nLab equivalence type

Idea

In dependent type theory, the equivalence type is to types what the identity type is to terms: it represents the collection of “equalities” between types (equality of types being given by the notion of equivalence in type theory), in the same way that the identity type represents the collection of equalities between terms (equality of terms being given by the notion of identity/identification/path).

Definition

In dependent type theory, the equivalence type between two types $A$ and $B$ is the type $A \simeq B$ whose terms are equivalences between $A$ and $B$. Like any other notion of type in dependent type theory, there are two different notions of equivalence types in type theory: strict and weak equivalence types. Strict equivalence types use judgmental equality in the conversion rules, while weak equivalence types use identity types in the conversion rules. Weak equivalence types could also be defined analytically from other type formers in the type theory.

Strict equivalence types

Given types $A$ and $B$, one could define the type of strict equivalences betwen $A$ and $B$. These are given by the following rules:

Formation rule for strict equivalence types:

$$\frac{\Gamma \vdash A \; \mathrm{type} \quad \Gamma \vdash B \; \mathrm{type}}{\Gamma \vdash A \simeq B \; \mathrm{type}}$$

Introduction rule for strict equivalence types:

$$\frac{\Gamma \vdash A \; \mathrm{type} \quad \Gamma \vdash B \; \mathrm{type} \quad \Gamma, x:A \vdash f(x):B \quad \Gamma, y:B \vdash g(y):A \quad \Gamma, x:A \vdash g(f(x)) \equiv x:A \quad \Gamma, y:B \vdash f(g(y)) \equiv y:B}{\Gamma \vdash \mathrm{toequiv}(x:A.f(x), y:B.g(y)):A \simeq B}$$

Elimination rules for strict equivalence types:

$$\frac{\Gamma \vdash A \; \mathrm{type} \quad \Gamma \vdash B \; \mathrm{type}}{\Gamma, e:A \simeq B, x:A \vdash \overrightarrow{e}(x):B} \qquad \frac{\Gamma \vdash A \; \mathrm{type} \quad \Gamma \vdash B \; \mathrm{type}}{\Gamma, e:A \simeq B, y:B \vdash \overleftarrow{e}(x):A}$$

$$\frac{\Gamma \vdash A \; \mathrm{type} \quad \Gamma \vdash B \; \mathrm{type}}{\Gamma, e:A \simeq B, x:A \vdash \overleftarrow{e}(\overrightarrow{e}(x)) \equiv x:A} \qquad \frac{\Gamma \vdash A \; \mathrm{type} \quad \Gamma \vdash B \; \mathrm{type}}{\Gamma, e:A \simeq B, y:B \vdash \overrightarrow{e}(\overleftarrow{e}(y)) \equiv y:B}$$

Computation rules for strict equivalence types:

$$\frac{\Gamma \vdash A \; \mathrm{type} \quad \Gamma \vdash B \; \mathrm{type} \quad \Gamma, x:A \vdash f(x):B \quad \Gamma, y:B \vdash g(y):A \quad \Gamma, x:A \vdash g(f(x)) \equiv x:A \quad \Gamma, y:B \vdash f(g(y)) \equiv y:B}{\Gamma, x:A \vdash \overrightarrow{\mathrm{toequiv}(x:A.f(x), y:B.g(y))}(x) \equiv f(x):B}$$

$$\frac{\Gamma \vdash A \; \mathrm{type} \quad \Gamma \vdash B \; \mathrm{type} \quad \Gamma, x:A \vdash f(x):B \quad \Gamma, y:B \vdash g(y):A \quad \Gamma, x:A \vdash g(f(x)) \equiv x:A \quad \Gamma, y:B \vdash f(g(y)) \equiv y:B}{\Gamma, y:B \vdash \overleftarrow{\mathrm{toequiv}(x:A.f(x), y:B.g(y))}(y) \equiv g(y):A}$$

