Adjunctions: Hom-Set and Unit-Counit Definitions

Intuition Beginner

A great many constructions in mathematics are best approximations. You have an object of one kind, and you want the most economical object of another kind built on top of it, one that wastes nothing and adds nothing it does not have to. Take a set of letters and build the most economical group you can from them, inventing only the products and inverses the group laws force. That output is the free group on those letters. The word "free" means free of extra relations: nothing holds among the letters except what every group must obey.

Now notice a second kind of object running alongside the first. A group can always be looked at as a plain set, by forgetting its multiplication. So there are two directions in play. One direction builds a group out of a set; the other strips a group back down to a set. An adjunction is the precise statement that these two directions fit together perfectly. It says that the maps out of the freely built group match up, one for one, with the maps out of the original set into any group you choose as a target.

That matching is the whole content. To send the free group somewhere, you only need to say where the original letters go, and after that the group laws carry the rest along with no further choices. Building the cheapest thing and naming where the raw ingredients land are two views of one situation.

Visual Beginner

Picture two columns. On the left is the world of sets; on the right is the world of groups. The "build" direction takes a set on the left and produces a free group on the right. The "forget" direction takes a group on the right and produces its underlying set on the left.

An adjunction lines up two collections of maps so that they are the same collection counted twice. One collection is the group maps out of the freely built group; the other is the set maps from your original set into the forgotten version of the target group. Every map in one collection corresponds to exactly one map in the other, and the matching is consistent as you vary the target.

feature	the build direction (free)	the forget direction (underlying)
starts in	sets	groups
ends in	groups	sets
role in the pair	the left side	the right side
what it optimises	cheapest group on a set	plain set under a group
matching of maps	group maps out of the free group	set maps into the forgotten set

The balance in the picture is the point. Neither side is privileged; the correspondence reads the same in both directions, and that even-handed matching is what the word adjunction names.

Worked example Beginner

Work with the free group on a single letter, $x$. Building the cheapest group on one letter gives all powers of $x$: the elements $\dots ,x^{-2},x^{-1},1,x,x^2,\dots$, with multiplication by adding exponents. This is the group of integers under addition, with $x$ playing the role of the number $1$.

Step 1. Pick a target group to map into. Take the group with two elements, written $0$ and $1$, where adding $1$ to $1$ loops back to $0$. Call it the flip group.

Step 2. To give a group map out of the free group on $x$, the matching says you only choose where the single letter $x$ lands. There are exactly two choices: send $x$ to $0$, or send $x$ to $1$. Each choice extends to a full group map in one and only one way, because once $x$ has a home, every power $x^n$ must go to the $n$-fold sum of that home.

Step 3. Count the maps the other way. A set map from the one-letter set into the underlying set of the flip group also has exactly two options: send the letter to $0$, or send it to $1$. Two and two.

What this tells us: the group maps out of the free group are matched one-for-one with the set maps from the letters into the target. Choosing a group homomorphism is the same labour as choosing where the raw letters go. That exact one-for-one matching, holding for every target group at once, is the adjunction between building and forgetting.

Check your understanding Beginner

Formal definition Intermediate+

Let $F:\mathcal{C}\to \mathcal{D}$ and $G:\mathcal{D}\to \mathcal{C}$ be functors. An adjunction $F⊣G$ ($F$ left adjoint to $G$, $G$ right adjoint to $F$) is a family of bijections

$${\phi }_{c,d}:\mathcal{D}(Fc,d)\stackrel{\cong }{\to }\mathcal{C}(c,Gd)$$

one for each pair of objects $c\in \mathcal{C}$, $d\in \mathcal{D}$, natural in $c$ and $d$ jointly ^{[Mac Lane 1998]}. Naturality means: for $u:c^{\prime }\to c$ in $\mathcal{C}$ and $v:d\to d^{\prime }$ in $\mathcal{D}$ and any $g:Fc\to d$,

$${\phi }_{c^{\prime },d^{\prime }}(v\circ g\circ Fu)=Gv\circ {\phi }_{c,d}(g)\circ u.$$

Equivalently, the two functors $(c,d)\mapsto \mathcal{D}(Fc,d)$ and $(c,d)\mapsto \mathcal{C}(c,Gd)$ from ${\mathcal{C}}^{\mathrm{op}}\times \mathcal{D}$ to $\mathbf{Set}$ are naturally isomorphic. The bijection $\phi$ is the adjunction isomorphism, and a morphism and its image $\phi (g)$ are adjoint transposes (or mates), written $\hat{g}$ when the indices are clear. Notation introduced here ($⊣$ for adjunction, $\phi$ for the transpose bijection, $\eta$ and $\epsilon$ below) is recorded in _meta/NOTATION.md.

