Natural Transformations, Functor Categories, and Equivalence of Categories

Intuition Beginner

A functor translates one category into another: it carries objects to objects and arrows to arrows. Once you have two such translations running side by side, a new question appears. Suppose two functors both send a category into the same target. Is there a way to slide from the output of the first translation to the output of the second, object by object, that respects all the arrows? That coordinated sliding is a natural transformation. It is a single rule that, at every object at once, nudges the first functor's answer over to the second functor's answer.

The word that matters is "coordinated". You are not allowed to make a separate, unrelated choice at each object. The choices must fit together so that following an arrow and then sliding gives the same result as sliding and then following the arrow. When that compatibility holds for every arrow, the family of slides is called natural. The slogan from the founders of the subject was that a construction is natural when it does not depend on arbitrary choices — when it works the same way everywhere.

Here is the everyday picture. A finite list can be reversed. Reversing a list of numbers, then adding one to each, gives the same outcome as adding one to each and then reversing. The "add one" step is an arrow; "reverse" is the family of slides. Because the order does not matter, reversal is natural. That order-does-not-matter property, drawn as a square that closes up, is the whole idea.

Visual Beginner

Picture two parallel filmstrips. The top strip is what functor $F$ produces; the bottom strip is what functor $G$ produces. For each object you draw a vertical connector — an arrow pointing from the top frame down to the bottom frame. That vertical connector is one component of the natural transformation.

Now follow any horizontal arrow on the top strip from one frame to the next. There is a matching horizontal arrow on the bottom strip. Going right-then-down must equal going down-then-right. Every little rectangle you can form this way closes up: the two paths around it agree.

feature	a functor	a natural transformation between two functors
connects	one category to another	one functor to another (same source and target)
made of	a target object and arrow for each input	one vertical arrow (component) per object
must respect	composition and identities	every naturality square must close up
reversible case	an isomorphism of categories	a natural isomorphism (every component reversible)

The picture says the vertical arrows are not independent decorations. They are tied together by the requirement that every rectangle commutes. That single requirement is what earns the family the name natural.

Worked example Beginner

Work with finite-dimensional real vector spaces and linear maps. To each space $V$ attach its double dual $V^{\ast \ast }$, the space of functions on functions on $V$. There is a standard rule that sends a vector $v$ in $V$ to the gadget "feed me a function and I hand back its value at $v$". Call this rule ${\eta }_V$, a linear map from $V$ to $V^{\ast \ast }$.

Step 1. Fix a concrete space, $V={\mathbb{R}}^2$. A vector is a pair, say $v=(3,5)$. Then ${\eta }_V(v)$ is the operation that takes any linear function $\varphi$ on ${\mathbb{R}}^2$ and returns the number $\varphi (3,5)$.

Step 2. Take a linear map to a second space, $T:{\mathbb{R}}^2\to {\mathbb{R}}^2$, say the map that swaps coordinates, $T(x,y)=(y,x)$. There are two routes from $v$ to the double dual of the target. Route A: first apply $\eta$, then push forward by $T$. Route B: first apply $T$ to get $(5,3)$, then apply $\eta$.

Step 3. Check the two routes agree on a sample function $\varphi (x,y)=2x+y$. Route B gives $\eta (5,3)$ evaluated at $\varphi$, which is $2\cdot 5+3=13$. Route A pushes $\eta (3,5)$ forward by $T$; pushing forward feeds $\varphi$ through $T$ first, giving $\varphi (T(3,5))=\varphi (5,3)=13$. The two answers match.

What this tells us: the map into the double dual is the same whether you transform first or embed first. No basis was chosen anywhere. That basis-free agreement, holding for every space and every map at once, is exactly what makes $\eta$ a natural transformation — the original example that the inventors of the subject built the definition to capture.

Check your understanding Beginner

Formal definition Intermediate+

Let $F,G:\mathcal{C}\to \mathcal{D}$ be functors. A natural transformation $\alpha :F\Rightarrow G$ assigns to each object $A$ of $\mathcal{C}$ a morphism ${\alpha }_A:F(A)\to G(A)$ in $\mathcal{D}$, its component at $A$, such that for every morphism $f:A\to B$ in $\mathcal{C}$ the naturality square

$$\begin{array}{ccc}F(A)& \stackrel{{\alpha }_A}{\to }& G(A)\\ F(f)\downarrow & & \downarrow G(f)\\ F(B)& \stackrel{{\alpha }_B}{\to }& G(B)\end{array}$$

commutes, that is $G(f)\circ {\alpha }_A={\alpha }_B\circ F(f)$ ^{[Mac Lane 1998]}. A natural transformation all of whose components are isomorphisms is a natural isomorphism, written $F\cong G$; the componentwise inverses then satisfy naturality and give an inverse transformation. Notation introduced here ($\Rightarrow$ for natural transformations, $\circ$ for composition, ${\alpha }_A$ for a component) is recorded in _meta/NOTATION.md.

