Yoneda, representability, and density — the depth

Intuition Beginner

An object is fixed in place by the web of arrows tying it to every other object. Two objects that relate to everything else in exactly the same way are the same object, renamed. This slogan, that an object is determined completely by its relationships to all others, is the single idea on which the whole chapter rests. This third unit gathers its consequences into one place and pushes them further.

The two earlier units built the machinery in pieces. The first defined a representable functor as the rule recording arrows into a fixed object, and a universal element as the identity arrow seen as a seed. The second proved the Yoneda lemma, the precise statement that this seed names every comparison, and deduced that the Yoneda embedding loses no information. Here we weld those results together and watch one fact do all the work.

Two deeper outcomes are the goal. First, the same lemma that pins an object down also lets us rebuild any measurement on objects out of the address books of single objects glued together, a statement called density. Second, the whole story survives unchanged when the sets of arrows are replaced by richer structures such as spaces or categories, the enriched and two-dimensional forms. Both follow the original lemma by direct argument.

Visual Beginner

Picture the address book of one fixed object: a one-element loop at the center, the identity, and around it a bundle of arrows arriving from every other object. The center is the seed; the bundles are its portrait. Density says every measurement, however intricate, is a patchwork of such portraits stitched along the arrows of the category.

direction	what it says	where it lands in this unit
read the book inward	the seed names every comparison	the Yoneda lemma, full faithfulness
read the book outward	every measurement is glued portraits	density and the co-Yoneda form
replace sets by spaces or categories	the argument is unchanged	enriched and 2-categorical Yoneda

The picture unifies the chapter. Reading the address book inward gives the embedding, full and faithful. Reading it outward gives density, the rebuilding of any measurement from portraits. The seed at the center, the identity arrow, governs both directions, and the same picture still reads correctly when the bundles become richer objects.

Worked example Beginner

Work in the walking-arrow category $2$. It has two objects, written $0$ and $1$, and exactly one arrow $a:0\to 1$ besides the two do-nothing identity arrows. Composition is forced: any arrow followed by an identity is itself.

Step 1. Build the address book of object $1$, the rule recording arrows into $1$. At object $0$ there is one arrow into $1$, namely $a$. At object $1$ there is one arrow into $1$, the identity on $1$. So the address book holds a one-element set at each object.

Step 2. Track how the arrow $a:0\to 1$ reshuffles the book. Pre-composing with $a$ turns an arrow into $1$ at object $1$ into an arrow into $1$ at object $0$: the identity on $1$ is sent to $a$. This is naturality at work, the arrow $a$ inducing the function compose-with-$a$ between the two one-element sets.

Step 3. Apply the slogan. A comparison from this address book to any other measurement is fixed by where it sends the identity on $1$. If the measurement gives a four-element set at object $1$, there are four comparisons, one per element. The address book is so rigid that one choice names a whole comparison.

What this tells us: in a category with just two objects and one arrow, the address book of $1$ already determines every comparison out of it. The same rigidity, scaled up, is what makes the embedding full and faithful and what makes density hold for every category.

Check your understanding Beginner

Formal definition Intermediate+

Throughout $\mathcal{C}$ is a locally small category, $\hat{\mathcal{C}}=[{\mathcal{C}}^{\mathrm{op}},\mathbf{Set}]$ is its presheaf category, and $よ:\mathcal{C}\to \hat{\mathcal{C}}$ is the Yoneda embedding sending $c$ to the contravariant hom-functor $よc=\mathcal{C}(-,c)$ and $f:c\to c^{\prime }$ to post-composition $\mathcal{C}(-,f)$, in the notation of 41.04.02. A presheaf is representable when it is naturally isomorphic to some $よc$; the covariant hom-functor $\mathcal{C}(c,-)$ is the representable on ${\mathcal{C}}^{\mathrm{op}}$. These definitions are recalled from 41.04.01, where a representation $\mathcal{C}(c,-)\cong F$ was shown to be the same data as a universal element $u={\Phi }_c({\mathrm{id}}_c)\in F(c)$, and as an initial object of the category of elements ${\int }_{\mathcal{C}}F$ ^{[Riehl 2016]}.

