category theory

# Equivalence of categories

## Idea

The concept of equivalence of categories is the correct category theoretic notion of “sameness” of categories.

Concretely, an equivalence between two categories is a pair of functors between them which are inverse to each other up to natural isomorphism of functors.

This is like an isomorphism, but weakened such as to accomodate for the fact that the correct ambient context for categories is not iself a 1-category, but is the 2-category Cat of all categories. Hence abstractly an equivalence of categories is just the special case of an equivalence in a 2-category specialized to Cat.

If some foundational fine print is taken care of, then a functor exhibits an equivalence of categories precisely if it is both essentially surjective and fully faithful. This is true in classical mathematics if the axiom of choice is assumed. It remains true non-classically, say for internal categories, if the concept of functor is suitably adapted (“anafunctors”).

Form the point of view of logic one may say that two categories are equivalent if they have the same properties — although this only applies (by definition) to properties that obey the principle of equivalence.

Just as equivalence of categories is the generalization of isomorphism of sets from sets to categories, so the concept generalizes further to higher categories (see e.g. equivalence of 2-categories, equivalence of (∞,1)-categories) and ultimately to equivalence of $\infty$-categories.

## Definitions

###### Definition

An equivalence between two categories $\mathcal{C}$ and $\mathcal{D}$ is an equivalence in the 2-category Cat of all categories, hence it is

1. a pair of functors

$\mathcal{C} \underoverset {\underset{\phantom{AA}F \phantom{AA}}{\longrightarrow}} {\overset{\phantom{AA}G\phantom{AA}}{\longleftarrow}} {} \mathcal{D},$
2. natural isomorphisms

$F \circ G \cong Id_{\mathcal{D}}$

and

$G \circ F \cong Id_{\mathcal{C}} \,.$

This is called an adjoint equivalence if the natural isomorphisms above satisfy the triangle identities, thus exhibiting $F$ and $G$ as a pair of adjoint functors.

Two categories are called equivalent if there exists an equivalence between them.

###### Proposition

Assume the ambient context is one of the following (see below at Variants for more):

Let $F \colon \mathcal{C} \to \mathcal{D}$ be a functor. Then the following are equivalent:

1. $F$ is part of an equivalence of categories in the sense of def. 1

2. $F$ is

## Variants

We discuss some possible variants of the definition of equivalence of categories, each of which comes naturally from a different view of Cat.

The first, isomorphism, comes from viewing $Cat$ as a mere 1-category; it is too strong and is really only of interest for strict categories. The next, strong equivalence, comes from viewing $Cat$ as a strict 2-category; it is the most common definition given and is correct if and only if the axiom of choice holds. The next definition, weak equivalence, comes from viewing $Cat$ as a model category; it is correct with or without choice and is about as simple to define as strong equivalence. The last, anaequivalence, comes from viewing $Cat$ as a bicategory that is not (without the axiom of choice) equivalent (as a bicategory!) to the strict $2$-category that defines strong equivalence; it is also always correct.

It is also possible to define ‘category’ in such a way that only a correct definition can be stated, but here we use the usual algebraic definitions of category, functor, and natural isomorphism.

### Isomorphism

Two strict categories $C$ and $D$ are isomorphic if there exist strict functors $F\colon C \to D$ and $G\colon D \to C$ such that $F G$ and $G F$ are each equal to the appropriate identity functor. In this case, we say that $F$ is an isomorphism from $C$ to $D$ (so $G$ is an isomorphism from $D$ to $C$) and call the pair $(F,G)$ an isomorphism between $C$ and $D$. The functor $G$ is called the strict inverse of $F$ (so $F$ is the strict inverse of $G$).

If you think of $Cat$ as the category of (strict) categories and functors, then this is the usual notion of isomorphism in a category. This is the most obvious notion of equivalence of categories and the first to be considered, but it is simply too strong for the purposes to which category theory is put.

#### Basic examples why strict isomorphism is too strong

Generally, (and to repeat what was said above), the reason to avoid isomorphisms in favor of equivalences of categories is the flexibility gained by replacing strict equations by isomorphism.

Specifically, a basic example is this: it is widely known that, given a field $K$, to most intents and purposes the abstract concept of a $2$-dimensional $K$-vector space (together with all $K$-linear maps between them) is equivalent to the concrete concept of the dimension $d$ itself (together with all $2\times 2$ matrices over $K$). The functor $F$ which sends an abstract $2$-dimensional $K$-vector spaces to the dimension $2$ itself (and does the usual thing to the linear maps) can be fitted into an equivalence of categories, but not into an isomorphism of categories, essentially for size reasons alone: the target category is a small category (and if the field is finite, even a finite category), while the source category is a large category. The source category (of all $K$-vector spaces) is large even if $K$ is a finite field, but for an inessential reason: in the usual contemporary implementation of mathematics, you are absolutely free how to set-theoretically implement your vector space (you may declare any set, no matter how large, to serve as your first basis vector, for example), but these implementation choices are essentially irrelevant. So the single number $2$ gets sent a proper class of objects, and when defining the functor $G$ (in the notation above) you both (0) have to make an arbitrary choice among a class of objects, and (1) cannot make that choice so as to have $G\circ F=Id_{\mathcal{C}}$ strictly.