The double-dual embedding ${\eta }_V:V\to V^{\ast \ast }$ is a natural transformation $\mathrm{Id}\Rightarrow (-{)}^{\ast \ast }$ on finite-dimensional vector spaces, and a natural isomorphism there; on all vector spaces it remains natural but ceases to be an isomorphism. The determinant is a natural transformation $\det :{\mathrm{GL}}_n\Rightarrow (-{)}^{\times }$ between functors from commutative rings to groups, since $\det (M)$ commutes with ring homomorphisms applied entrywise. Abelianization gives the unit ${\eta }_G:G\to G^{\mathrm{ab}}$ of the free-abelian-quotient construction, natural in $G$. The Hurewicz map $h:{\pi }_n\Rightarrow H_n$ is natural in the space.

Definition (vertical composition). Given $\alpha :F\Rightarrow G$ and $\beta :G\Rightarrow H$ between functors $\mathcal{C}\to \mathcal{D}$, their vertical composite $\beta \cdot \alpha :F\Rightarrow H$ has components $(\beta \cdot \alpha {)}_A={\beta }_A\circ {\alpha }_A$. The identity natural transformation $1_F$ has components ${\mathrm{id}}_{F(A)}$.

Definition (horizontal composition and whiskering). Given $\alpha :F\Rightarrow F^{\prime }$ for $F,F^{\prime }:\mathcal{C}\to \mathcal{D}$ and $\beta :G\Rightarrow G^{\prime }$ for $G,G^{\prime }:\mathcal{D}\to \mathcal{E}$, the horizontal composite $\beta \ast \alpha :GF\Rightarrow G^{\prime }{F}^{\prime }$ has component at $A$ given by either equal path around the square below, ${\beta }_{F^{\prime }(A)}\circ G({\alpha }_A)=G^{\prime }({\alpha }_A)\circ {\beta }_{F(A)}$. When one of the two is an identity transformation the operation is whiskering: $G\alpha :GF\Rightarrow G{F}^{\prime }$ has components $G({\alpha }_A)$, and $\beta F:GF\Rightarrow G^{\prime }F$ has components ${\beta }_{F(A)}$.

Definition (functor category). For categories $\mathcal{C}$ (small) and $\mathcal{D}$, the functor category $[\mathcal{C},\mathcal{D}]$ (also ${\mathcal{D}}^{\mathcal{C}}$ or $\mathrm{Fun}(\mathcal{C},\mathcal{D})$) has functors $\mathcal{C}\to \mathcal{D}$ as objects and natural transformations as morphisms, composed vertically; the identity on $F$ is $1_F$. An isomorphism in $[\mathcal{C},\mathcal{D}]$ is precisely a natural isomorphism. A presheaf category is the special case $[{\mathcal{C}}^{\mathrm{op}},\mathbf{Set}]$.

Definition (equivalence of categories). A functor $F:\mathcal{C}\to \mathcal{D}$ is an equivalence if there is a functor $G:\mathcal{D}\to \mathcal{C}$ (a quasi-inverse) together with natural isomorphisms $\eta :{\mathrm{Id}}_{\mathcal{C}}\cong GF$ and $\epsilon :FG\cong {\mathrm{Id}}_{\mathcal{D}}$. Equivalence is weaker than isomorphism of categories, which would demand $GF=\mathrm{Id}$ and $FG=\mathrm{Id}$ on the nose; equivalence allows them only up to natural isomorphism, so it ignores the distinction between isomorphic objects.