The two foundational facts, proved in 41.04.02, are restated here as the load-bearing input. The Yoneda lemma is the bijection

$${\Phi }_{F,c}:\mathrm{Nat}(よc,F)\stackrel{\cong }{\to }F(c),\alpha ⟼{\alpha }_c({\mathrm{id}}_c),$$

natural in the presheaf $F$ and the object $c$, with inverse $u\mapsto \Psi (u)$, $\Psi (u{)}_d(f)=F(f)(u)$. Specialising $F=よd$ yields $\mathrm{Nat}(よc,よd)\cong \mathcal{C}(c,d)$, which is the statement that $よ$ is full and faithful. These two facts are the sole ingredients of every theorem below, and the present unit assembles their deeper consequences. The notation $よ$ for the embedding, $\hat{\mathcal{C}}$ for the presheaf category, ${\int }_{\mathcal{C}}F$ for the category of elements, ${\int }^c$ for a coend, and $\mathcal{V}$ for a symmetric monoidal closed base is recorded in _meta/NOTATION.md.

Definition (density). The Yoneda embedding is dense (or adequate) when every presheaf $F$ is the canonical colimit of the composite

$${\int }_{\mathcal{C}}F\stackrel{\pi }{\to }\mathcal{C}\stackrel{よ}{\to }\hat{\mathcal{C}},$$

with colimiting cocone ${\lambda }_{(c,x)}=\Psi (x):よc\to F$ at $(c,x)\in {\int }_{\mathcal{C}}F$. Equivalently, by the standard rewriting of a colimit over a category of elements as a coend, $F\cong {\int }^{c\in \mathcal{C}}F(c)\cdot よc$, where $S\cdot X$ denotes the $S$-indexed copower (coproduct of $S$ copies of $X$); this coend form is the co-Yoneda lemma or ninja-Yoneda identity ^{[Mac Lane 1998]}. Density is the outward-reading half of the Yoneda lemma: where full faithfulness reads the address book inward, density reads it outward, rebuilding an arbitrary $F$ from the representables it contains.

Definition (enriched setting). Let $\mathcal{V}=({\mathcal{V}}_0,\otimes ,I,⊸)$ be a symmetric monoidal closed category, so that the internal hom $X⊸(-)$ is right adjoint to $X\otimes (-)$. A $\mathcal{V}$-category $\mathcal{C}$ has hom-objects $\mathcal{C}(c,d)\in {\mathcal{V}}_0$ in place of hom-sets, together with composition and unit morphisms in ${\mathcal{V}}_0$ satisfying the usual associativity and unit axioms. The enriched presheaf category $[{\mathcal{C}}^{\mathrm{op}},\mathcal{V}]$ has $\mathcal{V}$-functors ${\mathcal{C}}^{\mathrm{op}}\to \mathcal{V}$ as objects and $\mathcal{V}$-natural transformations as morphisms, and the enriched Yoneda embedding $\mathbf{Y}:\mathcal{C}\to [{\mathcal{C}}^{\mathrm{op}},\mathcal{V}]$ sends $c$ to $\mathcal{C}(-,c)$ regarded as a $\mathcal{V}$-functor. Taking $\mathcal{V}=\mathbf{Set}$ recovers the ordinary definitions; taking $\mathcal{V}=\mathbf{Cat}$ gives the 2-categorical setting, and $\mathcal{V}=\mathbf{Ab}$ the additive one ^{[Kelly 1982]}.

Counterexamples to common slips Intermediate+

• Density is a colimit, not a limit, of representables. Every presheaf is a colimit of representables indexed by its own category of elements; the dual statement with limits fails. A coproduct $よc⨿よd$ of two representables is already a presheaf that is not a finite limit of representables, so reversing the variance breaks the theorem.

• Full and faithful does not mean essentially surjective. The Yoneda embedding lands inside $\hat{\mathcal{C}}$ on the representables, a tiny full subcategory; most presheaves are not representable. Density repairs this asymmetry by exhibiting the non-representables as colimits of representables, not as representables themselves.

• The universal element is the Yoneda image of the identity, not an arbitrary choice. Re-reading ${\Phi }_{F,c}(\alpha )={\alpha }_c({\mathrm{id}}_c)$ when $\alpha$ is invertible gives the universal element of 41.04.01; the two units describe one datum in two vocabularies, and conflating "the representing object is unique" with "the representation is unique on the nose" forgets that the isomorphism is itself part of the data.

Key theorem with proof Intermediate+

The signature result is that full faithfulness of the embedding, the density theorem, and the co-Yoneda identity all flow from the single Yoneda bijection, so the three theorems of the chapter are one fact read in three directions.

Theorem (the Yoneda cycle). Let $\mathcal{C}$ be locally small. From the Yoneda lemma, namely the natural bijection $\mathrm{Nat}(よc,F)\cong F(c)$, $\alpha \mapsto {\alpha }_c({\mathrm{id}}_c)$, there follow both: (a) the embedding $よ:\mathcal{C}\to \hat{\mathcal{C}}$ is full and faithful; (b) density, that every presheaf $F$ is the colimit ${\mathrm{colim}}_{(c,x)\in {\int }_{\mathcal{C}}F}よc$, equivalently $F\cong {\int }^{c}F(c)\cdot よc$.

