Adjoint Functors and Freyd's Adjoint Functor Theorem

• Adjoint functors capture optimal solutions categorically via natural bijection: Hom_𝒟(F(c), d) ≅ Hom_𝒞(c, G(d)) between functors F and G
• Key examples: free groups adjoint to forgetful functor, tensor products adjoint to Hom, Stone-Čech Compactification adjoint to inclusion
• Freyd's Adjoint Functor Theorem (1964) provides criterion for adjoint existence based on completeness and size conditions, replacing explicit construction
• Left adjoints preserve colimits; right adjoints preserve limits—fundamental result enabling structural analysis across algebra, topology, geometry
• Adjunctions reveal universal properties underlying diverse mathematical constructions, making them a central organizing principle since development by Kan-Eilenberg in 1950s
• "Nice" categories automatically admit abundant adjoint functors, explaining prevalence of certain universal constructions in mathematics

Adjoint Functors and Freyd's Adjoint Functor Theorem

Overview

Adjoint functors form one of the central organizing principles of modern mathematics, capturing the notion of "optimal solution" in a categorical framework. An adjunction between categories $\mathcal{C}$ and $\mathcal{D}$ consists of a pair of functors $F:\mathcal{C}\to \mathcal{D}$ and $G:\mathcal{D}\to \mathcal{C}$ satisfying a natural bijection ${\text{Hom}}_{\mathcal{D}}(F(c),d)\cong {\text{Hom}}_{\mathcal{C}}(c,G(d))$. This relationship appears throughout mathematics: free groups are left adjoint to the forgetful functor from groups to sets, tensor products are left adjoint to internal hom functors, and Stone-Čech compactification is left adjoint to inclusion of compact Hausdorff spaces into all topological spaces. The ubiquity of adjunctions led Saunders Mac Lane to call them "a central organizing principle of mathematics."

The practical construction of adjoint functors, however, is not always straightforward. While many important functors have adjoints, proving their existence can require explicit construction of potentially complicated objects. Freyd's Adjoint Functor Theorem provides an elegant criterion: a functor has an adjoint if and only if it satisfies certain completeness and size conditions. This theorem transforms the problem from explicit construction to verification of abstract properties, making it a powerful tool in category theory, algebraic topology, algebraic geometry, and logic.

The development of adjoint functors in the 1950s by Daniel Kan and Samuel Eilenberg, followed by Peter Freyd's theorem in 1964, marked a turning point in categorical thinking. Rather than viewing categories merely as bookkeeping devices for mathematical structures, adjunctions revealed deep universal properties underlying diverse mathematical constructions. The theorem shows that "nice" categories automatically admit many adjoint functors, explaining why certain constructions (free objects, limits, colimits) exist more generally than one might initially expect.

An adjunction $F⊣G$ means $F$ provides the "best $\mathcal{D}$-approximation" to objects in $\mathcal{C}$, while $G$ provides the "best $\mathcal{C}$-approximation" to objects in $\mathcal{D}$. Freyd's theorem says that if a functor preserves all limits and satisfies a solution set condition, it automatically has a left adjoint—the universal construction exists without explicit construction.

1. $G$ preserves all (small) limits
2. $G$ satisfies the solution set condition: for each $c\in \mathcal{C}$, there exists a set $S$ of morphisms $\{f_i:c\to G(d_i){\}}_{i\in I}$ such that every morphism $g:c\to G(d)$ factors as $g=G(h)\circ f_i$ for some $i\in I$ and $h:d_i\to d$.

Core Theory

Main Result

We prove Theorem 1. The "only if" direction is straightforward: if $F⊣G$, then $G$ preserves limits because ${\text{Hom}}_{\mathcal{C}}(c,G(\lim d_j))\cong {\text{Hom}}_{\mathcal{D}}(F(c),\lim d_j)\cong \lim {\text{Hom}}_{\mathcal{D}}(F(c),d_j)\cong \lim {\text{Hom}}_{\mathcal{C}}(c,G(d_j))\cong {\text{Hom}}_{\mathcal{C}}(c,\lim G(d_j))$ by Yoneda lemma. The solution set exists by taking a singleton containing any component of the unit.

