41.03.02 · category-theory / adjunctions

RAPL, Reflective Subcategories, and the Adjoint Functor Theorems

shipped3 tiersLean: none

Anchor (Master): Mac Lane 1998 *Categories for the Working Mathematician* 2e (Springer GTM 5) Ch. V in full (preservation, the two adjoint functor theorems, reflective subcategories, Kan extensions as the limit form of adjoints); Borceux 1994 *Handbook of Categorical Algebra 1* (Cambridge) §3.3 (the adjoint functor theorems), Ch. 5 (reflective subcategories and localizations); Adámek-Rosický 1994 *Locally Presentable and Accessible Categories* (Cambridge LMS 189) Ch. 1-2 (the AFT in the locally presentable setting); Freyd 1964 *Abelian Categories* Ch. 3

Intuition Beginner

When you build the cheapest object of one kind on top of another, the building step has a partner that strips structure back down. The earlier unit on adjunctions named that partnership. This unit asks two follow-up questions. First: what does the stripping-down step do to "best objects over a pattern" — the meeting points, the things that sit consistently over a diagram? The answer is clean. It always carries a meeting point to a meeting point. Forgetting a group to its set, then taking the shared object, gives the same set as taking the shared object first and forgetting after. The order of operations loses nothing.

The second question is the reverse. Suppose you only have the stripping-down step and you want to know whether a cheapest-building partner even exists. Not every forgetting has a builder. So you need a test. The test says: if the forgetting respects all meeting points, and you can find a small enough catalogue of candidate approximations for each object, then a builder is guaranteed.

A third idea ties it together. Sometimes the smaller world sits inside the bigger world, and the builder pushes each object to its best version inside the smaller world. Abelianizing a group, completing a space with all its limits, turning a rough function-assignment into a genuine local-to-global one — each is a best-version map into a smaller world that lives inside a larger one.

Visual Beginner

Picture the forgetting step as a map from a richer world on the left to a plainer world on the right. Over in the richer world, draw a small pattern of objects with arrows and the best object sitting over it. The claim is that you can compute the best object two ways and get the same answer: take the best object first and then forget it, or forget every piece first and then take the best object on the right. The two routes meet.

route	step one	step two	result
forget last	take the best object over the pattern	forget that one object	a plain object
forget first	forget every piece of the pattern	take the best object on the right	the same plain object

The two routes give the same plain object, so the square commutes. That is the whole content of "the right-hand partner respects best objects."

The nested picture below the square is the reflective idea: a smaller world living inside a larger one, with an arrow sending each large object to its best version inside the small world.

Worked example Beginner

Take the forgetting step from groups to sets, and check that it respects a shared object. Use the simplest shared object: the meeting of two groups along nothing, which is their product. Take the two-element flip group, with elements $0$ and $1$ , and the three-element rotation group, with elements $0, 1, 2$ that add and wrap around.

Step 1. Form the product group first. Its elements are all pairs $(a, b)$ with $a$ from the flip group and $b$ from the rotation group. Counting, there are $2 \times 3 = 6$ pairs. Now forget down to the underlying set: a six-element set of pairs.

Step 2. Now do it the other order. Forget each group first: the flip group becomes a two-element set, the rotation group becomes a three-element set. Take the product of these two plain sets: all pairs of one element from each. That is again $2 \times 3 = 6$ pairs.

Step 3. Compare. Both routes give the same six pairs, matched name for name. The order did not matter.

What this tells us: forgetting a product of groups gives the product of the forgotten sets. The right-hand partner — the forgetting step — carried the shared object across without distortion. This is one small case of the general fact that a stripping-down partner always respects shared best objects.

Check your understanding Beginner

Formal definition Intermediate+

Throughout, categories are locally small unless stated otherwise, and $F ⊣ G$ denotes an adjunction with $F : C \to D$ left adjoint to $G : D \to C$ , with unit $η$ and counit $ε$ , in the notation of 41.03.01. The convention here is that limit means a limit of a small diagram $D : J \to D$ , $J$ small.

