Adjunctions — the unit-counit calculus and the adjoint functor theorems

Intuition Beginner

Adjoints are best approximations between two worlds. From a plain set of letters, the free-group construction builds the most economical group that contains them, inventing only the products and inverses the group laws force. The matching forgetful step strips a group back to its bare set. These two directions are not rivals: they are a paired team, and the precise way they cooperate is called an adjunction.

This unit studies the two pieces of structure every adjunction carries and the arithmetic for combining them. The first piece, the unit, records what the build step must insert. The second, the counit, records what the forget step can safely strip off. Between them they form a small toolkit, and the toolkit obeys two cancellation rules. Those two rules, the triangle identities, are the whole calculus.

Once the two rules are in hand, a great many facts about adjoints turn into bookkeeping. Slide a build past a forget, cancel a matched pair, read off the answer. This unit teaches that bookkeeping, then turns the question around: given only a forget step, when can a build step be guaranteed to exist at all? That reverse question leads to the adjoint functor theorems.

Visual Beginner

Picture the build and forget moves as arrows running between two columns, with the unit and counit as the labels that turn one direction into the other. The two cancellation rules are what keep the picture honest.

the pieces and rules	plain reading
the unit	the map inserting generators into the free object
the counit	the map multiplying out a rebuilt group
first cancellation	build, insert generators, then multiply out — back where you started
second cancellation	forget to a set, rebuild, then map back — the original group is unchanged

The cap-and-cup picture is the heart of the calculus. When a cap sits directly above a matching cup, the zigzag pulls straight: that straightening is each triangle identity, and every manipulation below is some combination of insert, multiply-out, and pull-straight.

Worked example Beginner

Build the free group on two letters, $a$ and $b$. Its elements are all reduced words in $a$, $b$, and their inverses $a^{-1}$ and $b^{-1}$: words like $ab{a}^{-1}$, $b^{2}a$, or the empty word. Multiplication concatenates and cancels adjacent inverse pairs. There is no relation between $a$ and $b$ except the group laws, so this group is free of extra equations.

The unit of the adjunction inserts the two letters as generators. Concretely it is the set map sending $a\mapsto a$ and $b\mapsto b$, regarded as landing in the underlying set of the free group. Nothing else is inserted — every other element is forced by the group laws once the two generators are named.

Step 1. Pick a small target. Take the flip group on two elements, written $0$ and $1$, where adding $1$ to $1$ returns to $0$. A group homomorphism from the free group on $a,b$ into the flip group must send each generator somewhere, and each choice extends in exactly one way.

Step 2. Count the choices. The generator $a$ has $2$ possible images ($0$ or $1$), and the generator $b$ independently has $2$ possible images. Together that is $2\times 2=4$ homomorphisms, one for each way to assign the two generators.

Step 3. Count the other way. A set map from the two-element set into the underlying set of the flip group also has $2\times 2=4$ options, since each letter chooses between two values. Four and four.

What this tells us: the four group homomorphisms match the four set maps, one for one, and this matching holds for every target group at once. The unit's job of inserting the two generators is exactly what makes a homomorphism the same labour as choosing where the raw letters go.

Check your understanding Beginner

Formal definition Intermediate+

Throughout, $F⊣G$ denotes an adjunction with $F:\mathcal{C}\to \mathcal{D}$ left adjoint to $G:\mathcal{D}\to \mathcal{C}$, unit $\eta :{\mathrm{Id}}_{\mathcal{C}}\Rightarrow GF$, and counit $\epsilon :FG\Rightarrow {\mathrm{Id}}_{\mathcal{D}}$, in the notation of 41.03.01. The unit and counit satisfy the triangle identities $(\epsilon F)\circ (F\eta )={\mathrm{id}}_F$ and $(G\epsilon )\circ (\eta G)={\mathrm{id}}_G$, and the hom-set bijection is recovered as $\phi (g)=Gg\circ {\eta }_c$ with inverse ${\phi }^{-1}(f)={\epsilon }_d\circ Ff$. The present unit develops the calculus of these data: a systematic equational and diagrammatic theory for sliding adjoints past each other.