Conversely (b) specialises at a representable to give back (a), and (a) together with the existence of the Yoneda inverse $\Psi$ recovers the lemma. So the three are equivalent formulations of one piece of structure ^{[Mac Lane 1998]}.

Proof. (Lemma $\Rightarrow$ (a).) Take $F=よd$ in the Yoneda bijection. Then $\mathrm{Nat}(よc,よd)\cong (よd)(c)=\mathcal{C}(c,d)$, and the inverse $\Psi$ sends $f:c\to d$ to the natural transformation with $e$-component $g\mapsto (よd)(f)(g)=f\circ g$, that is, to $よf$. Hence the function $\mathcal{C}(c,d)\to \mathrm{Nat}(よc,よd)$, $f\mapsto よf$, is a bijection, which is precisely full faithfulness of $よ$.

((a) $\Rightarrow$ (b) is not the route taken; the lemma gives (b) directly.) Define the cocone ${\lambda }_{(c,x)}=\Psi (x):よc\to F$, where $\Psi$ is the Yoneda inverse and $x\in F(c)$. For a morphism $f:(c,x)\to (c^{\prime },x^{\prime })$ in ${\int }_{\mathcal{C}}F$, that is $f:c\to c^{\prime }$ with $F(f)(x^{\prime })=x$, the cocone condition ${\lambda }_{(c,x)}={\lambda }_{(c^{\prime },x^{\prime })}\circ よf$ holds because both sides are transformations $よc\to F$ with the same Yoneda image: evaluating at ${\mathrm{id}}_c$ gives $({\lambda }_{(c^{\prime },x^{\prime })}\circ よf{)}_c({\mathrm{id}}_c)={\lambda }_{(c^{\prime },x^{\prime })c}(f)=F(f)(x^{\prime })=x={\lambda }_{(c,x)c}({\mathrm{id}}_c)$.

To see the cocone is colimiting, test it against an arbitrary presheaf $G$ and cocone ${\mu }_{(c,x)}:よc\to G$. By Yoneda each ${\mu }_{(c,x)}$ is the element $m_{(c,x)}={\mu }_{(c,x)c}({\mathrm{id}}_c)\in G(c)$, and the cocone condition on $\mu$ translates to $G(f)(m_{(c^{\prime },x^{\prime })})=m_{(c,x)}$ whenever $F(f)(x^{\prime })=x$. Define $\theta :F\Rightarrow G$ by ${\theta }_c(x)=m_{(c,x)}$; the displayed condition is exactly naturality of $\theta$, and $\theta \circ {\lambda }_{(c,x)}={\mu }_{(c,x)}$ by evaluating at ${\mathrm{id}}_c$. Any mediating $\theta$ is forced, since ${\theta }_c(x)={\theta }_c({\lambda }_{(c,x)c}({\mathrm{id}}_c))={\mu }_{(c,x)c}({\mathrm{id}}_c)=m_{(c,x)}$. So $F\cong {\mathrm{colim}}_{(c,x)}よc$, and rewriting the colimit over ${\int }_{\mathcal{C}}F$ as a coend gives $F\cong {\int }^{c}F(c)\cdot よc$.

((b) $\Rightarrow$ (a).) Specialise density to the representable $F=よd$. The category of elements ${\int }_{\mathcal{C}}よd$ has a terminal object $(d,{\mathrm{id}}_d)$, so the colimit collapses: $よd\cong よd$ with the colimiting cocone given by post-composition. Reading the cocone at this terminal object recovers the bijection $\mathrm{Nat}(よc,よd)\cong \mathcal{C}(c,d)$, which is (a). The equivalence of the three formulations is the Yoneda cycle. $\square$

Bridge. The cycle builds toward the entire colimit-based viewpoint of category theory and appears again in 41.06.02, where the free-cocompletion universal property $F\mapsto {\int }^{c}F(c)\cdot H(c)$ is recognised as the pointwise left Kan extension ${\mathrm{Lan}}_{よ}H$, and in 41.05.02, where Lawvere theories and monadic adjunctions are built from presheaf-style free constructions. The foundational reason one lemma does all this work is that the Yoneda inverse $\Psi (u{)}_d(f)=F(f)(u)$ reconstructs every component from a single value $u$, so a comparison and a value are the same thing; this is exactly the mechanism that makes the address book rigid enough to be full and faithful and rich enough to be dense, and putting these together, the bridge is that full faithfulness and density are the inward and outward readings of one bijection, so representability in 41.04.01, the lemma in 41.04.02, and the constructions of 41.03.01 all reduce to evaluating a natural transformation at an identity arrow.