This should not be a reason not to regard these two concepts to be equivalent, though, and category theory offers a precise instrument how to do so: whatever choice you make, you do get $G\circ F\cong Id_C$, since any two $K$-vector spaces of the same dimension are isomorphic. (Incidentally, this equivalence between a large category and a finite category also is, in a sense, a half-isomorphism of categories: whichever of the choices you make, you do get $F\circ G=Id_{\mathcal{D}}$ strictly.)

Another usual example is $\mathcal{C}:= \bullet\downarrow\mathsf{Sets}$ $=$ the coslice category ($=$ undercategory) of the category of sets under a terminal object. This is also known as the category of pointed sets and $\mathcal{D}$ the full subcategory of the category of spans in $\mathsf{Sets}$ consisting of precisely those spans at least one of whose legs is an injection. This is often interpreted as the category of partial functions of sets. See the section “The category of sets and partial functions” in partial function for this equivalence.

### Strong equivalence

Two strict categories $C$ and $D$ are strongly equivalent if there exist strict functors $F\colon C \to D$ and $G\colon D \to C$ such that $F G$ and $G F$ are each naturally isomorphic (isomorphic in the relevant functor category) to the appropriate identity functor. In this case, we say that $F$ is a strong equivalence from $C$ to $D$ (so $G$ is a strong equivalence from $D$ to $C$). The functor $G$ is called a weak inverse of $F$ (so $F$ is a weak inverse of $G$).

Note that an isomorphism is precisely a strong equivalence in which the natural isomorphisms are identity natural transformations.

If you think of $Cat$ as the strict 2-category of (strict) categories, functors, and natural transformations, then this is the usual notion of equivalence in a $2$-category. This is the first ‘correct’ definition of equivalence to be considered and the one most often seen today; it is only correct using the axiom of choice.

If possible, use or modify the counterexample to isomorphism to show how choice follows if strong equivalence is assumed correct.

### Weak equivalence

Two categories $C$ and $D$ are weakly equivalent if there exist a category $X$ and functors $F\colon X \to D$ and $G\colon X \to C$ that are essentially surjective and fully faithful. In this case, we say that $F$ is a weak equivalence from $X$ to $D$ (so $G$ is a weak equivalence from $X$ to $C$) and call the span $(X,F,G)$ a weak equivalence between $C$ and $D$. (It is not entirely trivial to check that such spans can be composed, but they can be.)

A strict functor with a weak inverse is necessarily essentially surjective and fully faithful; the converse is equivalent to the axiom of choice. Thus any strong equivalence becomes a weak equivalence in which $X$ is taken to be either $C$ or $D$ (or even built symmetrically out of $C$ and $D$ if you're so inclined); a weak equivalence becomes a strong equivalence using the axiom of choice to find weak inverses and composing across $X$.

If you think of $Cat$ as the model category of categories and functors with the canonical model structure, then this is the usual notion of weak equivalence in a model category.

### Anaequivalence

Two categories $C$ and $D$ are anaequivalent if there exist anafunctors $F\colon C \to D$ and $G\colon D \to C$ such that $F G$ and $G F$ are each ananaturally isomorphic (isomorphic in the relevant anafunctor category) to the appropriate identity anafunctor. In this case, we say that $F$ is an anaequivalence from $C$ to $D$ (so $G$ is an anaequivalence from $D$ to $C$). The functor $G$ is called an anainverse of $F$ (so $F$ is an anainverse of $G$). See also weak equivalence of internal categories.

Any strict functor is an anafunctor, so any strong equivalence is an anaequivalence. Using the axiom of choice, any anafunctor is ananaturally isomorphic to a strict functor, so any anaequivalence defines a strong equivalence. Using the definition of an anafunctor as an appropriate span of strict functors, a weak equivalence defines two anafunctors which form an anaequivalence; conversely, either anafunctor in an anaequivalence is (as a span) a weak equivalence.

If you think of $Cat$ as the bicategory of categories, anafunctors, and ananatural transformations, then this is the usual notion of equivalence in a $2$-category. It's fairly straightforward to turn any discussion of functors and strong equivalences in a context where the axiom of choice is assumed into a discussion of anafunctors and anaequivalences in a more general context.

We can also regard the $2$-category $Cat$ above as obtained from the $2$-category $Str Cat$ of strict categories, strict functors, and natural transformations by formally inverting the weak equivalences as in homotopy theory.

### Remarks

Note that weak inverses go with strong equivalences. The terminology isn't entirely inconsistent (weak inverses contrast with strict ones, while weak equivalences contrast with strong ones) but developed at different times. The prefix ‘ana‑’ developed last and is perfectly consistent.

If you accept the axiom of choice, then you don't have to worry about the different kinds of equivalence (as long as you don't use isomorphism). This is not just a question of foundations, however, since the axiom of choice usually fails in internal contexts.