Definition (preservation of limits). A functor $G : D \to C$ preserves the limit of $D : J \to D$ if, whenever $(lim D, π)$ is a limit cone, the image $(G lim D, G π)$ is a limit cone over $G D$ . $G$ is continuous if it preserves all small limits ^{[Mac Lane 1998]}.

Definition (reflective subcategory and reflector). A full subcategory $ι : A ↪ B$ is reflective if $ι$ has a left adjoint $L : B \to A$ , the reflector; the adjunction $L ⊣ ι$ is the reflection. Because $ι$ is fully faithful, the counit $ε : L ι \Rightarrow Id_{A}$ is an isomorphism, and the unit $η_{B} : B \to ι L B$ is the reflection map, universal among morphisms from $B$ into objects of $A$ : every $f : B \to ι A$ factors uniquely as $ι g \circ η_{B}$ . An object $B$ lies in (the essential image of) $A$ exactly when $η_{B}$ is an isomorphism. Dually, coreflective subcategories have a right-adjoint coreflector.

The reflective examples that recur across the corpus are these. Abelian groups sit reflectively in groups, the reflector being abelianization $G \mapsto G / [G, G]$ with $η_{G} : G \to G^{ab}$ the quotient. Complete metric spaces sit reflectively in metric spaces (with uniformly continuous maps), the reflector being Cauchy completion $X \mapsto X$ with $η_{X} : X \to X$ the dense isometric embedding. Sheaves on a site sit reflectively in presheaves, the reflector being sheafification $F \mapsto F^{+}$ , treated in 04.01.03. Compact Hausdorff spaces sit reflectively in Tychonoff (completely regular Hausdorff) spaces, the reflector being the Stone–Čech compactification $X \mapsto β X$ with $η_{X} : X \to β X$ the universal map into a compactum. Torsion-free abelian groups, complete lattices, and the localizations of a category at a class of maps are further instances.

Definition (solution-set condition). Let $G : D \to C$ and $c \in C$ . A solution set for $c$ is a set $I$ together with a family of objects $(d_{i})_{i \in I}$ in $D$ and morphisms $(f_{i} : c \to G d_{i})_{i \in I}$ such that every morphism $h : c \to G d$ factors as $h = Gt \circ f_{i}$ for some $i \in I$ and some $t : d_{i} \to d$ (the factorization need not be unique). $G$ satisfies the solution-set condition if every object $c \in C$ has a solution set. The set-theoretic point is that the comma category $c ↓ G$ may have a proper class of objects, and a solution set is a weakly initial set — a set of objects through which every object receives a map.

Definition (generators, cogenerators, well-powered). A set $G$ of objects of $D$ is a generating set (the $g \in G$ are generators) if the maps out of the $g$ jointly distinguish morphisms: for $u \neq = v : d ⇉ d^{'}$ there is $g \in G$ and $x : g \to d$ with $ux \neq = v x$ . Dually a cogenerating set $S$ detects morphisms by maps into its members: for $u \neq = v$ there is $s \in S$ and $y : d^{'} \to s$ with $y u \neq = y v$ . A category is well-powered if every object has only a set of subobjects (isomorphism classes of monomorphisms into it).

Counterexamples to common slips Intermediate+

Right adjoints need not preserve colimits. The forgetful functor $U : Grp \to Set$ is a right adjoint (free $⊣ U$ ), so it preserves limits, but it does not preserve coproducts: $U$ of a free product is far larger than the disjoint union of underlying sets. Preservation is tied to the side of the adjunction, and crosses sides only by accident.
The solution-set condition is genuine set-theoretic content, not a formality. Continuity alone does not force a left adjoint. The forgetful functor from complete Boolean algebras (with complete homomorphisms) to sets preserves all limits yet has no left adjoint, because the free complete Boolean algebra on a countably infinite set does not exist — there is no solution set. Dropping the condition breaks the theorem.
A reflective inclusion has an isomorphism counit, not merely an epimorphism unit. Full faithfulness of $ι$ is what forces $ε$ invertible; a left adjoint to a non-full inclusion is not a reflector and the localization need not land in the subcategory. The Stone–Čech unit $η_{X} : X \to β X$ is a dense embedding but is far from surjective, so "reflection map" must not be read as "quotient onto."