Definition (unit and counit). From the hom-set data, set $d=Fc$ and transpose the identity: the unit is ${\eta }_c={\phi }_{c,Fc}({\mathrm{id}}_{Fc}):c\to GFc$. Dually, set $c=Gd$ and transpose the identity backwards: the counit is ${\epsilon }_d={\phi }_{Gd,d}^{-1}({\mathrm{id}}_{Gd}):FGd\to d$. Naturality of $\phi$ makes $\eta :{\mathrm{Id}}_{\mathcal{C}}\Rightarrow GF$ and $\epsilon :FG\Rightarrow {\mathrm{Id}}_{\mathcal{D}}$ natural transformations. The transpose of a general $g:Fc\to d$ is recovered as $\phi (g)=Gg\circ {\eta }_c$, and its inverse as ${\phi }^{-1}(f)={\epsilon }_d\circ Ff$.

Definition (unit-counit adjunction; triangle identities). A unit-counit adjunction is a quadruple $(F,G,\eta ,\epsilon )$ with $\eta :{\mathrm{Id}}_{\mathcal{C}}\Rightarrow GF$ and $\epsilon :FG\Rightarrow {\mathrm{Id}}_{\mathcal{D}}$ natural transformations satisfying the triangle identities

$$(\epsilon F)\circ (F\eta )={\mathrm{id}}_F,(G\epsilon )\circ (\eta G)={\mathrm{id}}_G,$$

equalities of natural transformations $F\Rightarrow F$ and $G\Rightarrow G$. Componentwise the first reads ${\epsilon }_{Fc}\circ F{\eta }_c={\mathrm{id}}_{Fc}$ for every $c$, the second $G{\epsilon }_d\circ {\eta }_{Gd}={\mathrm{id}}_{Gd}$ for every $d$.

Definition (universal arrow). Given $G:\mathcal{D}\to \mathcal{C}$ and an object $c\in \mathcal{C}$, a universal arrow from $c$ to $G$ is a pair $(Fc,{\eta }_c)$ with $Fc\in \mathcal{D}$ and ${\eta }_c:c\to GFc$ such that every $f:c\to Gd$ factors as $f=Gg\circ {\eta }_c$ for a unique $g:Fc\to d$. This says $(Fc,{\eta }_c)$ is an initial object of the comma category $c\downarrow G$, whose objects are pairs $(d,c\to Gd)$. A left adjoint of $G$ is exactly a choice of universal arrow from each $c$.

The corpus is full of adjunctions. Each free-forgetful pair is one: free monoid $⊣$ forgetful $\mathbf{Mon}\to \mathbf{Set}$; free group $⊣$ forgetful $\mathbf{Grp}\to \mathbf{Set}$; free $R$-module $⊣$ forgetful $R\text{-}\mathbf{Mod}\to \mathbf{Set}$; free vector space $⊣$ forgetful ${\mathbf{Vect}}_k\to \mathbf{Set}$; free abelian group (the integral group ring on a set) $⊣$ forgetful $\mathbf{Ab}\to \mathbf{Set}$; tensor algebra $⊣$ forgetful. In topology the discrete-space functor is left adjoint and the indiscrete-space functor right adjoint to the underlying-set functor $\mathbf{Top}\to \mathbf{Set}$, giving a chain $\mathrm{disc}⊣|-|⊣\mathrm{indisc}$. Abelianization $(-{)}^{\mathrm{ab}}⊣$ inclusion $\mathbf{Ab}\hookrightarrow \mathbf{Grp}$; the Stone–Čech compactification $\beta ⊣$ inclusion of compact Hausdorff spaces into Tychonoff spaces. For a fixed module $M$ the tensor product $(-){\otimes }_{R}M⊣{\mathrm{Hom}}_R(M,-)$ is the tensor-hom adjunction.