Exercises Intermediate+

Advanced results Master

Theorem (the free cocompletion). For any locally small $\mathcal{C}$ the presheaf category $\hat{\mathcal{C}}=[{\mathcal{C}}^{\mathrm{op}},\mathbf{Set}]$ is the free colimit-completion of $\mathcal{C}$: it is cocomplete, and for every cocomplete category $\mathcal{E}$, restriction along $よ$ is an equivalence between cocontinuous functors $\hat{\mathcal{C}}\to \mathcal{E}$ and arbitrary functors $\mathcal{C}\to \mathcal{E}$ ^{[Mac Lane-Moerdijk 1992]}. The equivalence sends $H:\mathcal{C}\to \mathcal{E}$ to $\tilde{H}={\mathrm{Lan}}_{よ}H$ with $\tilde{H}(F)={\int }^{c}F(c)\cdot H(c)$, the density coend with $よc$ replaced by $H(c)$; on a representable the category of elements has a terminal object, so $\tilde{H}(よc)\cong H(c)$ and the unit of the universal property holds. Density is thus the local statement of which free cocompletion is the global one: every presheaf is rebuilt from representables, so a colimit-preserving functor is fixed by where it sends the representables, and conversely any choice on representables extends uniquely. A full proof is Proposition 2 below.

Theorem (the enriched Yoneda lemma). Let $\mathcal{V}$ be a symmetric monoidal closed category and $\mathcal{C}$ a $\mathcal{V}$-category. For every $\mathcal{V}$-functor $F:{\mathcal{C}}^{\mathrm{op}}\to \mathcal{V}$ and every object $c$ there is a $\mathcal{V}$-natural isomorphism

$$[{\mathcal{C}}^{\mathrm{op}},\mathcal{V}](\mathcal{C}(-,c),F)\cong F(c)$$

in ${\mathcal{V}}_0$, the enriched Yoneda lemma ^{[Kelly 1982]}. The bijection on underlying elements (maps $I\to F(c)$) is the ordinary one: a $\mathcal{V}$-natural transformation is determined by the image of ${\mathrm{id}}_c:I\to \mathcal{C}(c,c)$. The enriched embedding $\mathbf{Y}:\mathcal{C}\to [{\mathcal{C}}^{\mathrm{op}},\mathcal{V}]$ is full and faithful in the enriched sense, and enriched density holds: every $\mathcal{V}$-presheaf is a weighted colimit of enriched representables. Taking $\mathcal{V}=\mathbf{Set}$ recovers the entire chapter verbatim, so the ordinary theory is the $\mathbf{Set}$-enriched case of a uniform statement; taking $\mathcal{V}=\mathbf{Cat}$ yields the 2-categorical Yoneda below.

Theorem (the 2-categorical and bicategorical Yoneda). For a 2-category $\mathcal{K}$ and a 2-functor $F:{\mathcal{K}}^{\mathrm{op}}\to \mathbf{Cat}$, the category of 2-natural transformations $\mathcal{K}(-,A)\Rightarrow F$ is isomorphic to $F(A)$: this is the 2-Yoneda lemma, the $\mathcal{V}=\mathbf{Cat}$ instance of the enriched lemma. At the level of bicategories, Street's bicategorical Yoneda states that the bicategory of pseudofunctors ${\mathcal{B}}^{\mathrm{op}}\to \mathbf{Cat}$, pseudonatural transformations, and modifications is locally equivalent to $\mathcal{B}$ on the representables, and every pseudofunctor is a bicategorical colimit of representables ^{[Street 1980]}. This bicategorical form is what identifies fibrations with pseudofunctors through the Grothendieck construction and underlies the indexed reading of the category of elements; the ordinary Yoneda lemma is its truncation to the 1-categorical case.

Theorem (universal properties as natural isomorphisms of hom-functors). A universal property in the sense of 41.04.01 is equivalently an isomorphism in $\hat{\mathcal{C}}$ between a presheaf and a representable. Concretely, a limit of $D:\mathcal{J}\to \mathcal{C}$ is a representing object for the cone functor $C\mapsto \mathrm{Cone}(C,D)$, that is, an isomorphism $よ(\lim D)\cong \mathrm{Cone}(-,D)$; dually a colimit is a representation of the covariant cocone functor on ${\mathcal{C}}^{\mathrm{op}}$. Free constructions are the covariant side: the free group on a set $S$ is a representing object for $G\mapsto \mathbf{Set}(S,U(G))$, equivalently a natural isomorphism $\mathbf{Grp}(F(S),-)\cong \mathbf{Set}(S,U(-))$, which is the hom-set adjunction of 41.03.01. Thus every universal property is the assertion that some presheaf is representable, and the Yoneda lemma is what turns that assertion into a computable bijection.