Definition (mate of a natural transformation). Let $F⊣G$ as above and let $H:\mathcal{X}\to \mathcal{C}$ and $K:\mathcal{X}\to \mathcal{D}$ be functors. The mate correspondence (or transpose) is the bijection of natural transformations

$$\mathrm{Nat}(FH,K)\cong \mathrm{Nat}(H,GK),\alpha ⟼\hat{\alpha }=G\alpha \circ \eta H,$$

with inverse $\beta \mapsto {\hat{\beta }}^{-1}=\epsilon K\circ F\beta$, where $\eta H$ and $\epsilon K$ denote whiskering of the unit and counit with $H$ and $K$ ^{[Kelly 1982]}. Componentwise, ${\hat{\alpha }}_x=G({\alpha }_x)\circ {\eta }_{Hx}$. This is the hom-set adjunction $\mathcal{D}(Fc,d)\cong \mathcal{C}(c,Gd)$ of 41.03.01 lifted objectwise to functor categories, and it is the workhorse of the entire calculus: a natural transformation between composites of $F$ can be replaced, at no cost, by one between composites of $G$.

Definition (mate of a square). Let $F⊣G$ (between $\mathcal{A},\mathcal{B}$) and $F^{\prime }⊣G^{\prime }$ (between $\mathcal{C},\mathcal{D}$) be adjunctions, and let $U:\mathcal{A}\to \mathcal{C}$ and $V:\mathcal{B}\to \mathcal{D}$ be functors. For any $2$-cell $\sigma :F^{\prime }U\Rightarrow VF$, its mate is the $2$-cell $\bar{\sigma }:UG\Rightarrow G^{\prime }V$ given by the pasting

$$UG\stackrel{{\eta }^{\prime }UG}{\to }{G}^{\prime }{F}^{\prime }UG\stackrel{G^{\prime }\sigma G}{\to }{G}^{\prime }VFG\stackrel{G^{\prime }V\epsilon }{\to }{G}^{\prime }V,$$

using the unit ${\eta }^{\prime }$ of $F^{\prime }⊣G^{\prime }$ and the counit $\epsilon$ of $F⊣G$. The mate is computed by inserting one unit, sliding $\sigma$ across, and cancelling with one counit; the inverse mate inserts $\eta$ and cancels with ${\epsilon }^{\prime }$.

Definition (adjunction in a $2$-category). An adjunction in a $2$-category $\mathcal{K}$ consists of $1$-cells $f:A\to B$, $g:B\to A$ and $2$-cells $\eta :1_A\Rightarrow gf$, $\epsilon :fg\Rightarrow 1_B$ satisfying the triangle identities $(\epsilon f)\circ (f\eta )=1_f$ and $(g\epsilon )\circ (\eta g)=1_g$. Adjunctions in $\mathcal{K}=\mathbf{Cat}$ are the ordinary adjunctions of 41.03.01; in the $2$-category $\mathbf{Pos}$ of posets they are Galois connections; in the $2$-category of monoidal categories, monoidal functors, and monoidal natural transformations they are monoidal adjunctions. The calculus that follows is internal to any $\mathcal{K}$, because it uses only the two triangle equations.

Counterexamples to common slips Intermediate+

• The mate is not the inverse natural transformation. Transposing $\alpha :FH\Rightarrow K$ across $F⊣G$ produces $\hat{\alpha }:H\Rightarrow GK$ living between different functors; it is a bijection of sets of natural transformations, not an operation that returns $\alpha$ back. Confusing the two leads to type errors when chaining mates.

• A fully faithful left adjoint is not the same as an equivalence. $F⊣G$ with $F$ fully faithful forces the unit $\eta$ to be an isomorphism, but the counit need not be; the inclusion of abelian groups into groups has a fully faithful right adjoint (the forgetful direction) and a non-invertible counit. Both triangles must collapse to invertibility for an equivalence.

• The solution-set condition is not automatic from continuity. A continuous functor between complete categories need not satisfy it, and without it no left adjoint exists: the forgetful functor from complete Boolean algebras (with complete homomorphisms) to sets preserves all limits yet has no left adjoint. The calculus below constructs the unit and counit, but only once the condition supplies a foothold.