Definition (comma category). Given functors $F:\mathcal{A}\to \mathcal{C}$ and $G:\mathcal{B}\to \mathcal{C}$, the comma category $F\downarrow G$ has objects the triples $(A,B,f)$ with $A$ in $\mathcal{A}$, $B$ in $\mathcal{B}$, and $f:F(A)\to G(B)$ in $\mathcal{C}$; a morphism $(A,B,f)\to (A^{\prime },B^{\prime },f^{\prime })$ is a pair $(u:A\to A^{\prime },v:B\to B^{\prime })$ with $G(v)\circ f=f^{\prime }\circ F(u)$. Taking $\mathcal{B}=1$ recovers the slice $\mathcal{C}/c$ over an object $c$; taking $\mathcal{A}=1$ recovers the coslice $c/\mathcal{C}$; taking $\mathcal{A}=\mathcal{B}=\mathcal{C}$ with both functors the identity recovers the arrow category ${\mathcal{C}}^{\to }$.

Definition (category of elements). For a presheaf $F:{\mathcal{C}}^{\mathrm{op}}\to \mathbf{Set}$, the category of elements ${\int }_{\mathcal{C}}F$ (the Grothendieck construction) has objects the pairs $(A,x)$ with $A$ in $\mathcal{C}$ and $x\in F(A)$; a morphism $(A,x)\to (B,y)$ is a morphism $g:A\to B$ in $\mathcal{C}$ with $F(g)(y)=x$. The forgetful functor ${\int }_{\mathcal{C}}F\to \mathcal{C}$ exhibits a discrete fibration over $\mathcal{C}$.

Counterexamples to common slips Intermediate+

• A family of isomorphisms need not be natural. Every finite-dimensional vector space is isomorphic to its single dual $V^{\ast }$, but no choice of isomorphisms $V\to V^{\ast }$ is natural in $V$: any such family forces a choice of inner product or basis, and the naturality square fails for a generic linear map. Naturality is a genuine constraint, not a free consequence of pointwise isomorphism.

• Equivalence is not isomorphism of categories. The category of finite-dimensional vector spaces is equivalent to its skeleton with one object ${\mathbb{R}}^n$ per dimension, but it is not isomorphic to it: the original has a proper class of objects while the skeleton has countably many, so no bijection on objects exists. Equivalence permits collapsing isomorphic objects; isomorphism forbids it.

• The comma category depends on the functors, not just the categories. Writing $F\downarrow G$ without naming $F$ and $G$ is meaningless: changing either functor changes which triples $(A,B,f)$ exist. The slice and coslice are the special cases where one functor lands on a single chosen object.

Key theorem with proof Intermediate+

The signature result is the characterisation of equivalences without reference to an explicit quasi-inverse.

Theorem (characterisation of equivalences). A functor $F:\mathcal{C}\to \mathcal{D}$ is an equivalence of categories if and only if $F$ is full, faithful, and essentially surjective ^{[Mac Lane 1998]}. The construction of a quasi-inverse from the second condition uses the axiom of choice.

Proof. ($\Rightarrow$) Suppose $F$ is an equivalence with quasi-inverse $G$ and natural isomorphisms $\eta :{\mathrm{Id}}_{\mathcal{C}}\cong GF$, $\epsilon :FG\cong {\mathrm{Id}}_{\mathcal{D}}$. Essentially surjective: for any object $D$ of $\mathcal{D}$, ${\epsilon }_D:FG(D)\to D$ is an isomorphism, so $D$ is isomorphic to $F(G(D))$. Faithful: if $F(f)=F(g)$ for $f,g:A\to B$ then $GF(f)=GF(g)$; the naturality of $\eta$ gives $f={\eta }_B^{-1}\circ GF(f)\circ {\eta }_A={\eta }_B^{-1}\circ GF(g)\circ {\eta }_A=g$. The same argument with $\epsilon$ shows $G$ is faithful. Full: given $h:F(A)\to F(B)$, set $f={\eta }_B^{-1}\circ G(h)\circ {\eta }_A:A\to B$. Naturality of $\eta$ at $f$ reads $GF(f)\circ {\eta }_A={\eta }_B\circ f=G(h)\circ {\eta }_A$, and since ${\eta }_A$ is invertible, $GF(f)=G(h)$. As $G$ is faithful, $F(f)=h$, so $F$ is full.