Theorem (free constructions via left Kan extensions). Density makes every free construction a left Kan extension. The free cocompletion is ${\mathrm{Lan}}_{よ}{\mathrm{Id}}_{\mathcal{C}}$; the free finite-limit completion is the same construction on ${\mathcal{C}}^{\mathrm{op}}$ followed by finite-limit closure; the free monoid on an object of a monoidal category, the free $R$-module on a set, and the free group are each $\mathrm{Lan}$ of an inclusion along a Yoneda-type embedding, computed by the coend ${\int }^{c}F(c)\cdot H(c)$ of 41.06.02. This is why the left Kan extension formulas of 41.06.02 subsume the constructions of 41.05.02: a left adjoint is a left Kan extension of the identity along its right adjoint, and a free algebra functor is a left Kan extension of the generators along the forgetful functor. The co-Yoneda identity is the special case $H=よ$ in which the extension reproduces its input.

Synthesis. Putting these together, the Yoneda lemma is the single fact from which full faithfulness, density, the free cocompletion, the enriched lemma, and the bicategorical lemma all descend. The foundational reason one bijection does all of this is that the inverse $\Psi (u{)}_d(f)=F(f)(u)$ reconstructs an entire transformation from one value, so the address book is simultaneously rigid enough to be full and faithful and rich enough to be dense, and this is exactly the content of the Yoneda cycle proved above. Reading the bijection in $\mathcal{V}$-enriched form generalises it to spaces, categories, and abelian groups without change of argument, so the $\mathbf{Set}$-enriched theory of 41.04.01 and 41.04.02 is the base case of a uniform statement; reading it two-dimensionally gives Street's bicategorical Yoneda, which is dual throughout to the covariant, co-Yoneda side. The central insight is that density makes $\hat{\mathcal{C}}$ the free cocompletion, so every cocontinuous functor is a left Kan extension of its restriction to representables and the bridge is that adjunctions 41.03.01, monads 41.05.01, and Kan extensions 41.06.02 are each an instance of extending along $よ$, which is why all of category theory reduces to evaluating a natural transformation at an identity arrow.

Full proof set Master

Proposition 1 (co-Yoneda identity, full proof). For every presheaf $F\in \hat{\mathcal{C}}$ the canonical map ${\int }^{c}F(c)\cdot よc\to F$ with $e$-component sending $(x,f:e\to c)\mapsto F(f)(x)$ is a natural isomorphism.

Proof. The candidate $\theta :{\int }^{c}F(c)\cdot よc\to F$ is well-defined on the coend quotient because the two identifications agree: for $g:c\to c^{\prime }$, an element $(x\in F(c),f:e\to c)$ identified with $(F(g)(x),g\circ f)$ on one side and $(x,f)$ traced through $よ$ on the other both map to $F(f)(x)=F(g\circ f)(F(g^{-1}\text{-style}))$; concretely $F(f)(x)$ is invariant since $F(g\circ f)(-)$ agrees with $F(f)$ after the $F(g)$-quotient, which is the coend relation. Naturality in $e$ holds because $F$ is a functor. Define the inverse ${\iota }_e:F(e)\to {\int }^{c}F(c)\cdot よc$ by ${\iota }_e(x)=(x,{\mathrm{id}}_e)\in F(e)\cdot よe$. Then ${\theta }_e\circ {\iota }_e(x)=F({\mathrm{id}}_e)(x)=x$, so $\theta \circ \iota =\mathrm{id}$. For the other composite, an element $(x,f:e\to c)$ in the $F(c)\cdot よc$-summand is equal in the coend to $(F(f)(x),{\mathrm{id}}_e)\in F(e)\cdot よe$ via the morphism $f:(e,F(f)(x))\to (c,x)$ in ${\int }_{\mathcal{C}}F$, so ${\iota }_e\circ \theta$ fixes every representative, hence $\iota \circ \theta =\mathrm{id}$. Therefore $\theta$ is a natural isomorphism and $F\cong {\int }^{c}F(c)\cdot よc$. $\square$

Proposition 2 (free cocompletion universal property, full proof). Let $\mathcal{E}$ be cocomplete and $H:\mathcal{C}\to \mathcal{E}$ a functor. Then $\tilde{H}(F)={\mathrm{colim}}_{(c,x)\in {\int }_{\mathcal{C}}F}H(c)$ defines a cocontinuous functor $\tilde{H}={\mathrm{Lan}}_{よ}H:\hat{\mathcal{C}}\to \mathcal{E}$ with $\tilde{H}\circ よ\cong H$, and any cocontinuous $K:\hat{\mathcal{C}}\to \mathcal{E}$ with $K\circ よ\cong H$ satisfies $K\cong \tilde{H}$.