Key theorem with proof Intermediate+

The signature result of the unit-counit calculus is that transposing a whole natural transformation is a genuine, invertible operation governed entirely by the two triangle identities.

Theorem (the mate correspondence is a natural bijection). Let $F⊣G$ with unit $\eta$ and counit $\epsilon$, and let $H:\mathcal{X}\to \mathcal{C}$, $K:\mathcal{X}\to \mathcal{D}$. The assignments

$$\alpha :FH\Rightarrow K\mapsto \hat{\alpha }=G\alpha \circ \eta H,\beta :H\Rightarrow GK\mapsto \check{\beta }=\epsilon K\circ F\beta$$

are mutually inverse bijections between $\mathrm{Nat}(FH,K)$ and $\mathrm{Nat}(H,GK)$, natural in $H$ and $K$ ^{[Mac Lane 1998]}.

Proof. Both assignments are natural in $H$ and $K$ because they are built from the natural $\eta ,\epsilon$ by post- and pre-composition, and naturality in $H,K$ is whiskering. It remains to check the two round trips. For $\alpha :FH\Rightarrow K$,

$$\check{\hat{\alpha }}=\epsilon K\circ F(G\alpha \circ \eta H)=\epsilon K\circ FG\alpha \circ F\eta H=\alpha \circ \epsilon FH\circ F\eta H=\alpha ,$$

where the third equality is naturality of $\epsilon$ applied to $\alpha :FH\Rightarrow K$ (giving $\epsilon K\circ FG\alpha =\alpha \circ \epsilon FH$), and the last is the first triangle identity $(\epsilon F)\circ (F\eta )={\mathrm{id}}_F$ whiskered with $H$, so $\epsilon FH\circ F\eta H={\mathrm{id}}_{FH}$.

For the other round trip, take $\beta :H\Rightarrow GK$:

$$\hat{\check{\beta }}=G(\epsilon K\circ F\beta )\circ \eta H=G\epsilon K\circ GF\beta \circ \eta H=G\epsilon K\circ \eta GK\circ \beta =\beta ,$$

using naturality of $\eta$ at $\beta :H\Rightarrow GK$ (giving $GF\beta \circ \eta H=\eta GK\circ \beta$) and then the second triangle identity $(G\epsilon )\circ (\eta G)={\mathrm{id}}_G$ whiskered with $K$, so $G\epsilon K\circ \eta GK={\mathrm{id}}_{GK}$. Hence the two assignments are mutually inverse, and each is a bijection. The square-mate of the Formal definition is this same bijection iterated across two adjunctions. $\square$

Bridge. The mate correspondence builds toward every manipulation in the rest of this unit and appears again in 41.05.01, where the monad multiplication $G\epsilon F$ together with the unit $\eta$ is recognized as a monad, and again in 41.06.02, where Kan-extension formulas are computed as mates. The foundational reason the mate calculus works is that an adjunction is, objectwise, an isomorphism of hom-functors, so transposing a whole natural transformation amounts to transposing each component, and the triangle identities guarantee each round trip is the identity. This is exactly the two-dimensional lift of the hom-set bijection of 41.03.01, and the bridge is that every result below — the fully-faithful characterizations, the composed adjunctions, and the adjoint functor theorem — is a calculation with $\eta$ and $\epsilon$ pasted into commuting diagrams.

Exercises Intermediate+

Advanced results Master

Theorem (fully faithful adjunctions via unit and counit). For an adjunction $F⊣G$ with unit $\eta$ and counit $\epsilon$: the left adjoint $F$ is fully faithful if and only if $\eta$ is a natural isomorphism; dually the right adjoint $G$ is fully faithful if and only if $\epsilon$ is a natural isomorphism ^{[Mac Lane 1998]}. These are the first fruits of the calculus: invertibility of the unit or counit is detected by a representability property of the adjoint, and each is proved by pasting $\eta$ against $\epsilon$ and invoking a triangle identity. A full proof is Proposition 1 below.