($\Leftarrow$) Suppose $F$ is full, faithful, and essentially surjective. For each object $D$ of $\mathcal{D}$ choose an object $G(D)$ of $\mathcal{C}$ and an isomorphism ${\epsilon }_D:F(G(D))\to D$; this is the one appeal to choice. Extend $G$ to morphisms: for $h:D\to D^{\prime }$ the composite ${\epsilon }_{D^{\prime }}^{-1}\circ h\circ {\epsilon }_D:F(G(D))\to F(G(D^{\prime }))$ is, by fullness and faithfulness of $F$, equal to $F(g)$ for a unique $g:G(D)\to G(D^{\prime })$; set $G(h)=g$. Uniqueness forces $G(\mathrm{id})=\mathrm{id}$ and $G(h^{\prime }\circ h)=G(h^{\prime })\circ G(h)$, so $G$ is a functor, and $\epsilon :FG\Rightarrow {\mathrm{Id}}_{\mathcal{D}}$ is natural by construction with invertible components, hence a natural isomorphism. For $\eta$: each ${\epsilon }_{F(A)}:FGF(A)\to F(A)$ is an isomorphism between values of $F$, so by fullness and faithfulness equals $F({\eta }_A)$ for a unique isomorphism ${\eta }_A:A\to GF(A)$. Naturality of $\eta$ follows from naturality of $\epsilon$ together with faithfulness of $F$. Thus $G$ is a quasi-inverse and $F$ is an equivalence. $\square$

Bridge. This theorem builds toward every structural comparison made later by exhibiting an equivalence, and it appears again in 41.04.02, where the Yoneda embedding is shown full and faithful and the question of essential surjectivity becomes the representability problem. The foundational reason the trichotomy suffices is that fullness and faithfulness let an isomorphism between images be pulled back to a unique isomorphism upstream, so the quasi-inverse is forced on morphisms once chosen on objects; the central insight is that the only genuinely free datum is the choice in each isomorphism class, which is exactly where the axiom of choice enters. This is exactly the mechanism by which a category and its skeleton are equivalent, and putting these together shows equivalence is the correct notion of sameness — coarser than isomorphism of categories, fine enough to preserve every property stated in the categorical language, and the bridge from the duality of 41.01.01 to the universal constructions that follow, since limits, colimits, and adjoints are all defined only up to the natural isomorphism this theorem trades in.

Exercises Intermediate+

Advanced results Master

Theorem (interchange law). For categories $\mathcal{C},\mathcal{D},\mathcal{E}$, functors and natural transformations $\alpha :F\Rightarrow F^{\prime }$, ${\alpha }^{\prime }:F^{\prime }\Rightarrow F^{\prime \prime }$ in $[\mathcal{C},\mathcal{D}]$ and $\beta :G\Rightarrow G^{\prime }$, ${\beta }^{\prime }:G^{\prime }\Rightarrow G^{\prime \prime }$ in $[\mathcal{D},\mathcal{E}]$, the two compositions satisfy $$ (\beta'\cdot\beta)(\alpha'\cdot\alpha) = (\beta'\alpha')\cdot(\beta*\alpha). $$ Both sides are natural transformations $GF\Rightarrow G^{\prime \prime }{F}^{\prime \prime }$, and the identity says that horizontal and vertical composition commute. The proof is the naturality square of $\beta$ at the morphism ${\alpha }_A$, read componentwise: each component of either side equals ${\beta }_{F^{\prime \prime }(A)}^{\prime }\circ {\beta }_{F^{\prime \prime }(A)}\circ G({\alpha }_A^{\prime })\circ G({\alpha }_A)$ after using functoriality of $G,G^{\prime }$ and naturality of $\beta ,{\beta }^{\prime }$ to slide the components past one another. The interchange law is exactly the coherence that makes $\mathbf{Cat}$ a strict $2$-category: objects are small categories, $1$-morphisms are functors, $2$-morphisms are natural transformations, with the two composites of $2$-cells governed by this single equation.

Theorem (functor categories inherit structure). If $\mathcal{D}$ has all limits of a given shape, so does $[\mathcal{C},\mathcal{D}]$, computed pointwise: $({\lim }_i{F}_i)(A)={\lim }_i(F_i(A))$, with the universal property checked component by component. In particular presheaf categories $[{\mathcal{C}}^{\mathrm{op}},\mathbf{Set}]$ are complete and cocomplete, since $\mathbf{Set}$ is. Natural transformations between such limits are determined by their components, so the evaluation functors ${\mathrm{ev}}_A:[\mathcal{C},\mathcal{D}]\to \mathcal{D}$ jointly reflect and preserve the pointwise structure.