For the "if" direction, fix $c\in \mathcal{C}$. We construct $F(c)$ as a limit. Let $S=\{f_i:c\to G(d_i){\}}_{i\in I}$ be a solution set for $c$. Consider the comma category $(c\downarrow G)$ whose objects are pairs $(d,f:c\to G(d))$ and morphisms from $(d,f)$ to $(d^{\prime },f^{\prime })$ are arrows $h:d\to d^{\prime }$ in $\mathcal{D}$ with $G(h)\circ f=f^{\prime }$.

Define the projection functor $\pi :(c\downarrow G)\to \mathcal{D}$ sending $(d,f)\mapsto d$. Since $\mathcal{D}$ is complete, the limit $L=\lim \pi$ exists in $\mathcal{D}$. We claim $F(c)=L$ with universal morphism given by the composition

To verify universality, let $g:c\to G(d^{\prime })$ be any morphism. By the solution set condition, there exists $i\in I$ and $h:d_i\to d^{\prime }$ with $g=G(h)\circ f_i$. This shows $(d^{\prime },g)$ is in the comma category $(c\downarrow G)$, giving a projection $p_{d^{\prime },g}:L\to d^{\prime }$ from the limit. We claim $\bar{g}=p_{d^{\prime },g}$ satisfies $g=G(\bar{g})\circ {\eta }_c$.

Indeed, since $G$ preserves limits, $G(L)\cong {\lim }_{(d,f)\in (c\downarrow G)}G(d)$, and the component of this limit at $(d^{\prime },g)$ is precisely $G(p_{d^{\prime },g})$. The definition of ${\eta }_c$ and the universal property of the limit ensure that the component at $(d^{\prime },g)$ of $G(\text{lim proj})\circ \Delta$ is exactly $g$, so $G(p_{d^{\prime },g})\circ {\eta }_c=g$.

For uniqueness, suppose ${\bar{g}}^{\prime }:F(c)\to d^{\prime }$ also satisfies $g=G({\bar{g}}^{\prime })\circ {\eta }_c$. The morphism ${\bar{g}}^{\prime }$ induces a cone from $L$ to the diagram $\pi$. By the universal property of the limit, this cone is determined by its components, which are in turn determined by composing with $G$. Since $G({\bar{g}}^{\prime })\circ {\eta }_c=g=G(\bar{g})\circ {\eta }_c$ and ${\eta }_c$ is the universal cone to $G(L)$, we have ${\bar{g}}^{\prime }=\bar{g}$.

To extend this to a functor $F:\mathcal{C}\to \mathcal{D}$ and verify the adjunction is natural, we use similar limit constructions for morphisms in $\mathcal{C}$ and apply functoriality of limits. □

Examples and Applications

1

First, $\mathbf{Grp}$ is complete (products of groups are constructed componentwise, equalizers are kernels). Second, $U$ preserves limits: the underlying set of a product of groups is the product of underlying sets, and similarly for equalizers.

For the solution set condition, fix a set $X$. Consider the set $S=\{\iota :X\to U(F(X))\}$ where $\iota$ is the canonical inclusion and $F(X)$ is the free group on $X$ (which we construct explicitly). For any homomorphism $f:X\to U(G)$ where $G$ is a group, the universal property of free groups gives a unique homomorphism $\bar{f}:F(X)\to G$ with $f=U(\bar{f})\circ \iota$. Thus $S$ is a solution set (a singleton!).

By Freyd's theorem, $U$ has a left adjoint. The construction in the proof actually recovers the free group: for a set $X$, the free group $F(X)$ consists of reduced words in $X\cup X^{-1}$, and the unit ${\eta }_X:X\to U(F(X))$ sends each element to the corresponding word of length 1.

2

We verify this using Freyd's theorem. The category $\mathcal{D}$ is complete (limits of sheaves are computed as in presheaves, and the sheaf condition is preserved). The inclusion $i$ preserves limits trivially since it's a full inclusion and limits are computed the same way.