Key theorem with proof Intermediate+

The signature result of this unit is RAPL: right adjoints preserve limits, with the dual that left adjoints preserve colimits. It is the structural fact that makes adjunctions the organizing principle for which constructions commute with which.

Theorem (RAPL / LAPC). Let $F ⊣ G$ with $F : C \to D$ , $G : D \to C$ . Then $G$ preserves all limits that exist in $D$ , and $F$ preserves all colimits that exist in $C$ ^{[Mac Lane 1998]}.

Proof. Let $D : J \to D$ have a limit $lim D$ with limit cone $π_{j} : lim D \to D_{j}$ . Fix any $c \in C$ . The adjunction supplies, naturally in $c$ , a bijection $C (c, G lim D) ≅ D (F c, lim D)$ . The universal property of the limit in $D$ states that $D (F c, -)$ carries the limit cone to a limit cone of sets, so $D (F c, lim D) ≅ lim_{j} D (F c, D_{j})$ , the limit computed in $Set$ . Apply the adjunction in each $j$ : $D (F c, D_{j}) ≅ C (c, G D_{j})$ , and these isomorphisms are compatible with the maps of the diagram, so they assemble to $lim_{j} D (F c, D_{j}) ≅ lim_{j} C (c, G D_{j})$ . Chaining, $$ \mathcal{C}(c, G\lim D);\cong;\lim_j\mathcal{C}(c, GD_j), $$ naturally in $c$ . The right-hand side is, by definition, the set of cones over $G D$ with apex $c$ . So $G lim D$ represents the functor $c \mapsto {cones over G D with apex c}$ ; that is exactly the universal property of $lim G D$ . By the Yoneda lemma a natural isomorphism of representable functors is induced by a unique isomorphism of representing objects, and tracking $id_{G l i m D}$ through the chain shows the inducing map is the canonical comparison $G lim D \to lim G D$ built from the cone $G π_{j}$ . Hence $G lim D$ together with the cone $G π$ is a limit of $G D$ : $G$ preserves the limit. The colimit statement for $F$ is the same argument read in the opposite categories, where $F ⊣ G$ becomes $G^{op} ⊣ F^{op}$ and limits become colimits. $□$

Bridge. This theorem builds toward the existence question — given $G$ , when does a left adjoint $F$ exist? — and appears again in the General and Special Adjoint Functor Theorems below, where preservation of limits is the necessary half and the solution-set or cogenerator condition supplies the sufficient half. The foundational reason RAPL holds is that an adjunction is a natural isomorphism of hom-functors, and a limit is a representability statement about a hom-functor; preservation is therefore not a separate fact but the same isomorphism read across a diagram. This is exactly the move that the adjoint functor theorems invert: RAPL says a left adjoint forces continuity of $G$ , and the converse asks whether continuity of $G$ plus a smallness bound forces the left adjoint back into existence by exhibiting an initial object of each comma category $c ↓ G$ . Putting these together, the central insight is that " $G$ has a left adjoint", "each comma category $c ↓ G$ has an initial object", and "the functor $c \mapsto$ a suitable limit is representable" are one condition viewed three ways, and the bridge is that the reflective-subcategory special case — where $C$ already contains $D$ — is exactly the situation where the comma category has a most economical object built as a limit, generalises the completion constructions, and is dual to the coreflective case.

Exercises Intermediate+

Exercise 5 (medium, symbolic).

A reflective subcategory $A ↪ B$ with $B$ complete is itself complete, and limits in $A$ are computed by reflecting the limit taken in $B$ . Prove the completeness.

Hint

$ι$ is a right adjoint; apply it to a diagram in $A$ , take the limit in $B$ , and reflect.

Answer

Let $D : J \to A$ with $J$ small. Form $lim (ιD)$ in the complete category $B$ . Since each $ι D_{j}$ lies in the reflective subcategory and reflective subcategories are closed under limits taken in $B$ — because $ι$ is fully faithful and a right adjoint, so the limit of objects on which $η$ is invertible again has invertible $η$ — the object $lim (ιD)$ is (isomorphic to one) in $A$ . Its cone is a limit cone in $A$ by full faithfulness of $ι$ . Hence $A$ has all small limits. Colimits in $A$ are computed differently: take the colimit in $B$ , then apply the reflector $L$ .