Theorem (adjunctions in any $2$-category; the zigzag identities). The definition of an adjunction given above makes sense verbatim in any $2$-category $\mathcal{K}$, and in the string-diagram calculus for $\mathcal{K}$ the triangle identities become the zigzag identities: the unit $\eta$ is drawn as a cap and the counit $\epsilon$ as a cup, and the two triangles assert that a cap directly above a cup pulls straight to a vertical strand ^{[Joyal-Street 1991]}. This topological reading is why the calculus composes: pasting diagrams of units and counits reduce, by repeated straightening, to a normal form, so questions about whether two pastings are equal become questions about isotopy of strings. In $\mathcal{K}=\mathbf{Cat}$ this recovers ordinary adjunctions; in $\mathcal{K}=\mathbf{Pos}$ it gives Galois connections; in the $2$-category of monoidal categories it gives monoidal adjunctions, for which the unit and counit are themselves monoidal natural transformations and the triangle identities are monoidal. The monoidal adjunctions are exactly those preserving the tensor structure on both sides, and they are the structural backbone of duality in linear logic and of the tensor-hom adjunction in enriched category theory.

Theorem (the adjoint functor theorem via the unit-counit calculus). Let $\mathcal{D}$ be locally small and complete, $G:\mathcal{D}\to \mathcal{C}$ locally small. Then $G$ admits a left adjoint $F$ together with unit $\eta$ and counit $\epsilon$ satisfying the triangle identities if and only if $G$ preserves all small limits and satisfies the solution-set condition ^{[Freyd 1964]}. The necessity is the calculus read backwards: RAPL (right adjoints preserve limits, 41.03.02) is a single mate calculation, and the adjunction itself supplies the singleton solution set $\{{\eta }_c\}$. The sufficiency is the substantive direction and is where the calculus earns its keep: the solution-set condition plus completeness gives, via the weakly-initial-set lemma of 41.03.02, an initial object $(Fc,{\eta }_c)$ in each comma category $c\downarrow G$; one then constructs the counit ${\epsilon }_d$ by initiality at $c=Gd$ and verifies both triangle identities by initiality, exactly as in Exercise 8 and Proposition 3. This presentation complements the comma-category proof of 41.03.02: there the adjunction is built by exhibiting the hom-set bijection, here by building $\eta ,\epsilon$ and checking the two equations directly.

Theorem (adjoints, mates, and Kan extensions). A functor $G:\mathcal{D}\to \mathcal{C}$ has a left adjoint if and only if for each $c\in \mathcal{C}$ the functor $\mathcal{C}(c,G-):\mathcal{D}\to \mathbf{Set}$ is representable; the representing objects assemble into $F$ by uniqueness of representations (Yoneda, 41.04.02). Equivalently, the left adjoint is the pointwise left Kan extension of the identity along $G$, ${\mathrm{Lan}}_G{\mathrm{Id}}_{\mathcal{D}}$, when that extension exists and is preserved by $G$; the Kan-extension formulas of 41.06.02 are themselves mate calculations, transposing the defining $2$-cells of the extension across $F⊣G$. The calculus of mates is thus the common engine: adjoint existence, Kan extensions, and the preservation theorems are all instances of "slide a $2$-cell across an adjunction and cancel."

Synthesis. Putting these together, the unit-counit calculus reveals adjunctions as a single algebraic structure recurring across dimensions: the same $\eta ,\epsilon$ and triangle identities define adjunctions in $\mathbf{Cat}$, in any $2$-category, in monoidal categories, and in posets as Galois connections. The foundational reason the calculus is so effective is that the triangle identities are the zigzag identities of string diagrams, so pasting a unit against a counit topologically cancels, and fully-faithfulness, composed adjunctions, and RAPL all become isotopies of a diagram. This is exactly why the adjoint functor theorem splits so cleanly: the necessity (RAPL) is a mate calculation, and the sufficiency is the explicit construction of ${\eta }_c,{\epsilon }_d$ verified against the same two triangles, which generalises the comma-category argument of 41.03.02 by building the unit and counit directly, and is dual to the colimit-side version throughout. The central insight is that existence of a left adjoint is a representability condition, and the bridge is that the calculus of mates is the operating system on which monads 41.05.01, Kan extensions 41.06.02, and the derived adjunctions of homological algebra 01.06.01 all run.