Proof. The assignment is functorial: a natural transformation $\beta :F\Rightarrow G$ induces a map ${\int }_{\mathcal{C}}F\to {\int }_{\mathcal{C}}G$ sending $(c,x)\mapsto (c,{\beta }_c(x))$, hence a map between the colimits defining $\tilde{H}(F)$ and $\tilde{H}(G)$, and this preserves identities and composition by the functoriality of colimits. For a representable, ${\int }_{\mathcal{C}}よc$ has terminal object $(c,{\mathrm{id}}_c)$, so the colimit collapses: $\tilde{H}(よc)\cong H(c)$, giving the unit isomorphism $\tilde{H}\circ よ\cong H$. Cocontinuity: a colimit of presheaves is computed pointwise, and re-indexing the categories of elements shows ${\int }_{\mathcal{C}}({\mathrm{colim}}_i{F}_i)$ is the appropriate colimit of the ${\int }_{\mathcal{C}}{F}_i$; since colimits commute with colimits in $\mathcal{E}$, $\tilde{H}({\mathrm{colim}}_i{F}_i)\cong {\mathrm{colim}}_i\tilde{H}(F_i)$. For uniqueness, if $K$ is cocontinuous with $K\circ よ\cong H$, then by density $K(F)\cong K({\mathrm{colim}}_{(c,x)}よc)\cong {\mathrm{colim}}_{(c,x)}K(よc)\cong {\mathrm{colim}}_{(c,x)}H(c)\cong \tilde{H}(F)$, naturally in $F$, so $K\cong \tilde{H}$. Hence restriction along $よ$ is an equivalence between cocontinuous functors $\hat{\mathcal{C}}\to \mathcal{E}$ and functors $\mathcal{C}\to \mathcal{E}$. $\square$

Proposition 3 (the Yoneda embedding preserves limits and reflects isomorphisms). The functor $よ:\mathcal{C}\to \hat{\mathcal{C}}$ preserves every limit that exists in $\mathcal{C}$, and reflects isomorphisms: if $よf$ is an isomorphism of presheaves then $f$ is an isomorphism in $\mathcal{C}$.

Proof. For limits, let $L={\lim }_j{D}_j$. Since limits in $\hat{\mathcal{C}}$ are pointwise, $({\lim }_jよD_j)(d)={\lim }_j\mathcal{C}(d,D_j)$, and $\mathcal{C}(d,-)$ preserves limits because a cone over $D$ with apex $d$ is a compatible family $d\to D_j$ uniquely factoring through $L$, giving $\mathcal{C}(d,L)\cong {\lim }_j\mathcal{C}(d,D_j)=({\lim }_jよD_j)(d)=(よL)(d)$ naturally in $d$. So $よL\cong {\lim }_jよD_j$. For reflection, suppose $よf:よc\to よc^{\prime }$ is an isomorphism. Since $よ$ is full, the inverse $(よf{)}^{-1}=よg$ for some $g:c^{\prime }\to c$. Then $よ(g\circ f)=よg\circ よf={\mathrm{id}}_{よc}=よ({\mathrm{id}}_c)$, and faithfulness gives $g\circ f={\mathrm{id}}_c$; symmetrically $f\circ g={\mathrm{id}}_{c^{\prime }}$. Hence $f$ is an isomorphism with inverse $g$. A full and faithful functor always reflects isomorphisms by this argument. $\square$

Proposition 4 (the enriched Yoneda bijection). For a symmetric monoidal closed $\mathcal{V}$, a $\mathcal{V}$-category $\mathcal{C}$, a $\mathcal{V}$-functor $F:{\mathcal{C}}^{\mathrm{op}}\to \mathcal{V}$, and $c\in \mathcal{C}$, the map on underlying sets ${\mathcal{V}}_0(I,[{\mathcal{C}}^{\mathrm{op}},\mathcal{V}](\mathcal{C}(-,c),F))\to {\mathcal{V}}_0(I,F(c))$ sending a $\mathcal{V}$-natural transformation $\alpha$ to $I\stackrel{{\mathrm{id}}_c}{\to }\mathcal{C}(c,c)\stackrel{{\alpha }_c}{\to }F(c)$ is a bijection, natural in $c$ and $F$.