Exercise 6 (hard, symbolic).

State the solution-set condition for $U : Grp \to Set$ and exhibit an explicit solution set for a set $S$ , thereby giving the existence of free groups by the General Adjoint Functor Theorem.

Hint

Bound the cardinality of a group generated by the image of $S$ .

Answer

$U$ preserves limits (it is a right adjoint, but we may verify directly that it preserves products and equalizers). For the solution set at $S$ : any map $h : S \to U D$ has image generating a subgroup $⟨ h (S)⟩ \leq D$ of cardinality at most $κ = max (∣ S ∣, ℵ_{0})$ . Up to isomorphism there is only a set of groups of cardinality $\leq κ$ ; pick one representative $d_{i}$ of each, together with all set maps $f_{i} : S \to U d_{i}$ (a set, since $U d_{i}$ is a set). Every $h : S \to U D$ factors as $U t \circ f_{i}$ where $t : d_{i} ↪ D$ is the inclusion of the generated subgroup and $f_{i}$ corestricts $h$ . This is a solution set, so by GAFT $U$ has a left adjoint, the free-group functor.

Exercise 7 (hard, symbolic).

Verify that $Set$ has a cogenerating set (a single cogenerator), enabling the Special Adjoint Functor Theorem, and identify it.

Hint

Two distinct parallel maps differ at a point; map that point apart in a two-element set.

Answer

The two-element set $2 = {0, 1}$ is a cogenerator of $Set$ . If $u \neq = v : X ⇉ Y$ then $u (x) \neq = v (x)$ for some $x \in X$ ; define $y : Y \to 2$ by $y (u (x)) = 0$ and $y = 1$ elsewhere, so $y u (x) \neq = y v (x)$ , hence $y u \neq = y v$ . So maps into $2$ detect distinct morphisms. With $Set$ complete and well-powered (a subset of a set forms a set of subobjects), SAFT applies: any limit-preserving functor out of $Set$ has a left adjoint. The same cogenerator $2$ underlies the SAFT proof of Stone–Čech compactness, where the unit interval $[0, 1]$ replaces $2$ as the cogenerator of compact Hausdorff spaces.

Advanced results Master

Theorem (General Adjoint Functor Theorem). Let $D$ be locally small and complete, and $G : D \to C$ a functor with $C$ locally small. Then $G$ has a left adjoint if and only if $G$ preserves all small limits and satisfies the solution-set condition ^{[Freyd 1964]}.

The necessity is RAPL together with the observation that an adjunction supplies the singleton solution set ${η_{c} : c \to GF c}$ . The sufficiency is the substantive direction and rests on a representability principle: a left adjoint exists at $c$ exactly when the comma category $c ↓ G$ has an initial object, equivalently when the functor $d \mapsto C (c, G d)$ is representable. Completeness of $D$ makes $c ↓ G$ complete; the solution-set condition supplies a weakly initial set; and a complete category with a weakly initial set has an initial object, obtained by taking the limit of the weakly initial set and cutting it down by a joint equalizer to kill endomorphisms. This last step is the categorical form of the principle that a complete poset with a small cofinal-from-below family has a least element.

Theorem (Special Adjoint Functor Theorem). Let $D$ be locally small, complete, well-powered, and equipped with a cogenerating set $S$ , and let $C$ be locally small. Then a functor $G : D \to C$ has a left adjoint if and only if $G$ preserves all small limits ^{[Mac Lane 1998]}.

The hypotheses on $D$ replace the per-object solution-set condition by a uniform one: in a well-powered complete category with a cogenerating set, every object embeds into a product of cogenerators, and the subobjects of such a product form a set, so the solution set is manufactured for free. SAFT is the workhorse, because its hypotheses are satisfied by $Set$ (cogenerator $2$ ), by compact Hausdorff spaces (cogenerator $[0, 1]$ ), by every Grothendieck topos, and by every locally presentable category. The applications are immediate. The Stone–Čech compactification is the left adjoint to the inclusion of compact Hausdorff spaces into Tychonoff spaces, produced by SAFT on the complete, well-powered category $CHaus$ with cogenerator $[0, 1]$ . Free objects in any variety of algebras, the existence of arbitrary colimits in algebraic categories (left adjoint to the diagonal), and the reflectivity of localizations all follow by exhibiting the relevant functor as continuous on a SAFT-eligible category.