Uniqueness rules for strict equivalence types:

$$\frac{\Gamma \vdash A \; \mathrm{type} \quad \Gamma \vdash B \; \mathrm{type}}{\Gamma, e:A \simeq B \vdash \mathrm{toequiv}(x:A.\overrightarrow{e}(x), y:B.\overleftarrow{e}(y)) \equiv e:A \simeq B}$$

Weak equivalence types

As a dependent sum type of the isEquiv type family

Given a notion of the isEquiv type family on the function type $A \to B$, the equivalence type is defined by

$$A \simeq B \coloneqq \sum_{f:A \to B} \mathrm{isEquiv}(f)$$

Locally small equivalence types

Given a type universe $U$ and a notion of a $U$-small isEquiv type family for some type $F_U(A, B)$, the locally $U$-small equivalence type is defined by

$$A \simeq_U B \coloneqq \sum_{f:F_U(A, B)} \mathrm{isEquiv}_U(f)$$

$F_U(A, B)$ could be the type of $U$-small spans, the type of $U$-small multivalued partial functions, or the type of $U$-small correspondences.

Rules for weak equivalence types

In the same way that isEquiv could be defined in a way such that given the function $f:A \to B$ and a witness $p:\mathrm{isEquiv}(f)$, $(B, f, p)$ satisfies the universal property of a wrapped copy of $A$, one could define weak equivalence types such that given an equivalence $R:A \simeq B$, $(B, R)$ satsifes the universal property of a wrapped copy of $A$:

Formation rules for equivalence types:

$$\frac{\Gamma \vdash A \; \mathrm{type} \quad \Gamma \vdash B \; \mathrm{type}}{\Gamma \vdash A \simeq B \; \mathrm{type}}$$

Introduction rules for equivalence types:

$$\frac{ \begin{array}{l} \Gamma \vdash A \; \mathrm{type} \quad \Gamma \vdash B \; \mathrm{type} \quad \Gamma, f:A \to B \quad \Gamma \vdash a:A \quad \Gamma \vdash b:\prod_{y:B} f(a) =_B y \\ \Gamma \vdash \tau_A:\prod_{x:A} \prod_{y:B} (f(x) =_B y) \to (a =_A x) \quad \Gamma \vdash \tau_B:\prod_{x:A} \prod_{y:B} \prod_{z:f(x) =_B y} b(y) =_B^{\tau_A(x, y, z)} z \end{array} }{\Gamma \vdash \mathrm{equiv}(f, a, b, \tau_A, \tau_B):A \simeq B}$$

Elimination rules for equivalence types:

$$\frac{\Gamma \vdash A \; \mathrm{type} \quad \Gamma \vdash B \; \mathrm{type} \quad \Gamma, R:A \simeq B}{\Gamma, x:A \vdash f_R(x):B}$$

$$\frac{ \begin{array}{l} \Gamma \vdash A \; \mathrm{type} \quad \Gamma \vdash B \; \mathrm{type} \quad \Gamma, R:A \simeq B \\ \Gamma, y:B \vdash C(y) \; \mathrm{type} \quad \Gamma \vdash c_R:\prod_{x:A} C(f_R(x)) \quad \Gamma \vdash b:B \end{array} }{\Gamma \vdash \mathrm{ind}_{A \simeq B}^C(R, c_R, b):C(b)}$$

Computation rules for equivalence types:

$$\frac{ \begin{array}{l} \Gamma \vdash A \; \mathrm{type} \quad \Gamma \vdash B \; \mathrm{type} \quad \Gamma \vdash R:A \simeq B \\ \Gamma, y:B \vdash C(y) \; \mathrm{type} \quad \Gamma \vdash c_R:\prod_{x:A} C(f_R(x)) \quad \Gamma \vdash a:A \end{array} }{\Gamma \vdash \beta_{A \simeq B}(R, c_R, a):\mathrm{ind}_{A \simeq B}^C(R, c_R, f_R(a)) =_{C(f_R(a))} c_R(a)}$$

