The Yoneda Lemma: How Objects Are Known by Their Relationships

The precise statement

Fix a locally small category C — meaning Hom(x, y) is a genuine set for all objects x, y. For each object c, the assignment x ↦ Hom(c, x) is a functor hᶜ = Hom(c, −): C → Set, sending a morphism g: x → y to post-composition g ∘ (−).

The Yoneda Lemma asserts: for every functor F: C → Set,

• there is a bijection Nat(Hom(c, −), F) ≅ F(c);
• this bijection is natural in c ∈ C and in F ∈ [C, Set].

The left side is a set of natural transformations; the right side is a plain set. The content is that specifying an entire natural transformation α: Hom(c, −) ⇒ F requires no more data than choosing a single element of F(c). The contravariant twin replaces Hom(c, −) by Hom(−, c) and functors C → Set by presheaves Cᵒᵖ → Set.

Intuition: an object is its web of relationships

Think of Hom(c, −) as a universal probe. To understand any structure F, you interrogate it with maps: an element of F(x) is a "generalized point of shape c" once you know how c maps around. The lemma says the probe hᶜ is so faithful that a compatible family of answers over all objects and morphisms — a natural transformation — collapses to one answer at c itself.

The slogan is: you know an object by its relationships. Two objects with the same pattern of incoming (or outgoing) morphisms, functorially, are isomorphic. This is the categorical version of "a point is determined by the functions on it" (Gelfand duality), "a space by its open sets," or "a group by its representations." In physics-flavored language, you never observe the object directly; you observe how it interacts, and Yoneda promises those interactions carry the complete information — nothing is lost by refusing to look inside.

The key idea of the proof

The proof is short and is driven entirely by one element: the identity morphism id_c.

Given a natural transformation α: Hom(c, −) ⇒ F, its component α_c: Hom(c, c) → F(c) can eat id_c. Define

• Φ(α) = α_c(id_c) ∈ F(c).

Conversely, given u ∈ F(c), define a transformation Ψ(u) whose component at x sends f: c → x to F(f)(u) ∈ F(x).

The naturality square for α says exactly that α_x(f) = α_x(f ∘ id_c) = F(f)(α_c(id_c)). That is, α is forced everywhere by its value on id_c — so Ψ ∘ Φ = id. And Φ(Ψ(u)) = F(id_c)(u) = u, since F preserves identities, so Φ ∘ Ψ = id. The clever step is realizing that naturality does all the work: the single value at the identity propagates by functoriality to a fully determined transformation.

Worked example: the fullness of the embedding

The most-used special case takes F = Hom(d, −) itself. Then the lemma reads

• Nat(Hom(c, −), Hom(d, −)) ≅ Hom(d, c).

So natural transformations between representable functors are exactly morphisms of C (with direction reversed). Unwinding Φ and Ψ: a morphism k: d → c corresponds to the transformation "pre-compose with k," i.e. f ↦ f ∘ k. This says the Yoneda embedding よ: C ↪ [Cᵒᵖ, Set], c ↦ Hom(−, c), is full and faithful.

Concretely in Set: take c = 1, the one-point set. Then Hom(1, −) is (naturally) the identity functor, and Yoneda gives Nat(Hom(1,−), F) ≅ F(1) = the elements of F(1). "A generalized element of shape 1" is just an ordinary element — the lemma recovers the naïve notion of an element as a map from a point.

Why the hypotheses matter, and connections

Local smallness is essential. If Hom(c, x) is a proper class rather than a set, then Hom(c, −) is not a functor to Set, Nat(Hom(c,−), F) may not even form a set, and the statement fails to typecheck. For large categories one must pass to a larger universe or restrict to Set-valued functors carefully; this is why size conditions pervade category theory.

The target must be Set. The proof crucially evaluates α_c at the specific element id_c ∈ Hom(c, c) — an argument about elements. Enriched versions (Yoneda over a monoidal base V) hold, but you must reformulate "elements" via the unit object and prove an enriched analogue; you cannot just repeat the set-level argument.

Connections: Yoneda underlies the uniqueness of objects satisfying a universal property (a representing object is unique up to unique isomorphism), the theory of adjunctions (Hom(Fa, b) ≅ Hom(a, Gb) naturally), and density: every presheaf is a colimit of representables. It is the categorical shadow of Cayley's theorem — every group embeds into a symmetric group by acting on itself.

Applications and significance

Yoneda is the workhorse lemma of modern structural mathematics. A few payoffs:

• Universal properties are rigorous. Products, limits, tensor products, free objects, and quotients are defined by representing a functor; Yoneda guarantees the representing object is unique up to canonical isomorphism, so "the" product is well-defined.
• Proof by representables. To show two constructions agree, it suffices to show they induce the same functor of points Hom(−, X). This is the daily technique in algebraic geometry: a scheme is studied through its functor of points, and moduli problems are phrased as representability questions.
• Adjunctions and the calculus of Kan extensions are built on the embedding being full and faithful.
• Presheaf topoi. [Cᵒᵖ, Set] is the free cocompletion of C; the Yoneda embedding is the universal way to add colimits.

Philosophically, Yoneda formalizes structuralism: identity is relational. That single, one-page proof is why category theorists can insist you never need to know what an object is, only what it does.

Covariant vs. contravariant Yoneda, and the embedding it yields

Version	Probe functor	Statement	Embedding consequence
Covariant Yoneda	hᶜ = Hom(c, −): C → Set	Nat(Hom(c,−), F) ≅ F(c)	c ↦ Hom(c,−) gives Cᵒᵖ ↪ [C, Set], full & faithful
Contravariant Yoneda	h_c = Hom(−, c): Cᵒᵖ → Set	Nat(Hom(−,c), G) ≅ G(c)	c ↦ Hom(−,c) gives C ↪ [Cᵒᵖ, Set], the presheaf embedding
Specialize F = Hom(d,−)	hᶜ probing hᵈ	Nat(Hom(c,−), Hom(d,−)) ≅ Hom(d,c)	Morphisms d→c ↔ natural transformations; fullness & faithfulness
Naturality corollary	vary c and F	Bijection commutes with C- and Set-morphisms	Yoneda is an isomorphism of functors, not just of sets