For the solution set condition, fix a presheaf $\mathcal{F}$. Consider the étalé space construction $\text{Et}(\mathcal{F})$, which gives a sheaf ${\mathcal{F}}^{+}$, together with the natural morphism $\theta :\mathcal{F}\to i({\mathcal{F}}^{+})$ defined by taking germs. The set $S=\{\theta \}$ is a solution set: if $\varphi :\mathcal{F}\to i(\mathcal{G})$ for any sheaf $\mathcal{G}$, then since $\mathcal{G}$ already satisfies the sheaf condition, $\varphi$ factors uniquely through $\theta$ by the universal property of taking stalks and sections.

The sheafification functor $(-{)}^{+}:\mathcal{C}\to \mathcal{D}$ is the left adjoint guaranteed by Freyd's theorem, giving the natural bijection ${\text{Hom}}_{\mathcal{D}}({\mathcal{F}}^{+},\mathcal{G})\cong {\text{Hom}}_{\mathcal{C}}(\mathcal{F},i(\mathcal{G}))$.

The functor $G$ preserves all limits (limits decrease or preserve cardinality in $\mathbf{Set}$, so limits in $\mathcal{D}$ of sets smaller than $\kappa$ remain smaller than $\kappa$). However, $G$ does not have a left adjoint.

The issue is the solution set condition. Consider the singleton set $\{\ast \}\in \mathcal{C}$. For $G$ to have a left adjoint, we'd need a "free" object $F(\{\ast \})$ and a universal map $\eta :\{\ast \}\to G(F(\{\ast \}))$. But for any set $X$ with $|X|\ge \kappa$, we have $G(X)=\varnothing$, so there can be no morphism from $\{\ast \}$ to $G(X)$ factoring through the unit. For sets $X$ with $|X|<\kappa$, the morphisms $\{\ast \}\to X$ are just elements of $X$, and there are unboundedly many such sets below $\kappa$ (since $\kappa$ is inaccessible). No set of morphisms can serve as a solution set because we'd need representatives for arbitrary large cardinalities below $\kappa$, which would form a proper class, not a set.

Think of an adjunction $F⊣G$ as an "optimal approximation" relationship. If $G:\mathcal{D}\to \mathcal{C}$ "forgets" structure (like the forgetful functor from groups to sets), then its left adjoint $F$ "freely adds" that structure back in the most economical way. The unit ${\eta }_c:c\to GF(c)$ embeds your object into the underlying object of its free version.

Geometrically, imagine $\mathcal{C}$ as a "simple" space and $\mathcal{D}$ as a "structured" space. The right adjoint $G$ projects from structured to simple. The left adjoint $F$ lifts from simple to structured, choosing the "freest" or "most general" structured object that maps back down to your simple object.

Freyd's theorem says: if $G$ respects all limits (it's "continuous" in a categorical sense) and there's only a set's worth of "essentially different" ways to approximate any object (solution set condition), then the optimal approximations $F(c)$ exist automatically as certain limits. You don't need to construct them by hand—the theorem guarantees their existence through abstract nonsense.

Historical Development

The concept of adjoint functors emerged in the mid-1950s from the work of Daniel Kan on homotopy theory. Kan introduced "Kan extensions" in 1958, which are examples of adjoint functors, though the general definition was formulated independently by Samuel Eilenberg and Mac Lane around the same time. The terminology "adjoint" was chosen by analogy with adjoint operators in Hilbert space theory, where the relationship $⟨Ax,y⟩=⟨x,A^{\ast }y⟩$ mirrors the bijection $\text{Hom}(Fc,d)\cong \text{Hom}(c,Gd)$.

Peter Freyd's 1964 doctoral thesis at Princeton, published as the monograph "Abelian Categories," revolutionized the field by introducing the General Adjoint Functor Theorem. Freyd's work was motivated by questions in homological algebra about when certain functors (particularly those between abelian categories) admit adjoints. His solution set condition provided the first general size-theoretic criterion for existence of adjoints, complementing the structural condition of limit preservation.

The Special Adjoint Functor Theorem, which replaces the solution set condition with structural properties like well-poweredness and existence of a cogenerator, was developed shortly after by various authors including Freyd and John Isbell. Modern formulations using locally presentable categories, developed by Gabriel-Ulmer in the 1970s and refined by Adámek-Rosický in the 1990s, show that in many "naturally occurring" categories, the solution set condition is automatic.

Connections to Other Areas

What You Need First

What This Opens Up

See Also