Theorem (skeleton and the role of choice). Every category $\mathcal{C}$ has a skeleton $\mathrm{sk}(\mathcal{C})$: a full subcategory containing exactly one object from each isomorphism class, and the inclusion $\mathrm{sk}(\mathcal{C})\hookrightarrow \mathcal{C}$ is an equivalence. Constructing the skeleton selects a representative per class, an appeal to the axiom of choice; conversely, the statement "every fully faithful essentially surjective functor is an equivalence" is equivalent over $\mathbf{ZF}$ to a choice principle. Two categories are equivalent if and only if their skeletons are isomorphic, so equivalence classes of categories are isomorphism classes of skeletons.

Theorem (category of elements as a comma category; the Grothendieck construction). For a presheaf $F:{\mathcal{C}}^{\mathrm{op}}\to \mathbf{Set}$, the category of elements satisfies ${\int }_{\mathcal{C}}F\cong (y\downarrow F)$, the comma category of the Yoneda embedding $y:\mathcal{C}\to [{\mathcal{C}}^{\mathrm{op}},\mathbf{Set}]$ over the object $F$. The forgetful functor ${\int }_{\mathcal{C}}F\to \mathcal{C}$ is a discrete fibration, and $F$ is recovered as the presheaf of fibres. The general Grothendieck construction upgrades a functor ${\mathcal{C}}^{\mathrm{op}}\to \mathbf{Cat}$ into a fibration over $\mathcal{C}$, the indexed-versus-fibred-category equivalence that organises descent and stack theory.

Synthesis. Putting these together, natural transformations are the $2$-cells that turn $\mathbf{Cat}$ from a category into a $2$-category, and the interchange law is the single coherence that makes the two ways of composing them agree. The foundational reason the functor category $[\mathcal{C},\mathcal{D}]$ behaves so well is that everything in it is checked pointwise: limits, isomorphisms, and naturality all reduce to the target $\mathcal{D}$ object by object, which is exactly why presheaf categories inherit completeness for free and why the Yoneda machinery of 41.04.02 lives there. This generalises the duality of 41.01.01: where duality came from the involution $({\mathcal{C}}^{\mathrm{op}}{)}^{\mathrm{op}}=\mathcal{C}$, equivalence comes from allowing identities only up to natural isomorphism, and the central insight is that the correct notion of sameness for categories is the one that ignores the difference between isomorphic objects — the skeleton makes this precise and exposes where the axiom of choice is spent. The bridge is that equivalence, comma categories, and the category of elements are all assembled from the same primitive, the natural transformation: limits and colimits in 41.02.01 are defined up to the natural isomorphisms catalogued here, adjunctions in 41.03.01 are pairs of natural transformations (unit and counit) subject to the triangle identities, and the Yoneda embedding in 41.04.02 is full and faithful precisely because natural transformations between representables are computed by this same pointwise discipline.

Full proof set Master

Proposition 1 (vertical composition is associative and unital; $[\mathcal{C},\mathcal{D}]$ is a category). Vertical composition of natural transformations is associative, has the identity transformations as units, and produces a category $[\mathcal{C},\mathcal{D}]$.

Proof. For $\alpha :F\Rightarrow G$, $\beta :G\Rightarrow H$, $\gamma :H\Rightarrow K$, the component of $(\gamma \cdot \beta )\cdot \alpha$ at $A$ is $({\gamma }_A\circ {\beta }_A)\circ {\alpha }_A$ and of $\gamma \cdot (\beta \cdot \alpha )$ is ${\gamma }_A\circ ({\beta }_A\circ {\alpha }_A)$; these agree by associativity in $\mathcal{D}$. The composite is natural: pasting the naturality squares of $\alpha$ and $\beta$ along their shared edge $G(f)$ gives the square for $\beta \cdot \alpha$. The identity $1_F$ with components ${\mathrm{id}}_{F(A)}$ satisfies $\alpha \cdot 1_F=\alpha =1_G\cdot \alpha$ componentwise. Hence functors $\mathcal{C}\to \mathcal{D}$ and natural transformations form a category. $\square$

Proposition 2 (whiskering associates with composition). For functors $F,F^{\prime }:\mathcal{C}\to \mathcal{D}$, a transformation $\alpha :F\Rightarrow F^{\prime }$, and functors $G:\mathcal{B}\to \mathcal{C}$, $H:\mathcal{D}\to \mathcal{E}$, one has $H(\alpha G)=(H\alpha )G$ as transformations $HFG\Rightarrow H{F}^{\prime }G$.