Uniqueness rules for equivalence types:

$$\frac{ \begin{array}{l} \Gamma \vdash A \; \mathrm{type} \quad \Gamma \vdash B \; \mathrm{type} \quad \Gamma \vdash R:A \simeq B \\ \Gamma, y:B \vdash C(y) \; \mathrm{type} \quad \Gamma \vdash c:\prod_{y:B} C(y) \quad \Gamma \vdash b:B \end{array} }{\Gamma \vdash \eta_{A \simeq B}(R, c, b):c(b) =_{C(b)} \mathrm{ind}_{A \simeq B}^C(R, c, b)}$$

Coinductive definition

Given two types $A$ and $B$ and two functions $f:A \to B$ and $g:B \to A$, $f$ and $g$ are inverses of each other if for every element $a:A$ and $b:A$, there is an equivalence of types between $f(a) =_B b$ and $a =_A g(b)$:

$$\mathrm{isInv}(f, g) \coloneqq \prod_{a:A} \prod_{b:B} (f(a) =_B b) \simeq (a =_A g(b))$$

This leads to a coinductive definition of the type of equivalences

$$A \simeq B \coloneqq \sum_{f:A \to B} \sum_{g:B \to A} \prod_{a:A} \prod_{b:B} (f(a) =_B b) \simeq (a =_A g(b))$$

Formation rule for equivalence types:

$$\frac{\Gamma \vdash A \; \mathrm{type} \quad \Gamma \vdash B \; \mathrm{type}}{\Gamma \vdash A \simeq B \; \mathrm{type}}$$

Introduction rule for equivalence types:

$$\frac{\Gamma \vdash A \; \mathrm{type} \quad \Gamma \vdash a:A \quad \Gamma \vdash b:A \quad \Gamma \vdash p:a =_A b \quad \Gamma, x:A \vdash B(x) \; \mathrm{type}}{\Gamma \vdash \mathrm{tr}_B(a, b, p):B(a) \simeq B(b) \; \mathrm{type}}$$

Elimination rules for equivalence types:

$$\frac{\Gamma \vdash A \; \mathrm{type} \quad \Gamma \vdash B \; \mathrm{type} \quad \Gamma \vdash f:A \simeq B \quad \Gamma \vdash x:A}{\Gamma \vdash f(x):B} \qquad \frac{\Gamma \vdash A \; \mathrm{type} \quad \Gamma \vdash B \; \mathrm{type} \quad \Gamma \vdash f:A \simeq B \quad \Gamma \vdash y:B}{\Gamma \vdash f^{-1}(y):A}$$

$$\frac{\Gamma \vdash A \; \mathrm{type} \quad \Gamma \vdash B \; \mathrm{type} \quad \Gamma \vdash f:A \simeq B \quad \Gamma \vdash x:A \quad \Gamma \vdash y:B}{\Gamma\vdash \mathrm{coh}(f, x, y):(f(x) =_B y) \simeq (x =_A f^{-1}(y))}$$

Computation rules for equivalence types:

$$\frac{\Gamma \vdash A \; \mathrm{type} \quad \Gamma \vdash a:A \quad \Gamma \vdash b:A \quad \Gamma \vdash p:a =_A b \quad \Gamma, x:A \vdash B(x) \; \mathrm{type} \quad \Gamma, x:A \vdash w:B(x)}{\Gamma \vdash \mathrm{apdl}_B(a, b, p, w):w(a) =_{B(a)} \mathrm{tr}_B(a, b, p)^{-1}(w(b))}$$

$$\frac{\Gamma \vdash A \; \mathrm{type} \quad \Gamma \vdash a:A \quad \Gamma \vdash b:A \quad \Gamma \vdash p:a =_A b \quad \Gamma, x:A \vdash B(x) \; \mathrm{type} \quad \Gamma, x:A \vdash w:B(x)}{\Gamma \vdash \mathrm{apdr}_B(a, b, p, w):\mathrm{tr}_B(a, b, p)(w(a)) =_{B(b)} w(b)}$$