Counterexamples to common slips Intermediate+

• Not every functor with both a notion of "free" and "underlying" gives an adjunction. The naturality of $\phi$ in both variables is a real constraint; a bijection of hom-sets that is not natural fails to come from a unit and counit. The field-of-fractions construction sends an integral domain to a field but has no left or right adjoint as a functor on all rings, because the required factorization fails.

• Left and right adjoints are not symmetric roles. If $F⊣G$, it does not follow that $G⊣F$. The free-group functor is left adjoint to the forgetful functor, but the forgetful functor is not left adjoint to the free-group functor — there is no functor making the hom-sets match in the reversed order.

• The triangle identities are two equations, not one. A pair $(\eta ,\epsilon )$ satisfying only one triangle is an adjunction up to one side; both are needed for $\phi (g)=Gg\circ {\eta }_c$ and ${\phi }^{-1}(f)={\epsilon }_d\circ Ff$ to be mutually inverse. A bare unit with no counit gives only a reflection onto a subcategory when $G$ is fully faithful.

Key theorem with proof Intermediate+

The signature result is that the hom-set presentation and the unit-counit presentation carry the same information.

Theorem (equivalence of the two definitions). Let $F:\mathcal{C}\to \mathcal{D}$ and $G:\mathcal{D}\to \mathcal{C}$ be functors. The following data are interchangeable, each determining the other ^{[Mac Lane 1998]}:

(i) a natural family of bijections ${\phi }_{c,d}:\mathcal{D}(Fc,d)\cong \mathcal{C}(c,Gd)$;

(ii) natural transformations $\eta :{\mathrm{Id}}_{\mathcal{C}}\Rightarrow GF$ and $\epsilon :FG\Rightarrow {\mathrm{Id}}_{\mathcal{D}}$ satisfying the triangle identities.

Proof. (i) $\Rightarrow$ (ii). Define ${\eta }_c={\phi }_{c,Fc}({\mathrm{id}}_{Fc})$ and ${\epsilon }_d={\phi }_{Gd,d}^{-1}({\mathrm{id}}_{Gd})$. Naturality of $\phi$ in $c$, applied to ${\mathrm{id}}_{Fc}$ with a map $u:c^{\prime }\to c$, gives $\phi (Fu)=GFu\circ {\eta }_{c^{\prime }}$ on one side and ${\eta }_c\circ u$ on the other after transposing, so $GFu\circ {\eta }_{c^{\prime }}={\eta }_c\circ u$: the unit is natural. Dually $\epsilon$ is natural. For a general $g:Fc\to d$, naturality in $d$ along $g:Fc\to d$ applied to ${\mathrm{id}}_{Fc}$ yields $\phi (g)=\phi (g\circ {\mathrm{id}}_{Fc})=Gg\circ \phi ({\mathrm{id}}_{Fc})=Gg\circ {\eta }_c$. Setting $g={\epsilon }_d$ with $c=Gd$ gives $\phi ({\epsilon }_d)=G{\epsilon }_d\circ {\eta }_{Gd}$; but $\phi ({\epsilon }_d)=\phi ({\phi }^{-1}({\mathrm{id}}_{Gd}))={\mathrm{id}}_{Gd}$, which is the second triangle identity. Symmetrically, ${\phi }^{-1}(f)={\epsilon }_d\circ Ff$ and putting $f={\eta }_c$ gives the first triangle identity ${\epsilon }_{Fc}\circ F{\eta }_c={\mathrm{id}}_{Fc}$.