Proof. The component of $\alpha G$ at an object $B$ of $\mathcal{B}$ is ${\alpha }_{G(B)}$. Applying $H$ gives component $H({\alpha }_{G(B)})$ for $H(\alpha G)$. The component of $H\alpha$ at an object $C$ of $\mathcal{C}$ is $H({\alpha }_C)$, and post-whiskering by $G$ evaluates at $C=G(B)$, giving $H({\alpha }_{G(B)})$. The components coincide, and naturality of each side was established by Exercise 3 and its mirror. $\square$

Proposition 3 (the interchange law). With notation as in the Advanced-results statement, $(\beta'\cdot\beta)(\alpha'\cdot\alpha)=(\beta'\alpha')\cdot(\beta\alpha)$.*

Proof. It suffices to treat the basic case $\beta \ast \alpha$ for $\alpha :F\Rightarrow F^{\prime }$, $\beta :G\Rightarrow G^{\prime }$ and verify the two definitions of $\ast$ agree, then assemble. Fix $A$. Naturality of $\beta$ at the morphism ${\alpha }_A:F(A)\to F^{\prime }(A)$ reads $G^{\prime }({\alpha }_A)\circ {\beta }_{F(A)}={\beta }_{F^{\prime }(A)}\circ G({\alpha }_A)$; this is the common value defining $(\beta \ast \alpha {)}_A$. For the interchange identity, expand the left side at $A$: $({\beta }^{\prime }\cdot \beta )\ast ({\alpha }^{\prime }\cdot \alpha )$ has component the common value of $(G^{\prime \prime }({\alpha }_A^{\prime }\circ {\alpha }_A))\circ ({\beta }_{F(A)}^{\prime }\circ {\beta }_{F(A)})$ and its transpose. Expand the right side: vertical composition of $({\beta }^{\prime }\ast {\alpha }^{\prime }{)}_A=G^{\prime \prime }({\alpha }_A^{\prime })\circ {\beta }_{F^{\prime }(A)}^{\prime }$ after $(\beta \ast \alpha {)}_A=G^{\prime }({\alpha }_A)\circ {\beta }_{F(A)}$. Using functoriality of $G^{\prime },G^{\prime \prime }$ and naturality of ${\beta }^{\prime }$ at ${\alpha }_A$ to commute ${\beta }_{F^{\prime }(A)}^{\prime }$ past $G^{\prime }({\alpha }_A)$, both expansions reduce to $G^{\prime \prime }({\alpha }_A^{\prime })\circ G^{\prime \prime }({\alpha }_A)\circ {\beta }_{F(A)}^{\prime }\circ {\beta }_{F(A)}$. The components agree for all $A$, so the transformations are equal. $\square$

Proposition 4 (pointwise limits in functor categories). If $\mathcal{D}$ has limits of shape $\mathcal{J}$, then $[\mathcal{C},\mathcal{D}]$ has them, and they are computed pointwise.

Proof. Let $D:\mathcal{J}\to [\mathcal{C},\mathcal{D}]$ be a diagram, so each $D_j$ is a functor $\mathcal{C}\to \mathcal{D}$ and each arrow of $\mathcal{J}$ gives a natural transformation. For each object $A$ of $\mathcal{C}$ form the limit $L(A)={\lim }_j{D}_j(A)$ in $\mathcal{D}$, with limiting cone ${\pi }_{j,A}:L(A)\to D_j(A)$. For a morphism $f:A\to B$, the maps $D_j(f)\circ {\pi }_{j,A}$ form a cone over the diagram at $B$, inducing a unique $L(f):L(A)\to L(B)$; functoriality of $L$ follows from uniqueness. The ${\pi }_j$ are natural by construction, and given any cone in $[\mathcal{C},\mathcal{D}]$ over $D$ its components factor uniquely through each $L(A)$, the factorisations assembling into a natural transformation by uniqueness. So $L$ with the ${\pi }_j$ is the limit, computed pointwise. $\square$

Connections Master

• Categories, functors, and duality 41.01.01. This unit is the direct continuation of the functor theory developed there: the faithful/full/essentially-surjective trichotomy introduced in 41.01.01 is assembled here into the characterisation of equivalence, and the opposite category and slice constructions of 41.01.01 reappear as the special comma categories $F\downarrow G$. The duality involution $({\mathcal{C}}^{\mathrm{op}}{)}^{\mathrm{op}}=\mathcal{C}$ is the degenerate case of the equivalence relation studied here, and the survey 01.02.09 states the natural-transformation definition more briefly within its single-sweep treatment.