Theorem (reflective localizations and idempotent monads). A reflective subcategory $A ↪ B$ corresponds exactly to an idempotent monad $T = ι L$ on $B$ : the monad whose multiplication $μ : T^{2} \Rightarrow T$ is an isomorphism. The category of $T$ -algebras is then equivalent to $A$ , and the reflection $η : Id \Rightarrow T$ exhibits $A$ as the local objects — the $B$ with $η_{B}$ invertible. Sheafification realizes the sheaf condition as locality for the idempotent monad $(-)^{++}$ on presheaves, the link developed in 04.01.03; localization of modules at a multiplicative set, profinite completion, and the localization of a category at a class $W$ of weak equivalences (when it is reflective) are the same pattern. Idempotency is the categorical content of "completing twice changes nothing."

Theorem (adjoints, representability, and Kan extensions). A functor $G : D \to C$ has a left adjoint if and only if for each $c \in C$ the functor $C (c, G -) : D \to Set$ is representable; the representing objects assemble into the left adjoint by uniqueness of representations. Equivalently, the left adjoint is the pointwise left Kan extension of the identity along $G$ , when that extension exists and is preserved by $G$ . This recasts adjoint existence as a family of representability problems, tying the present unit to the Yoneda machinery of 41.04.02: representability is the engine, and the adjoint functor theorems are the conditions under which the representing object can be constructed as a limit.

Synthesis. Putting these together, the existence theory of adjoints is the existence theory of representable functors, and RAPL is the necessary condition that prunes the search: a candidate left adjoint can exist only for a continuous $G$ , since continuity is preservation of the limits out of which the representing object is built. The foundational reason GAFT works is that completeness lets the comma category $c ↓ G$ have all limits, so a weakly initial set — the solution-set condition — can be sharpened to an initial object by the limit-then-equalize construction; SAFT is the case where a cogenerating set manufactures that solution set uniformly, which is why it, not GAFT, supplies Stone–Čech, free algebras, and the cocompleteness of algebraic categories. This generalises the completion constructions: Cauchy completion, abelianization, and sheafification are reflections, hence idempotent-monad localizations, hence the universal best-approximation inside a reflective subcategory. The central insight is that limits, colimits, free objects, completions, and localizations are not separate existence problems but one — the representability of a hom-functor — and the adjoint functor theorems are the precise hypotheses under which the representing object materializes. This is dual to the colimit side throughout, $D \mapsto D^{op}$ turning right adjoints into left adjoints, and the bridge to monad theory is that every adjunction generates a monad while reflective ones generate the idempotent monads whose algebras recover the subcategory, opening the monadicity question of 41.05.02.

Full proof set Master

Proposition 1 (RAPL via representability). If $F ⊣ G$ and $D : J \to D$ has a limit, then $G lim D ≅ lim G D$ canonically.

Proof. For every $c \in C$ , natural in $c$ , $$ \mathcal{C}(c, G\lim D)\cong\mathcal{D}(Fc,\lim D)\cong\lim_j\mathcal{D}(Fc, D_j)\cong\lim_j\mathcal{C}(c, GD_j)\cong\mathcal{C}\big(c,\lim_j GD_j\big), $$ the first and third isomorphisms being the adjunction, the second the universal property of the limit in $D$ (representables preserve limits, sending the limit cone to a limit cone of sets), and the fourth the universal property of $lim G D$ when it exists; when $lim G D$ is not assumed a priori, the chain exhibits $G lim D$ as representing $c \mapsto lim_{j} C (c, G D_{j})$ , the cone functor of $G D$ , so $G lim D$ is that limit. The composite carries $id_{G l i m D}$ to the canonical comparison map induced by $G π$ , so by the Yoneda lemma the comparison is an isomorphism. $□$

Proposition 2 (initial object from a weakly initial set in a complete category). Let $E$ be locally small and complete with a weakly initial set ${e_{i}}_{i \in I}$ — a set such that every object receives a morphism from some $e_{i}$ . Then $E$ has an initial object.