Full proof set Master

Proposition 1 (fully faithful adjoints via unit and counit). For $F⊣G$ with unit $\eta$ and counit $\epsilon$, the left adjoint $F$ is fully faithful if and only if $\eta$ is a natural isomorphism. Dually, $G$ is fully faithful if and only if $\epsilon$ is a natural isomorphism.

Proof. ($\eta$ iso $\Rightarrow$ $F$ fully faithful.) We show $F_{x,y}:\mathcal{C}(x,y)\to \mathcal{D}(Fx,Fy)$ is a bijection. For $g:x\to y$, naturality of $\eta$ gives ${\eta }_y\circ g=GFg\circ {\eta }_x=\phi (Fg)$, where $\phi (h)=Gh\circ {\eta }_x$ is the adjunction transpose at $z=Fy$. Since ${\eta }_y$ is invertible, $g={\eta }_y^{-1}\circ \phi (Fg)$, so $g$ is uniquely determined by $Fg$: $F_{x,y}$ is injective. For surjectivity, given $h:Fx\to Fy$, set $g={\eta }_y^{-1}\circ \phi (h)={\eta }_y^{-1}\circ Gh\circ {\eta }_x$. Then $\phi (Fg)={\eta }_y\circ g=\phi (h)$, and since $\phi$ is a bijection, $Fg=h$. Hence $F_{x,y}$ is bijective, and $F$ is fully faithful.

($F$ fully faithful $\Rightarrow$ $\eta$ iso.) For each $x$, the component ${\epsilon }_{Fx}:FGFx\to Fx$ lies in $\mathcal{D}(FGFx,Fx)$; since $F$ is full and faithful, ${\epsilon }_{Fx}=F({\theta }_x)$ for a unique ${\theta }_x:GFx\to x$. Now $F({\theta }_x\circ {\eta }_x)=F{\theta }_x\circ F{\eta }_x={\epsilon }_{Fx}\circ F{\eta }_x={\mathrm{id}}_{Fx}=F({\mathrm{id}}_x)$ by the first triangle identity, and faithfulness of $F$ gives ${\theta }_x\circ {\eta }_x={\mathrm{id}}_x$. For the other composite, the second triangle identity at $Fx$ reads $G{\epsilon }_{Fx}\circ {\eta }_{GFx}={\mathrm{id}}_{GFx}$; substituting ${\epsilon }_{Fx}=F{\theta }_x$ gives $GF{\theta }_x\circ {\eta }_{GFx}={\mathrm{id}}_{GFx}$. By naturality of $\eta$ at ${\theta }_x$, ${\eta }_x\circ {\theta }_x=GF{\theta }_x\circ {\eta }_{GFx}$. Hence ${\eta }_x\circ {\theta }_x={\mathrm{id}}_{GFx}$, and ${\theta }_x$ is a two-sided inverse of ${\eta }_x$. So $\eta$ is a natural isomorphism. The dual statement for $G$ and $\epsilon$ is obtained by reversing every $2$-cell. $\square$

Proposition 2 (adjoint equivalence). An adjunction $F⊣G$ is an equivalence of categories if and only if both the unit $\eta$ and the counit $\epsilon$ are natural isomorphisms. Moreover, any equivalence of categories can be improved to an adjoint equivalence (with both $\eta ,\epsilon$ invertible) by modifying the counit.