(ii) $\Rightarrow$ (i). Given $\eta ,\epsilon$ with the two triangles, define $\phi (g)=Gg\circ {\eta }_c$ for $g:Fc\to d$ and $\psi (f)={\epsilon }_d\circ Ff$ for $f:c\to Gd$. Both are natural in $c$ and $d$ because $\eta ,\epsilon$ are. They are mutually inverse: $$ \psi(\varphi(g))=\varepsilon_d\circ F(Gg\circ\eta_c)=\varepsilon_d\circ FGg\circ F\eta_c =g\circ\varepsilon_{Fc}\circ F\eta_c=g\circ\mathrm{id}_{Fc}=g, $$ using naturality of $\epsilon$ at $g$ (to move $\epsilon$ past $FGg$) and then the first triangle. The computation $\phi (\psi (f))=f$ is dual, using naturality of $\eta$ and the second triangle. Hence $\phi$ is a natural bijection. The two constructions are mutually inverse by the displayed formulas, so the data sets coincide. $\square$

Bridge. This theorem builds toward the entire apparatus of adjoint functors and appears again in 41.05.01, where the composite $GF$ equipped with $\eta$ and $G\epsilon F$ is shown to be the monad of the adjunction, and the triangle identities become the monad's unit and associativity laws. The foundational reason the two presentations agree is that an adjunction is determined by what it does to identities: transposing ${\mathrm{id}}_{Fc}$ produces the unit and transposing ${\mathrm{id}}_{Gd}$ produces the counit, and naturality forces every other transpose to be reconstructed from these by the formula $\phi (g)=Gg\circ {\eta }_c$. This is exactly the move that the universal-arrow presentation packages geometrically, where ${\eta }_c$ is the initial object of the comma category $c\downarrow G$; putting these together, the central insight is that "left adjoint", "natural hom-set bijection", "universal arrow from each object", and "unit-counit with triangles" are four names for one structure, and the bridge is that uniqueness of adjoints — proved in the Master tier via Yoneda and cross-referenced to 41.04.02 — follows the instant any one presentation is pinned down.

Exercises Intermediate+

Advanced results Master

Theorem (uniqueness of adjoints). If $G:\mathcal{D}\to \mathcal{C}$ has left adjoints $F$ and $F^{\prime }$, then $F\cong F^{\prime }$ by a unique natural isomorphism compatible with the units. Dually, right adjoints are unique up to unique natural isomorphism. The proof is an application of the Yoneda lemma: for each $c$, the representable $\mathcal{D}(Fc,-)\cong \mathcal{C}(c,G-)\cong \mathcal{D}(F^{\prime }c,-)$ are naturally isomorphic, so $Fc\cong F^{\prime }c$ by Yoneda, and the isomorphisms are natural in $c$. The full Yoneda argument is developed in 41.04.02; uniqueness is the first corollary that representability buys.

Theorem (composition of adjunctions). If $F⊣G$ with unit $\eta$, counit $\epsilon$ and $F^{\prime }⊣G^{\prime }$ with unit ${\eta }^{\prime }$, counit ${\epsilon }^{\prime }$, where $F:\mathcal{C}\to \mathcal{D}$, $F^{\prime }:\mathcal{D}\to \mathcal{E}$, then $F^{\prime }F⊣G{G}^{\prime }$, with unit $G{\eta }_F^{\prime }\circ \eta$ and counit ${\epsilon }^{\prime }\circ F^{\prime }{\epsilon }_{G^{\prime }}$. The hom-set bijection is the composite $\mathcal{E}(F^{\prime }Fc,e)\cong \mathcal{D}(Fc,G^{\prime }e)\cong \mathcal{C}(c,G{G}^{\prime }e)$. Adjunctions therefore compose, and identity functors are self-adjoint, so categories, functors, and adjunctions assemble into a $2$-category-like structure (a double category when both directions are tracked); left adjoints compose to left adjoints.

Theorem (RAPL / LAPC: right adjoints preserve limits, left adjoints preserve colimits). Let $G:\mathcal{D}\to \mathcal{C}$ be a right adjoint, $F⊣G$. If $D:\mathcal{J}\to \mathcal{D}$ has a limit, then $G$ preserves it: $G({\lim }_j{D}_j)\cong {\lim }_{j}G(D_j)$, the canonical comparison being an isomorphism. Dually a left adjoint preserves all colimits that exist. The mechanism is hom-set: $\mathcal{C}(c,G{\lim }_j{D}_j)\cong \mathcal{D}(Fc,{\lim }_j{D}_j)\cong {\lim }_j\mathcal{D}(Fc,D_j)\cong {\lim }_j\mathcal{C}(c,G{D}_j)\cong \mathcal{C}(c,{\lim }_{j}G{D}_j)$, natural in $c$, so the two objects are isomorphic by Yoneda. The detailed proof and the partial converse — that a continuous functor between suitable categories has a left adjoint (the adjoint functor theorems) — are developed in 41.03.02.