It's also possible to use foundations (such as homotopy type theory, some other forms of type theory, or FOLDS) in which isomorphism and strong equivalence are impossible to state. In such a case, one usually drops the prefixes ‘weak’ and ‘ana‑’. In the $n$-Lab, we prefer to remain agnostic about foundations but usually drop these prefixes as well, leaving it up to the reader to insert them if necessary.

Any equivalence can be improved to an adjoint equivalence: a strong equivalence or anaequivalence whose natural isomorphisms satisfy the triangle identities. In that case, $G$ is called the adjoint inverse of $F$ (so $F$ is the adjoint inverse of $G$). When working in the $2$-category $Cat$, a good rule of thumb is that it is okay to consider either

• a functor with the property of being a general equivalence or
• a functor with the structure of being an adjoint equivalence (that is, a functor $G$ and a pair of natural isomorphisms $F G \cong 1$ and $1 \cong G F$ satisfying the triangle identities),

whereas considering

• a functor with the structure of being a general equivalence (that is, merely a functor $G$ and a pair of natural isomorphisms $F G \cong 1$ and $1 \cong G F$)

is fraught with peril. For instance, an adjoint inverse is unique up to unique isomorphism (much as a strict inverse is unique up to equality), while a weak inverse or anainverse is not. Including adjoint equivalences is also the only way to make a higher-categorical structure completely algebraic.

### An example of a non-adjoint equivalence

Identify a group $H$ with its delooping. One can check the following:

• Any equivalence $F : H \leftrightarrows H : G$ of a group with itself comprises two automorphisms $F, G$, such that $F G$ and $G F$ are inner. The unit and counit are the group elements $g_{\rho}$ such that $GF(k) = g_{\rho} k g_{\rho}^{-1}$ and $g_{\sigma}$ such that $FG(k) = g_{\sigma}^{-1} k g_{\sigma}$ for any $k \in H$.

• Any equivalence of $H$ with itself where $F$ and $G$ are themselves also inner is an adjoint equivalence.

• If $H$ has trivial center, then any equivalence of $H$ with itself is an adjoint equivalence.

• To obtain a non-adjoint equivalence, we therefore need a group $H$ with nontrivial center and nontrivial outer automorphisms, such that we can pick two whose products are inner.

• So take $H = K$ the Klein 4-group. This is a product of abelian groups, so abelian, so is its own center. In fact, it’s $\mathbb{Z}/2 \mathbb{Z} \times \mathbb{Z} / 2 \mathbb{Z}$, so let $F = G$ the automorphism which interchanges coordinates. $FG = GF = \operatorname{id}_{K}$, which is given by conjugation by any element.

• If this were adjoint, the triangle equality for $F$ will stipulate that $F(g_{\rho}) = g_{\sigma}^{-1}$. We can pick $g_{\rho}$ and $g_{\sigma}$ to break this. For example, let $g_{\rho}$ be $(1,1)$ and let $g_{\sigma}$ be $(0,1)$.

This is a special case of the fact that, given a non-adjoint equivalence, you can always replace its unit with another unit (which determines the counit) to improve the equivalence to an adjoint equivalence.

## In higher categories

All of the above types of equivalence make sense for $n$-categories and $\infty$-categories defined using an algebraic definition of higher category; again, it is the weak notion that is usually wanted. When using a geometric definition of higher category, often isomorphism cannot even be written down, so equivalence comes more naturally.

In particular, one expects (although a proof depends on the exact definition and hasn't always been done) that in any $(n+1)$-category of $n$-categories, every equivalence (in the sense of an $(n+1)$-category) will be essentially $k$-surjective for all $0\le k\le n+1$; this is the $n$-version of “full, faithful, and essentially surjective.” The converse should be true assuming that

• we have an axiom of choice and use weak (pseudo) $n$-functors, or
• we use $n$-anafunctors? (which are automatically weak).

If we use too strict a notion of $n$-functor, then we will not get the correct notion of equivalence; if we use weak $n$-functors but not anafunctors, then we will get the correct notion of equivalence only if the axiom of choice holds, although again this can be corrected by moving to a span. Note that even strict $n$-categories need weak $n$-functors to get the correct notion of equivalence between them!

For example, assuming choice, a strict 2-functor between strict $2$-categories is an equivalence in $Bicat$ if and only if it is essentially (up to equivalence) surjective on objects, locally essentially surjective, and locally fully faithful. However, its weak inverse may not be a strict $2$-functor, and even if it is, the equivalence transformations need not be strictly $2$-natural. Thus, it need not be an equivalence in the strict 3-category $Str 2 Cat$ of $2$-categories, strict $2$-functors, and strict $2$-natural transformations, or even in the semi-strict 3-category? $Gray$ of strict $2$-categories, strict $2$-functors, and pseudonatural transformations.

As with $Cat$, we can recover $Bicat$ as a full subtricategory of $Gray$ by formally inverting all such weak equivalences. Note that even with the axiom of choice, $Bicat$ is not equivalent (as a tricategory) to $Gray$, even though by the coherence theorem for tricategories it is equivalent to some $Gray$-category; see here.

Revised on June 16, 2017 05:18:01 by Peter Heinig (84.183.88.168)