Proof. If $F⊣G$ has both $\eta ,\epsilon$ isomorphisms, then $\eta :{\mathrm{Id}}_{\mathcal{C}}\cong GF$ and $\epsilon :FG\cong {\mathrm{Id}}_{\mathcal{D}}$ are natural isomorphisms exhibiting $F,G$ as quasi-inverses, so $F,G$ form an equivalence. Conversely, suppose $F,G$ are an equivalence, so there exist natural isomorphisms $\alpha :{\mathrm{Id}}_{\mathcal{C}}\cong GF$ and $\beta :FG\cong {\mathrm{Id}}_{\mathcal{D}}$. Keep $\eta =\alpha$ as the unit. Define a new counit $\epsilon$ by transporting $\beta$ so the second triangle holds: set, componentwise, ${\epsilon }_d={\beta }_d\circ F({\alpha }_{Gd}^{-1}):FGD\to d$; this is a natural isomorphism (composite of isomorphisms). The second triangle $G{\epsilon }_d\circ {\eta }_{Gd}={\mathrm{id}}_{Gd}$ holds by construction and the naturality of $\alpha$: $G{\epsilon }_d\circ {\alpha }_{Gd}=G{\beta }_d\circ GF({\alpha }_{Gd}^{-1})\circ {\alpha }_{Gd}$, and $GF({\alpha }_{Gd}^{-1})\circ {\alpha }_{Gd}={\alpha }_{Gd}^{-1}\circ {\alpha }_{Gd}$-style cancellation (using the coherence of the original equivalence, $G\beta \circ \alpha G=\mathrm{id}$) yields the identity. The first triangle ${\epsilon }_{Fc}\circ F{\eta }_c={\mathrm{id}}_{Fc}$ then follows by cancelling the invertible $\eta$ and applying naturality — a standard triangle-from-triangle reduction. Hence $(F,G,\eta ,\epsilon )$ is an adjoint equivalence with both unit and counit invertible. $\square$

Proposition 3 (General Adjoint Functor Theorem, unit-counit form). Let $\mathcal{D}$ be locally small and complete and $G:\mathcal{D}\to \mathcal{C}$ preserve all small limits and satisfy the solution-set condition. Then there exist a functor $F:\mathcal{C}\to \mathcal{D}$ and natural transformations $\eta :{\mathrm{Id}}_{\mathcal{C}}\Rightarrow GF$, $\epsilon :FG\Rightarrow {\mathrm{Id}}_{\mathcal{D}}$ satisfying both triangle identities.

Proof. Fix $c\in \mathcal{C}$. The comma category $c\downarrow G$ has objects $(d,h:c\to Gd)$ and morphisms $t:d\to d^{\prime }$ with $Gt\circ h=h^{\prime }$. Because $G$ preserves limits, limits in $c\downarrow G$ are created from $\mathcal{D}$ (take the limit in $\mathcal{D}$; $G$-continuity supplies the structure map from $c$), so $c\downarrow G$ is complete. The solution-set condition at $c$ is a weakly initial set in $c\downarrow G$. By the lemma of 41.03.02 (a complete locally small category with a weakly initial set has an initial object), $c\downarrow G$ has an initial object; write it $(Fc,{\eta }_c:c\to GFc)$.

Define $F$ on maps: for $u:c\to c^{\prime }$, the composite ${\eta }_{c^{\prime }}\circ u:c\to GF{c}^{\prime }$ is an object of $c\downarrow G$, so by initiality of $(Fc,{\eta }_c)$ there is a unique $Fu:Fc\to F{c}^{\prime }$ with $GFu\circ {\eta }_c={\eta }_{c^{\prime }}\circ u$. Uniqueness makes $F$ a functor and $\eta :\mathrm{Id}\Rightarrow GF$ natural. For the counit, fix $d\in \mathcal{D}$: the component ${\eta }_{Gd}:Gd\to GFGd$ is an object of $Gd\downarrow G$, whose initial object is $(FGd,{\eta }_{Gd})$, so initiality gives a unique ${\epsilon }_d:FGd\to d$ with $G{\epsilon }_d\circ {\eta }_{Gd}={\mathrm{id}}_{Gd}$ — the second triangle identity. Naturality of $\epsilon$ follows from uniqueness in the defining factorization.