• Limits and colimits 41.02.01. Limits and colimits are defined only up to the natural isomorphisms catalogued in this unit, and Proposition 4 shows functor categories inherit them pointwise — the mechanism by which presheaf categories become complete and cocomplete. The category of elements and the comma categories here are precisely the shapes over which limit and colimit cones are taken, so this unit supplies the $2$-categorical language in which 41.02.01 states universal properties.

• The Yoneda lemma and representability 41.04.02. The Yoneda lemma computes the natural transformations between a representable functor and an arbitrary presheaf, so it is a theorem about the functor category $[{\mathcal{C}}^{\mathrm{op}},\mathbf{Set}]$ built here; the Yoneda embedding is full and faithful, hence an equivalence onto its image by the theorem of this unit, and the category of elements ${\int }_{\mathcal{C}}F\cong (y\downarrow F)$ is the comma-category presentation that makes representability the existence of a terminal element.

• Adjunctions 41.03.01. An adjunction is a pair of natural transformations — the unit and counit — subject to the triangle identities, all of which are equations in functor categories defined here. Every equivalence is in particular an adjunction (an adjoint equivalence after adjusting the isomorphisms), so the characterisation theorem of this unit is the entry point to the adjoint-functor theory of 41.03.01.

Historical & philosophical context Master

Natural transformations were the phenomenon that category theory was invented to make precise. In their 1945 paper General theory of natural equivalences ^{[Eilenberg-Mac Lane 1945]}, Samuel Eilenberg and Saunders Mac Lane set out to define the word "natural" in the statement that a finite-dimensional vector space is naturally isomorphic to its double dual but only unnaturally isomorphic to its single dual. Defining "natural transformation" required first defining functor, and defining functor required first defining category, so the three notions entered the literature in the reverse of their conceptual order. The naturality square — the commuting rectangle that the components must satisfy — is already the central definition of that founding paper.

The functor category and the equivalence of categories belong to the consolidation of the subject in Mac Lane's Categories for the Working Mathematician (1971; 2nd ed. 1998) ^{[Mac Lane 1998]}, where the interchange law, the $2$-categorical structure of $\mathbf{Cat}$, comma categories, and the characterisation of equivalences by the full-faithful-essentially-surjective trichotomy are organised as standard tools. The Grothendieck construction and the category of elements emerged from Alexander Grothendieck's reworking of algebraic geometry in the early 1960s, where fibred categories over a base encode descent data; the equivalence between indexed categories and fibrations is the structural core of that theory. The dependence of the equivalence characterisation on the axiom of choice was isolated in the categorical-foundations literature and is the reason constructive category theory distinguishes equivalences from adjoint equivalences with specified data.

Bibliography Master

@article{EilenbergMacLane1945,
  author  = {Eilenberg, Samuel and Mac Lane, Saunders},
  title   = {General theory of natural equivalences},
  journal = {Transactions of the American Mathematical Society},
  volume  = {58},
  year    = {1945},
  pages   = {231--294}
}

@book{MacLane1998,
  author    = {Mac Lane, Saunders},
  title     = {Categories for the Working Mathematician},
  edition   = {2},
  publisher = {Springer},
  series    = {Graduate Texts in Mathematics 5},
  year      = {1998}
}

@book{Riehl2016,
  author    = {Riehl, Emily},
  title     = {Category Theory in Context},
  publisher = {Dover},
  year      = {2016}
}

@book{Awodey2010,
  author    = {Awodey, Steve},
  title     = {Category Theory},
  edition   = {2},
  publisher = {Oxford University Press},
  series    = {Oxford Logic Guides 52},
  year      = {2010}
}

@book{Leinster2014,
  author    = {Leinster, Tom},
  title     = {Basic Category Theory},
  publisher = {Cambridge University Press},
  series    = {Cambridge Studies in Advanced Mathematics 143},
  year      = {2014}
}

@book{Kelly1982,
  author    = {Kelly, G. Max},
  title     = {Basic Concepts of Enriched Category Theory},
  publisher = {Cambridge University Press},
  series    = {London Mathematical Society Lecture Note Series 64},
  year      = {1982}
}