Proof. Form the product $P = \prod_{i \in I} e_{i}$ , which exists by completeness. $P$ is weakly initial: for any object $x$ , choose $i$ with a map $e_{i} \to x$ and precompose with the projection $P \to e_{i}$ . To remove the multiplicity of maps, take the limit of the (large but cone-bounded) diagram of all endomorphisms of $P$ — concretely, let $w : W \to P$ be the joint equalizer of the set $E (P, P)$ of all endomorphisms of $P$ , which is a small set since $E$ is locally small. For any object $x$ and any two maps $f, g : W ⇉ x$ : composing the weak-initiality map $P \to x$ (call it $u$ ) we get that $f w^{'}, g w^{'}$ ... more directly, $W$ is weakly initial via $P$ , and any endomorphism $k$ of $W$ satisfies $w k = w$ because $w$ equalizes all endomorphisms of $P$ and $w k$ is such after pushing through; since $w$ is a (joint) equalizer it is monic, so $k = id_{W}$ . Thus $W$ has a unique endomorphism. For maps $f, g : W \to x$ , the equalizer $eq (f, g) \to W$ receives a map from $W$ by weak initiality, giving a section, so $eq (f, g) \to W$ is a split monic with a unique endomorphism forcing $f = g$ . Hence $W$ is initial. $□$

Proposition 3 (General Adjoint Functor Theorem). Let $D$ be locally small and complete and $G : D \to C$ preserve all small limits and satisfy the solution-set condition. Then $G$ has a left adjoint.

Proof. It suffices to construct, for each $c \in C$ , an initial object of the comma category $c ↓ G$ , since a left adjoint is exactly a choice of universal arrow, i.e. an initial object of each $c ↓ G$ (the universal-arrow characterization of 41.03.01). The category $c ↓ G$ has objects $(d, h : c \to G d)$ and morphisms $t : d \to d^{'}$ with $Gt \circ h = h^{'}$ . Limits in $c ↓ G$ are created from $D$ because $G$ preserves them: given a small diagram in $c ↓ G$ , take the limit in $D$ , and $G$ -continuity provides the structure map from $c$ . So $c ↓ G$ is complete. The solution-set condition at $c$ is precisely a weakly initial set in $c ↓ G$ . By Proposition 2, $c ↓ G$ has an initial object $(F c, η_{c})$ . Initiality gives, for each $(d, h)$ , a unique $t : F c \to d$ with $Gt \circ η_{c} = h$ , i.e. the natural bijection $D (F c, d) ≅ C (c, G d)$ . Hence $F ⊣ G$ . $□$

Proposition 4 (Special Adjoint Functor Theorem). Let $D$ be locally small, complete, well-powered, with a cogenerating set $S$ , and let $G : D \to C$ preserve all small limits. Then $G$ has a left adjoint.

Proof. By Proposition 3 it suffices to produce a solution set for each $c$ . Fix $c$ and consider all $h : c \to G d$ . Form the product $Q_{d} = \prod_{s \in S} \prod_{G d \to G s} s$ over the cogenerators, with the canonical map $ρ_{d} : d \to Q_{S}$ into the product $Q_{S} = \prod_{s \in S} s^{C (c, G s)}$ assembled from all maps $c \to G s$ ; the cogenerating property makes the comparison detecting morphisms, and factoring $d$ through the image of its induced map exhibits each $h$ as factoring through a subobject of the fixed object $Q_{S}$ . Well-poweredness makes the subobjects of $Q_{S}$ a set; choosing a representative monomorphism $m_{k} : d_{k} ↣ Q_{S}$ for each subobject and, for each, the maps $c \to G d_{k}$ , yields a set ${(d_{k}, c \to G d_{k})}$ . Every $h : c \to G d$ factors through some $d_{k}$ via the image factorization of the induced $d \to Q_{S}$ , so this is a solution set. Proposition 3 then delivers the left adjoint. $□$

Proposition 5 (a reflective subcategory is the algebras of an idempotent monad). If $L ⊣ ι$ is a reflection with $ι$ fully faithful, the monad $T = ι L$ is idempotent, and $D^{T} ≃ A$ .