Proof. A $\mathcal{V}$-natural transformation $\alpha :\mathcal{C}(-,c)\Rightarrow F$ is, in particular, a family of ${\mathcal{V}}_0$-morphisms ${\alpha }_d:\mathcal{C}(d,c)\to F(d)$ satisfying the $\mathcal{V}$-naturality condition that for all $d,e$ a certain square built from composition and $F$ commutes. The composite $u_{\alpha }={\alpha }_c\circ {\mathrm{id}}_c:I\to F(c)$ records the image of the identity. Conversely, given $u:I\to F(c)$, define ${\alpha }_d^u:\mathcal{C}(d,c)\to F(d)$ as the composite $\mathcal{C}(d,c)\cong I\otimes \mathcal{C}(d,c)\stackrel{u\otimes \mathrm{id}}{\to }F(c)\otimes \mathcal{C}(d,c)\stackrel{\mathrm{comp}}{\to }F(d)$, where the last map is the $F$-action on the hom-object, the enriched analogue of $f\mapsto F(f)(u)$. The $\mathcal{V}$-naturality of ${\alpha }^u$ follows from the associativity of the $F$-action and the composition coherence of $\mathcal{C}$. The two constructions are mutually inverse: ${\alpha }_d^{u_{\alpha }}$ evaluated via the unit recovers ${\alpha }_d$ by the enriched naturality of $\alpha$ at the identity, and $u_{{\alpha }^u}={\alpha }_c^u\circ {\mathrm{id}}_c=u$ by the unit coherence. Hence the map on underlying elements is a bijection, and the $\mathcal{V}$-naturality in $c$ and $F$ is the enriched analogue of the ordinary naturality square, transported along the closed structure of $\mathcal{V}$. For $\mathcal{V}=\mathbf{Set}$ with $I=1$ this reduces verbatim to the ordinary Yoneda lemma. $\square$

Connections Master

• The Yoneda lemma and the embedding 41.04.02. The sibling unit states and proves the Yoneda lemma and the full faithfulness of $よ$ as its central results; the present unit takes those as input and establishes the Yoneda cycle — that the lemma, full faithfulness, density, and the co-Yoneda identity are one fact in four directions — and pushes into the enriched, 2-categorical, and free-cocompletion consequences. Propositions 1-2 below are the deeper forms of the density and free-cocompletion statements sketched in 41.04.02.

• Representable functors and universal elements 41.04.01. The representability theorem of 41.04.01 identifies a representation, a universal element, and a terminal object of the category of elements. The present unit reads that theorem through the Yoneda embedding: a universal property is the assertion that a presheaf is representable, and the cycle proved here is what turns that assertion into the computable bijection $\mathrm{Nat}(よc,F)\cong F(c)$.

• Limits and universal cones 41.02.01. A limit is a representing object for the cone functor, equivalently an isomorphism $よ(\lim D)\cong \mathrm{Cone}(-,D)$ in $\hat{\mathcal{C}}$, and Proposition 3 shows $よ$ preserves the limits of 41.02.01 because each $\mathcal{C}(d,-)$ does. Density is the complementary theorem that $よ$ does not preserve colimits but instead freely adjoins them, making $\hat{\mathcal{C}}$ the universal home of formal colimits over the limits of 41.02.01.

• Adjunctions as representability 41.03.01, 41.03.03. A left adjoint to $U$ is a choice of representation of $\mathcal{C}(c,U(-))$ for every $c$, with the universal elements assembling into the unit of the adjunction; the unit-counit calculus of 41.03.03 is the two-dimensional shadow of the Yoneda bijection applied componentwise. The free-construction theorem above restates this as: a left adjoint is a left Kan extension of the identity along its right adjoint, computed by the density coend.

• Monads and algebraic free constructions 41.05.01, 41.05.02. Every adjunction $F⊣G$ generates the monad $GF$ of 41.05.01, and the free-algebra functor of 41.05.02 is the left Kan extension of the generators along the forgetful functor, the same coend formula ${\int }^{c}F(c)\cdot H(c)$ that gives free cocompletion. Lawvere theories are presheaf-style free constructions on a skeleton, so the density theorem governs their completion to categories of algebras.

• Ends, coends, and Kan extensions 41.06.01, 41.06.02. The co-Yoneda identity $F\cong {\int }^{c}F(c)\cdot よc$ is given its full calculus in 41.06.01, where the Fubini and continuity rules make the formula computational, and the free-cocompletion extension ${\mathrm{Lan}}_{よ}H$ is the leading example of the pointwise Kan extensions of 41.06.02. The Yoneda cycle proved here is the lemma from which those general formulas descend by replacing $よc$ with an arbitrary weight.

