective on objects — most presheaves are not representable, and a presheaf is representable iff its category of elements has a terminal object. "Naturally isomorphic" $\neq =$ "isomorphic in each component." Two functors can have

F (X) ≅ G (X)

for all

X

(componentwise bijections) without being naturally isomorphic. The Yoneda isomorphism's force comes from naturality — the commuting squares — not mere cardinality matching. Forgetting naturality is the single most common error in applying the lemma.

Consider $C = Grp$ and let $F (G) = Aut (G)$ — but treat it naively as a set-valued assignment ignoring that automorphisms don't transform functorially along arbitrary homomorphisms. There is no functor $Grp \to Set$ with $G \mapsto Aut (G)$ that is covariant in homomorphisms (a non-iso $ϕ : G \to H$ gives no induced map on automorphism groups), so asking whether " $Aut$ is representable" is ill-posed until one fixes the variance/functoriality. The lesson: representability questions require a bona fide functor first; many natural-looking assignments are not functorial at all, so Yoneda does not apply.

Connections

The Yoneda Lemma is the technical heart of Universal Properties: a universal element of $F$ is precisely a representation, and uniqueness up to unique isomorphism is the statement that representing objects are unique. It underlies Adjoint Functors — an adjunction $L ⊣ R$ is equivalent to natural isomorphisms $D (L A, B) ≅ C (A, R B)$ , i.e. representability of certain hom-functors. It powers Limits and Colimits, since a limit of $D$ represents the functor $X \mapsto lim C (X, D (-))$ , and Yoneda gives the uniqueness. Through the embedding $C ↪ [C^{op}, Set]$ it connects to Presheaves and Sheaves and the theory of Density and the Co-Yoneda Lemma, where every presheaf is a colimit of representables.

Real-World Applications

Algebraic geometry — functor of points. A scheme $X$ is studied via its functor $R \mapsto X (R) = Hom (Spec R, X)$ ; Grothendieck's foundations identify schemes with representable functors on rings, and moduli problems (e.g. the Hilbert scheme, moduli of elliptic curves) are representability questions.
Homotopy theory — Brown representability. Cohomology theories $H^{n} (-; A)$ are representable by Eilenberg–MacLane spaces $K (A, n)$ : $H^{n} (X; A) ≅ [X, K (A, n)]$ , a Yoneda-style identification of a functor with maps into a representing object.
Type theory and proof assistants. In Homotopy Type Theory and Coq/Agda formalizations, the Yoneda Lemma justifies "parametricity" reasoning and the equivalence of an object with its system of maps; it is a standard formalized result.
Database theory. Categorical databases (Spivak's work) model schemas as categories and instances as $Set$ -valued functors; queries are representable functors and the Yoneda Lemma describes data migration.
Programming language semantics. The Yoneda embedding appears as the CPS/continuation transform and as the "Yoneda trick" in Haskell, where Yoneda f a ≅ f a enables fusion optimizations.
Representation theory. Studying a group/algebra through its modules is studying it through hom-functors; Morita equivalence is naturally phrased via representability and equivalences of presheaf-like module categories.
Tannakian reconstruction. A group (or groupoid/gerbe) is recovered from its category of representations via a representability/Yoneda-type theorem — reconstructing the object from the functor of fiber points.

Historical Development

1942–1945 — Eilenberg and Mac Lane introduce categories, functors, and natural transformations, making "naturality" a precise notion and setting the stage.
1950s — Hom-functors and representability emerge implicitly in homological algebra (Cartan–Eilenberg).
c. 1954 — Nobuo Yoneda explains the lemma to Saunders Mac Lane at the Gare du Nord in Paris; Mac Lane disseminates and names it.
1958 — Kan introduces adjoint functors, whose theory is intertwined with representability.
1960s — Alexander Grothendieck founds the functor-of-points viewpoint in algebraic geometry (EGA/SGA), making representability the organizing principle of scheme theory and moduli.
1962 — E. H. Brown proves the Brown Representability Theorem in homotopy theory.
1971 — Mac Lane's Categories for the Working Mathematician gives the canonical textbook treatment, cementing notation and the embedding.
1990s–2000s — Representability and Yoneda become central to higher category theory (∞-categories, Lurie), via $\infty$ -presheaves and the $\infty$ -Yoneda Lemma.
2010s–present — Formalization in proof assistants (Coq, Lean, Agda) and the HoTT program restate Yoneda constructively and homotopically.

Subtleties & Limits

Size and largeness. For locally small but large categories, the presheaf category

[C^{op}, Set]

may itself fail to be locally small, and care with Grothendieck universes is needed; controlling these size issues (accessible/presentable categories, the small object argument) is a delicate ongoing concern in foundations. Higher Yoneda. Stating and proving the Yoneda Lemma for

(\infty, 1)

-categories required substantial machinery (simplicial/quasicategorical models); fully synthetic proofs in directed type theory and simplicial type theory are an active research frontier. Constructive and enriched versions. The enriched Yoneda Lemma (over a monoidal

V

) and its constructive content — when does representability hold without choice, and what is its computational meaning — remain areas of refinement, especially as proof assistants demand fully explicit witnesses.