Proof. The monad $T = ι L$ has unit $η$ and multiplication $μ = ι ε L$ . Since $ι$ is fully faithful, the counit $ε : L ι \Rightarrow Id_{A}$ is an isomorphism, so $μ = ι ε L$ is an isomorphism: $T$ is idempotent. For idempotent $T$ , an object carries at most one $T$ -algebra structure, namely an inverse to $η$ , and it carries one exactly when $η_{B}$ is invertible. The full subcategory of such $B$ is the essential image of $ι$ , which is $A$ by full faithfulness. The comparison functor $A \to D^{T}$ is therefore essentially surjective and fully faithful, an equivalence. $□$

Connections Master

Adjunctions: hom-set and unit-counit 41.03.01. This unit is the existence-and-preservation sequel to the structural development of 41.03.01: the universal-arrow presentation proved there is the exact mechanism by which the General Adjoint Functor Theorem constructs a left adjoint, since a left adjoint is a choice of initial object in each comma category $c ↓ G$ . RAPL, sketched in the Advanced results of 41.03.01, is proved in full here, and the solution-set condition is the smallness hypothesis that turns the universal-arrow characterization into an existence criterion.
Constructing limits and preservation 41.02.02. The completeness construction of 41.02.02 — every limit as the equalizer of two maps between products — is what makes the comma category $c ↓ G$ complete in the GAFT proof, and the limit-then-equalize step of Proposition 2 is exactly that products-and-equalizers reduction applied to a weakly initial set. RAPL here is the headline preservation theorem whose hom-set proof 41.02.02 anticipates in its own Advanced results.
Sheafification and sheaves 04.01.03. Sheafification is the reflector exhibiting sheaves as a reflective subcategory of presheaves, and the idempotent-monad description of Proposition 5 is the abstract content of "sheafify twice changes nothing"; the plus-construction iterated twice is the multiplication $μ$ becoming an isomorphism. The Special Adjoint Functor Theorem gives the abstract existence of sheafification on any Grothendieck topos, complementing the explicit plus-construction of 04.01.03.
The Yoneda lemma and representability 41.04.02. Adjoint existence is representability of the hom-functors $C (c, G -)$ , and the construction of the representing object as a limit is the technical core of both adjoint functor theorems; the co-produced 41.04.02 supplies the Yoneda machinery that turns "represents the same functor" into "is canonically isomorphic," used in the RAPL proof and in the uniqueness of the constructed adjoint.
Monads, monadicity, and algebras 41.05.02. Every adjunction generates a monad, and the reflective adjunctions of this unit generate precisely the idempotent monads, whose Eilenberg–Moore algebras recover the reflective subcategory (Proposition 5); the co-produced monadicity unit 41.05.02 gives Beck's criterion for when a right adjoint exhibits its domain as the algebras of its monad, the non-idempotent generalization of the reflective case treated here.

Historical & philosophical context Master

The theorem that right adjoints preserve limits was among the first structural consequences drawn from Daniel Kan's 1958 definition of adjoint functors ^{[Mac Lane 1998]}, and it explained uniformly a collection of facts that had been observed case by case — that the underlying set of a product of algebraic structures is the product of the underlying sets, that an inverse limit of modules is a submodule of the product, that a representable functor preserves limits. The reflective-subcategory language organizes the completion constructions of nineteenth and early twentieth century mathematics: the Cauchy completion of a metric space, formalized by Cantor and Méray in the 1870s, and the Stone–Čech compactification, constructed by Marshall Stone and Eduard Čech independently in 1937, are recognized as instances of one universal property.