• The functor of points in algebraic geometry 04.02.01. The geometric unit 04.02.01 develops the reading under which a scheme is its representable presheaf $\mathbf{Sch}(-,S)$ and moduli problems are functors to be proved representable; the density theorem here is what allows arbitrary presheaves to be handled as colimits of representables even when not themselves representable, the entry point to algebraic stacks where representability fails.

Historical & philosophical context Master

The bijection at the heart of this unit is attributed to Nobuo Yoneda from a 1954 conversation at the Gare du Nord in Paris, reported by Saunders Mac Lane, and entered the literature through Yoneda's paper on the homology theory of modules ^{[Yoneda 1954]}. Mac Lane's Categories for the Working Mathematician (1971; 2nd ed. 1998) fixed the lemma, the embedding, and the observation that an object is determined up to isomorphism by the functor it represents as the central results of Chapter III ^{[Mac Lane 1998]}; the coend form, often called the co-Yoneda or ninja-Yoneda identity, is developed there in Chapter X alongside the ends-and-coends calculus that makes density computational.

The density theorem and the free-cocompletion reading of presheaves were systematised in topos theory: Mac Lane and Moerdijk's Sheaves in Geometry and Logic (1992) presents every presheaf as a colimit of representables in its first chapter and uses it to build the sheafification adjunction ^{[Mac Lane-Moerdijk 1992]}. The enriched and weighted-colimit forms are due to Max Kelly, whose Basic Concepts of Enriched Category Theory (1982) gives the $\mathcal{V}$-enriched Yoneda lemma and the density theorem for $\mathcal{V}$-presheaves as the uniform backbone of enriched theory ^{[Kelly 1982]}. The bicategorical refinement — the Yoneda lemma for bicategories, under which every pseudofunctor is a bicategorical colimit of representables — is due to Ross Street, appearing in his work on fibrations in bicategories ^{[Street 1980]}, and is the formal setting that identifies fibrations with pseudofunctors. Grothendieck's functor-of-points philosophy, in which a geometric object is the functor it represents on all test objects ^{[Grothendieck 1960]}, is the geometric avatar of the same slogan: the address book of a scheme is its complete portrait, and density is what licenses the passage from representable functors to arbitrary moduli functors and on to stacks.

Bibliography Master

@article{Yoneda1954depth,
  author  = {Yoneda, Nobuo},
  title   = {On the homology theory of modules},
  journal = {Journal of the Faculty of Science, University of Tokyo, Section I},
  volume  = {7},
  year    = {1954},
  pages   = {193--227}
}

@book{MacLane1998depth,
  author    = {Mac Lane, Saunders},
  title     = {Categories for the Working Mathematician},
  edition   = {2},
  publisher = {Springer},
  series    = {Graduate Texts in Mathematics 5},
  year      = {1998}
}

@book{Riehl2016depth,
  author    = {Riehl, Emily},
  title     = {Category Theory in Context},
  publisher = {Dover},
  year      = {2016}
}

@book{Kelly1982depth,
  author    = {Kelly, Gregory M.},
  title     = {Basic Concepts of Enriched Category Theory},
  publisher = {Cambridge University Press},
  series    = {London Mathematical Society Lecture Note Series 64},
  year      = {1982},
  note      = {Reprinted as Theory and Applications of Categories, No. 10, 2005}
}

@book{MacLaneMoerdijk1992depth,
  author    = {Mac Lane, Saunders and Moerdijk, Ieke},
  title     = {Sheaves in Geometry and Logic: A First Introduction to Topos Theory},
  publisher = {Springer},
  series    = {Universitext},
  year      = {1992}
}

@incollection{Street1980depth,
  author    = {Street, Ross},
  title     = {Fibrations in bicategories},
  booktitle = {Cahiers de topologie et géométrie différentielle catégoriques},
  volume    = {21},
  year      = {1980},
  pages     = {111--160}
}

@incollection{Grothendieck1961depth,
  author    = {Grothendieck, Alexander},
  title     = {Techniques de construction et théorèmes d'existence en géométrie algébrique IV: Les schémas de Hilbert},
  booktitle = {Séminaire Bourbaki},
  volume    = {6, Exp. 221},
  publisher = {Société Mathématique de France},
  year      = {1961},
  pages     = {249--276}
}

@book{Awodey2010depth,
  author    = {Awodey, Steve},
  title     = {Category Theory},
  edition   = {2},
  publisher = {Oxford University Press},
  series    = {Oxford Logic Guides 52},
  year      = {2010}
}