Theorem (Galois connections and reflective subcategories as adjunctions). A Galois connection between posets $P,Q$ — order-preserving $f:P\to Q$, $g:Q\to P$ with $f(p)\le q⟺p\le g(q)$ — is exactly an adjunction $f⊣g$ between the posets viewed as categories, the hom-sets being at most one element. A reflective subcategory $\mathcal{A}\hookrightarrow \mathcal{B}$ is one whose inclusion has a left adjoint (the reflector); abelianization, sheafification, the Stone–Čech compactification, and completion of a metric space are all reflectors, and the counit is an isomorphism precisely because the inclusion is fully faithful.

Synthesis. Putting these together, an adjunction is the single most pervasive organising structure in the categorical corpus: every free construction, every forgetful functor with a best-approximation partner, every Galois connection, and every reflective localization is one instance. The foundational reason adjoints are unique is that each is determined by a representability condition, and representable functors are rigid — Yoneda turns the hom-set bijection into an isomorphism of objects, which is why uniqueness, composition, and preservation all reduce to manipulations of $\mathcal{C}(c,G-)\cong \mathcal{D}(Fc,-)$. This generalises the equivalence theory of 41.01.02: an equivalence is an adjunction whose unit and counit are isomorphisms (an adjoint equivalence), so the trichotomy characterisation is the degenerate case where both triangles collapse to invertibility. The central insight is that limits and colimits, defined up to isomorphism in 41.02.01, are preserved by the correct half of any adjoint pair — right adjoints carry limits, left adjoints carry colimits — and this is exactly the hom-set calculation read in two variables. The bridge is that the composite $GF$ of an adjoint pair, with $\eta$ and $G\epsilon F$, is a monad, opening the algebra of 41.05.01, while the universal-arrow presentation ties every adjunction back to initial objects in comma categories, the same comma machinery the Yoneda development of 41.04.02 uses to make representability geometric.

Full proof set Master

Proposition 1 (the transpose formulas determine each other). In an adjunction with unit $\eta$ and counit $\epsilon$, the maps $\phi (g)=Gg\circ {\eta }_c$ and $\psi (f)={\epsilon }_d\circ Ff$ are mutually inverse natural bijections, and conversely a natural bijection $\phi$ recovers $\eta ,\epsilon$ as transposes of identities.

Proof. Assume $\eta ,\epsilon$ natural with both triangle identities. For $g:Fc\to d$, $$ \psi(\varphi(g))=\varepsilon_d\circ F(Gg\circ\eta_c)=\varepsilon_d\circ FGg\circ F\eta_c=g\circ\varepsilon_{Fc}\circ F\eta_c=g, $$ where the third equality is naturality of $\epsilon$ at $g:Fc\to d$ (which gives ${\epsilon }_d\circ FGg=g\circ {\epsilon }_{Fc}$) and the last is the first triangle identity. For $f:c\to Gd$, $$ \varphi(\psi(f))=G(\varepsilon_d\circ Ff)\circ\eta_c=G\varepsilon_d\circ GFf\circ\eta_c=G\varepsilon_d\circ\eta_{Gd}\circ f=f, $$ by naturality of $\eta$ at $f$ and then the second triangle identity. Both $\phi ,\psi$ are natural since they are built by post- and pre-composition with the natural $\eta ,\epsilon$. Conversely, given a natural bijection $\phi$, define ${\eta }_c=\phi ({\mathrm{id}}_{Fc})$ and ${\epsilon }_d={\phi }^{-1}({\mathrm{id}}_{Gd})$; naturality of $\phi$ in $d$ along $g$ gives $\phi (g)=\phi (g\circ \mathrm{id})=Gg\circ {\eta }_c$, recovering the first formula, and dually the second. $\square$

Proposition 2 (uniqueness of adjoints via Yoneda). If $F⊣G$ and $F^{\prime }⊣G$ share the right adjoint $G$, there is a unique natural isomorphism $\theta :F\cong F^{\prime }$ with $G\theta \circ \eta ={\eta }^{\prime }$.