Peter Freyd proved the adjoint functor theorems in his 1964 Abelian Categories ^{[Freyd 1964]}, where the solution-set condition first appears as the precise set-theoretic hypothesis separating "preserves limits" from "has a left adjoint." The Special Adjoint Functor Theorem, with its well-powered-plus-cogenerator hypotheses, became the standard existence tool for free objects and compactifications; Mac Lane's Categories for the Working Mathematician (1971; 2nd ed. 1998) gave the formulation now standard, in Chapter V. The recognition that reflective localizations are idempotent monads, and the general localization theory of categories at a class of weak equivalences, was developed by Pierre Gabriel and Michel Zisman in their 1967 Calculus of Fractions and Homotopy Theory and extended in the locally presentable setting by Gabriel and Ulmer in 1971, the framework in which the adjoint functor theorems acquire their cleanest modern form.

Bibliography Master

@book{MacLane1998aft,
  author    = {Mac Lane, Saunders},
  title     = {Categories for the Working Mathematician},
  edition   = {2},
  publisher = {Springer},
  series    = {Graduate Texts in Mathematics 5},
  year      = {1998}
}

@book{Freyd1964aft,
  author    = {Freyd, Peter},
  title     = {Abelian Categories: An Introduction to the Theory of Functors},
  publisher = {Harper and Row},
  year      = {1964}
}

@book{Riehl2016aft,
  author    = {Riehl, Emily},
  title     = {Category Theory in Context},
  publisher = {Dover},
  year      = {2016}
}

@book{Borceux1994aft,
  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{AdamekRosicky1994aft,
  author    = {Ad\'amek, Ji\v{r}\'i and Rosick\'y, Ji\v{r}\'i},
  title     = {Locally Presentable and Accessible Categories},
  publisher = {Cambridge University Press},
  series    = {London Mathematical Society Lecture Note Series 189},
  year      = {1994}
}

@book{GabrielZisman1967,
  author    = {Gabriel, Pierre and Zisman, Michel},
  title     = {Calculus of Fractions and Homotopy Theory},
  publisher = {Springer},
  series    = {Ergebnisse der Mathematik und ihrer Grenzgebiete 35},
  year      = {1967}
}

@article{Kan1958aft,
  author  = {Kan, Daniel M.},
  title   = {Adjoint functors},
  journal = {Transactions of the American Mathematical Society},
  volume  = {87},
  year    = {1958},
  pages   = {294--329}
}

Prerequisites

41.03.01

Tier anchors

beginner: Riehl 2016 *Category Theory in Context* (Dover) §4.5-4.6 (right adjoints preserve limits as the slogan 'taking limits commutes with the right adjoint'; reflective subcategories as best-approximation, completion and abelianization as worked motivation); Leinster 2014 *Basic Category Theory* (Cambridge) §6.3 (adjoints and limits, the existence question posed informally)
intermediate: Mac Lane 1998 *Categories for the Working Mathematician* 2e (Springer GTM 5) Ch. V §4-5 (right adjoints preserve limits; the General Adjoint Functor Theorem and the solution-set condition), Ch. V §6 (the Special Adjoint Functor Theorem); Riehl 2016 *Category Theory in Context* (Dover) §4.5-4.6, §5.4; Awodey 2010 *Category Theory* 2e (Oxford) §9.5-9.6
master: Mac Lane 1998 *Categories for the Working Mathematician* 2e (Springer GTM 5) Ch. V in full (preservation, the two adjoint functor theorems, reflective subcategories, Kan extensions as the limit form of adjoints); Borceux 1994 *Handbook of Categorical Algebra 1* (Cambridge) §3.3 (the adjoint functor theorems), Ch. 5 (reflective subcategories and localizations); Adámek-Rosický 1994 *Locally Presentable and Accessible Categories* (Cambridge LMS 189) Ch. 1-2 (the AFT in the locally presentable setting); Freyd 1964 *Abelian Categories* Ch. 3

References

Mac Lane — Categories for the Working Mathematician · 2nd ed., Springer GTM 5, Ch. V §4-6
Freyd — Abelian Categories · Harper and Row 1964, Ch. 3
Riehl — Category Theory in Context · Dover 2016, §4.5-4.6, §5.4
Borceux — Handbook of Categorical Algebra 1 · Cambridge 1994, §3.3, Ch. 5
Adámek-Rosický — Locally Presentable and Accessible Categories · Cambridge LMS 189, 1994, Ch. 1-2

Estimated time

beginner: 20m
intermediate: 50m
master: 90m