It remains to verify the first triangle ${\epsilon }_{Fc}\circ F{\eta }_c={\mathrm{id}}_{Fc}$. Apply $G$: $G({\epsilon }_{Fc}\circ F{\eta }_c)=G{\epsilon }_{Fc}\circ GF{\eta }_c$. By naturality of $\eta$ at ${\eta }_c:c\to GFc$, we have $GF{\eta }_c\circ {\eta }_c={\eta }_{GFc}\circ {\eta }_c$. Using the second triangle at $Fc$, $G{\epsilon }_{Fc}\circ {\eta }_{GFc}={\mathrm{id}}_{GFc}$, whence $G{\epsilon }_{Fc}\circ GF{\eta }_c\circ {\eta }_c=G{\epsilon }_{Fc}\circ {\eta }_{GFc}\circ {\eta }_c={\eta }_c$. So $G({\epsilon }_{Fc}\circ F{\eta }_c)\circ {\eta }_c={\eta }_c={\mathrm{id}}_{GFc}\circ {\eta }_c=G({\mathrm{id}}_{Fc})\circ {\eta }_c$. Initiality of $(Fc,{\eta }_c)$ in $c\downarrow G$ makes the factorization through ${\eta }_c$ unique, so ${\epsilon }_{Fc}\circ F{\eta }_c={\mathrm{id}}_{Fc}$. Both triangles hold, so $(F,G,\eta ,\epsilon )$ is unit-counit adjunction data. $\square$

Proposition 4 (adjoint existence is representability). A functor $G:\mathcal{D}\to \mathcal{C}$ has a left adjoint if and only if for each $c\in \mathcal{C}$ the functor $\mathcal{C}(c,G-):\mathcal{D}\to \mathbf{Set}$ is representable.

Proof. ($\Rightarrow$) If $F⊣G$, then $\mathcal{C}(c,Gd)\cong \mathcal{D}(Fc,d)=\mathcal{D}(Fc,-)(d)$ naturally in $d$, so $\mathcal{C}(c,G-)$ is represented by $Fc$ with universal element ${\eta }_c\in \mathcal{C}(c,GFc)$ corresponding to ${\mathrm{id}}_{Fc}$ under the bijection — exactly the unit.

($\Leftarrow$) Suppose each $\mathcal{C}(c,G-)$ is represented by some object $F_c\in \mathcal{D}$ with universal element ${\eta }_c\in \mathcal{C}(c,G{F}_c)$. Representability gives, for each $d$, a natural bijection $\mathcal{D}(F_c,d)\cong \mathcal{C}(c,Gd)$ carrying ${\mathrm{id}}_{F_c}$ to ${\eta }_c$. Set $Fc=F_c$. For $u:c\to c^{\prime }$, the natural transformation $\mathcal{C}(c^{\prime },G-)\stackrel{u^{\ast }}{\to }\mathcal{C}(c,G-)$ between representables corresponds, by the Yoneda lemma (41.04.02), to a unique map $Fu:Fc\to F{c}^{\prime }$ with $\mathcal{D}(Fu,d)$ realizing $u^{\ast }$; functoriality of $F$ follows from uniqueness. The resulting family of bijections $\mathcal{D}(Fc,d)\cong \mathcal{C}(c,Gd)$ is natural in $c$ and $d$, exhibiting $F⊣G$. $\square$

Connections Master

• Adjunctions: hom-set and unit-counit 41.03.01. The definitions of unit, counit, triangle identities, and the equivalence of the hom-set and unit-counit presentations are established in 41.03.01; the present unit is the equational and diagrammatic calculus that operates on those definitions, treating $\eta$ and $\epsilon$ as generators of a rewriting system whose relations are the two triangles.

• RAPL, reflective subcategories, and the adjoint functor theorems 41.03.02. The General and Special Adjoint Functor Theorems are proved in 41.03.02 by constructing initial objects of comma categories and reading off the hom-set bijection; this unit reconstructs the General theorem by directly building the unit and counit and verifying the triangle identities, a complementary proof that exposes the theorem as a statement about existence of unit-counit data. RAPL, proved there, is the necessity half read here as a single mate calculation.

• Limits, colimits, and their construction 41.02.01, 41.02.02. Limits are themselves adjoints to the diagonal functor $\lim ⊣\Delta ⊣\mathrm{colim}$, so the unit-counit calculus applies to cones and cocones uniformly; the completeness of $\mathcal{D}$ used in Proposition 3, and the products-and-equalizers reduction of 41.02.02, are what make the comma category $c\downarrow G$ complete enough to host an initial object.