Proof. For each $c$ and all $d$, the adjunctions give natural isomorphisms of functors $\mathcal{D}(Fc,-)\cong \mathcal{C}(c,G-)\cong \mathcal{D}(F^{\prime }c,-)$ in the variable $d$. By the Yoneda lemma, a natural isomorphism between representable functors $\mathcal{D}(Fc,-)\cong \mathcal{D}(F^{\prime }c,-)$ is induced by a unique isomorphism ${\theta }_c:F^{\prime }c\to Fc$ (Yoneda is contravariant on the representing object), equivalently ${\theta }_c^{-1}:Fc\cong F^{\prime }c$. Naturality of the hom-set isomorphisms in $c$ forces $\theta$ to be a natural transformation, hence a natural isomorphism. The identity $G\theta \circ \eta ={\eta }^{\prime }$ holds because both transpose to $\mathrm{id}$ under the respective bijections, and Yoneda makes the inducing isomorphism unique. $\square$

Proposition 3 (right adjoints preserve limits). If $F⊣G$ and $D:\mathcal{J}\to \mathcal{D}$ has limit $\lim D$, then $G(\lim D)$ is a limit of $GD$, with the canonical comparison an isomorphism.

Proof. For every $c\in \mathcal{C}$, natural in $c$, $$ \mathcal{C}(c,G\lim D)\cong\mathcal{D}(Fc,\lim D)\cong\lim_{j}\mathcal{D}(Fc,D_j)\cong\lim_{j}\mathcal{C}(c,GD_j)\cong\mathcal{C}(c,\lim_j GD_j), $$ where the first and third isomorphisms are the adjunction, the second is the defining universal property of the limit in $\mathcal{D}$ (representable functors send limits to limits of sets), and the fourth is again the limit's universal property in $\mathcal{C}$, provided ${\lim }_{j}G{D}_j$ exists; when it does not a priori, the chain exhibits $G\lim D$ as representing the functor $c\mapsto {\lim }_j\mathcal{C}(c,G{D}_j)$, so it is the limit. The composite isomorphism is natural in $c$ and sends $\mathrm{id}$ to the canonical comparison, so by Yoneda the comparison is an isomorphism. $\square$

Proposition 4 (composition of adjunctions is an adjunction). With $F⊣G$, $F^{\prime }⊣G^{\prime }$ composable as above, $F^{\prime }F⊣G{G}^{\prime }$, and the unit and counit are the stated composites.

Proof. The hom-set bijection is the composite of natural bijections $$ \mathcal{E}(F'Fc,e)\xrightarrow{\ \varphi'\ }\mathcal{D}(Fc,G'e)\xrightarrow{\ \varphi\ }\mathcal{C}(c,GG'e), $$ natural in $c$ and $e$ as a composite of natural maps. Transposing ${\mathrm{id}}_{F^{\prime }Fc}$ gives the unit: $\phi ({\phi }^{\prime }(\mathrm{id}))=\phi ({\eta }_{Fc}^{\prime })=G{\eta }_{Fc}^{\prime }\circ {\eta }_c=(G{\eta }_F^{\prime }\circ \eta {)}_c$. Transposing ${\mathrm{id}}_{G{G}^{\prime }e}$ backwards gives the counit ${\epsilon }_e^{\prime }\circ F^{\prime }{\epsilon }_{G^{\prime }e}=({\epsilon }^{\prime }\circ F^{\prime }{\epsilon }_{G^{\prime }}{)}_e$. The triangle identities for the composite follow from those of the two factors together with naturality, by Proposition 1 applied to the composite bijection. $\square$

Connections Master

• Natural transformations and equivalence 41.01.02. An adjunction is by definition a pair of natural transformations — the unit and counit — living in the functor categories built in 41.01.02, and the triangle identities are equations between vertical and whiskered composites of exactly the kind catalogued there. An adjoint equivalence is the special adjunction whose unit and counit are natural isomorphisms, so the full-faithful-essentially-surjective characterisation of equivalences is the degenerate case of adjointness where both triangles witness invertibility.