• Representable functors and the Yoneda lemma 41.04.01, 41.04.02. Proposition 4 reduces adjoint existence to representability of the functors $\mathcal{C}(c,G-)$, and the mate correspondence is the two-dimensional shadow of Yoneda: transposing a natural transformation across $F⊣G$ is a representable-functor bijection applied componentwise, with uniqueness of representations assembling the local data into a global adjoint.

• Monads and their algebras 41.05.01. Every adjunction $F⊣G$ generates the monad $GF$ with unit $\eta$ and multiplication $G\epsilon F$; the triangle identities verified throughout this unit become the monad's unit and associativity laws, and the calculus of mates governs monad morphisms and the distributive laws between monads developed there.

• Kan extensions 41.06.02. Pointwise Kan extensions are computed by mate calculations, and a left adjoint is the left Kan extension of the identity along its right adjoint; the formulas of 41.06.02 are the transpose-and-cancel manipulations of this unit applied to the defining $2$-cells of an extension.

• Homological algebra and derived adjunctions 01.06.01. The tensor-hom adjunction and its derived counterparts underlie derived functors and the long exact sequences of 01.06.01; the unit-counit calculus is the syntactic backbone of the adjunctions between derived categories, where mates become the natural transformations governing connecting homomorphisms.

Historical & philosophical context Master

Daniel Kan introduced adjoint functors in 1958 in Adjoint functors ^{[Kan 1958]}, where the hom-set bijection $\mathcal{D}(Fc,d)\cong \mathcal{C}(c,Gd)$ is the defining condition and the term "adjoint" is borrowed, by analogy with adjoint operators on Hilbert spaces, from the hom-functor's role as a categorical inner product. The recognition that the unit and counit carry the same data, and that the triangle identities are the operative equations, was systematized by Saunders Mac Lane in Categories for the Working Mathematician (1971; 2nd ed. 1998) ^{[Mac Lane 1998]}, Chapter IV, where fully faithful adjunctions, composed adjunctions, and the reflection of equivalences as adjunctions with invertible unit and counit are given their standard form.

Peter Freyd's adjoint functor theorems appeared in his 1964 Abelian Categories ^{[Freyd 1964]}, casting the existence of a left adjoint as a completeness-plus-smallness condition — the solution-set condition being the precise set-theoretic hypothesis separating "preserves limits" from "has a left adjoint." The calculus of mates, phrased as a bijection of $2$-cells under adjunctions internal to a $2$-category, was developed by Max Kelly in his work on enriched category theory ^{[Kelly 1982]}, and the string-diagram reformulation in which the triangle identities become the zigzag identities is due to André Joyal and Ross Street ^{[Joyal-Street 1991]}, giving the calculus its topological face: adjunctions are the categorical avatar of duality, the cap and cup of knot theory, and the straightening of a zigzag is the move that underlies every manipulation above.

Bibliography Master

@article{Kan1958calc,
  author  = {Kan, Daniel M.},
  title   = {Adjoint functors},
  journal = {Transactions of the American Mathematical Society},
  volume  = {87},
  year    = {1958},
  pages   = {294--329}
}

@book{MacLane1998calc,
  author    = {Mac Lane, Saunders},
  title     = {Categories for the Working Mathematician},
  edition   = {2},
  publisher = {Springer},
  series    = {Graduate Texts in Mathematics 5},
  year      = {1998}
}

@book{Freyd1964calc,
  author    = {Freyd, Peter},
  title     = {Abelian Categories: An Introduction to the Theory of Functors},
  publisher = {Harper and Row},
  year      = {1964}
}

@book{Riehl2016calc,
  author    = {Riehl, Emily},
  title     = {Category Theory in Context},
  publisher = {Dover},
  year      = {2016}
}

@book{Awodey2010calc,
  author    = {Awodey, Steve},
  title     = {Category Theory},
  edition   = {2},
  publisher = {Oxford University Press},
  series    = {Oxford Logic Guides 52},
  year      = {2010}
}

@book{Kelly1982,
  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}
}

@article{JoyalStreet1991,
  author  = {Joyal, Andr\'e and Street, Ross},
  title   = {The geometry of tensor calculus, I},
  journal = {Advances in Mathematics},
  volume  = {88},
  number  = {1},
  year    = {1991},
  pages   = {55--112}
}