• Limits and colimits 41.02.01. Right adjoints preserve limits and left adjoints preserve colimits, established here as a hom-set calculation and proved in full in 41.03.02; conversely, a limit of $D$ can be described as the value of a right adjoint to the constant-diagram functor $\Delta :\mathcal{D}\to [\mathcal{J},\mathcal{D}]$, so $\lim ⊣\Delta ⊣\mathrm{colim}$ exhibits limits and colimits themselves as adjoints to the diagonal. The universal cones of 41.02.01 are thus instances of the universal arrows of this unit.

• The Yoneda lemma and representability 41.04.02. Uniqueness of adjoints is a corollary of Yoneda — a left adjoint represents the functor $d\mapsto \mathcal{C}(c,Gd)$, and representable functors determine their representing object up to unique isomorphism. The universal-arrow presentation here is the same comma-category machinery ${\int }_{\mathcal{C}}F\cong (y\downarrow F)$ that makes representability geometric in 41.04.02.

• Monads and their algebras 41.05.01. Every adjunction $F⊣G$ generates a monad $GF$ on $\mathcal{C}$ with unit $\eta$ and multiplication $G\epsilon F$, and dually a comonad $FG$ on $\mathcal{D}$; the triangle identities of this unit become the monad's unit laws. The Eilenberg–Moore and Kleisli constructions of 41.05.01 show every monad arises from an adjunction, closing the loop opened here.

• Adjoint functor theorems and RAPL 41.03.02. The preservation theorem stated and sketched here is proved there, together with its partial converse: a limit-preserving functor between locally small, complete categories satisfying a solution-set condition has a left adjoint. That unit also formalizes RAPL/LAPC and the General and Special Adjoint Functor Theorems referenced in this unit's Advanced results.

Historical & philosophical context Master

Adjoint functors were introduced by Daniel Kan in 1958 in Adjoint functors ^{[Kan 1958]}, where the hom-set bijection $\mathcal{D}(Fc,d)\cong \mathcal{C}(c,Gd)$ appears as the defining condition and the term "adjoint" is borrowed by analogy with adjoint operators on Hilbert spaces, the hom-functor playing the role of the inner product. Kan's paper also introduced what are now called Kan extensions, and the two notions are intertwined: an adjoint is a special Kan extension. The recognition that free constructions, tensor-hom duality, and limit-colimit preservation were all the same phenomenon followed quickly and reorganized large parts of algebra and topology around the single notion.

The unit-counit formulation with the triangle identities, and the equivalence with the hom-set definition, were systematized in Mac Lane's Categories for the Working Mathematician (1971; 2nd ed. 1998) ^{[Mac Lane 1998]}, Chapter IV, alongside the universal-arrow presentation and the slogan, often attributed to the categorical tradition, that adjoint functors arise everywhere. The poset case had been known much earlier as the Galois connection, studied by Évariste Galois in the correspondence between subgroups and subfields and abstracted by Garrett Birkhoff in lattice theory in the 1930s; the categorical definition subsumes it as the case where all hom-sets have at most one element. Peter Freyd's adjoint functor theorems, giving conditions under which a left adjoint exists, appeared in his 1964 Abelian Categories and made adjointness an existence question answerable by the solution-set condition.

Bibliography Master

@article{Kan1958,
  author  = {Kan, Daniel M.},
  title   = {Adjoint functors},
  journal = {Transactions of the American Mathematical Society},
  volume  = {87},
  year    = {1958},
  pages   = {294--329}
}

@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{Borceux1994,
  author    = {Borceux, Francis},
  title     = {Handbook of Categorical Algebra 1: Basic Category Theory},
  publisher = {Cambridge University Press},
  series    = {Encyclopedia of Mathematics and its Applications 50},
  year      = {1994}
}

@book{Freyd1964,
  author    = {Freyd, Peter},
  title     = {Abelian Categories: An Introduction to the Theory of Functors},
  publisher = {Harper and Row},
  year      = {1964}
}