$$\frac{\Gamma \vdash A \; \mathrm{type} \quad \Gamma \vdash a:A \quad \Gamma \vdash b:A \quad \Gamma \vdash p:a =_A b \quad \Gamma, x:A \vdash B(x) \; \mathrm{type} \quad \Gamma, x:A \vdash w:B(x)}{\Gamma \vdash \mathrm{cohapd}_B(a, b, p, w):\left(\mathrm{coh}(\mathrm{tr}_B(a, b, p), w(a), w(b))(\mathrm{apdr}_B(a, b, p, w)) =_{w(a) =_{B(a)} \mathrm{tr}_B(a, b, p)^{-1}(w(b))} \mathrm{apdl}_B(a, b, p, w)\right) \simeq \left(\mathrm{apdr}_B(a, b, p, w) =_{\mathrm{tr}_B(a, b, p)(w(a)) =_{B(b)} w(b)} \mathrm{coh}(\mathrm{tr}_B(a, b, p), w(a), w(b))^{-1}(\mathrm{apdl}_B(a, b, p, w))\right)}$$

Properties

Relation to interval types

Given types $A$ and $B$ and an equivalence $f:A \simeq B$, one could define the dependent type $x:\mathbb{I} \vdash C(x)$ indexed by the interval type $\mathbb{I}$ as $C(0) \equiv A$, $C(1) \equiv B$, and $\mathrm{tr}_C(0, 1, p) \equiv f$.

One-to-one correspondences

Given types $A$ and $B$ and an equivalence $f:A \simeq B$, one could define a correspondence $x:A, y:B \vdash R(x, y)$ as the dependent identity type

$$R(x, y) \coloneqq x =_C^p y$$

where $x:\mathbb{I} \vdash C(x)$ is defined as in the previous section. By the properties of dependent identity types, the correspondence is always a one-to-one correspondence.

Quasi-inverse functions with contractible fibers

By the rules for function types, given an equivalence $R:A \simeq B$, one could derive functions $\rho(R):A \to B$ and $\lambda(R):B \to A$. One could show that these functions are quasi-inverse functions of each other: for all $x:A$ and $y:B$ and equivalences $R:A \simeq B$, there are identities

$$\rho_\kappa(R, \lambda(R, y), y, \lambda_\tau(R, y)):\rho(R)(\lambda(R)(y)) =_B y$$

$$\lambda_\kappa(R, x, \rho(R)(x), \rho_\tau(R, x)^{-1}):\lambda(R)(\rho(R)(x)) =_A x$$

where $p^{-1}:b =_A a$ is the inverse identity of $p:a =_A b$. By the introduction rule for dependent product types, there are homotopies

$$\lambda y.\rho_\kappa(R, \lambda(R, y), y, \lambda_\tau(R, y)):\prod_{y:B} \rho(R)(\lambda(R)(y)) =_B y$$

$$\lambda x.\lambda_\kappa(R, x, \rho(R)(x), \rho_\tau(R, x)^{-1}):\prod_{x:A} \lambda(R)(\rho(R)(x)) =_A x$$

which indicate that $\rho(R)$ and $\lambda(R)$ are quasi-inverse functions of each other.

By the rules for dependent sum types and dependent product types, one could show that the above functions each have contractible fibers, making both of them coherent inverse functions of each other.

Heterogeneous identity types

Given the definition of the equivalence type as the type of encodings for one-to-one correspondences, the heterogeneous identity type is defined by the rule

$$\frac{\Gamma \vdash A \; \mathrm{type} \quad \Gamma \vdash a:A \quad \Gamma \vdash b:A \quad \Gamma \vdash p:a =_A b \quad \Gamma, x:A \vdash B \; \mathrm{type}}{\Gamma \vdash (x =_B^p y) \equiv (x =_{B(a), B(b)}^{\mathrm{tr}_B(p)} y) \; \mathrm{type}}$$

Identity equivalences, inverse equivalences, and composition of equivalences

The identity equivalence on a type $A$ is defined as an equivalence $\mathrm{id}_A:A \simeq A$ such that for all elements $a:A$,

$$\lambda(\mathrm{id}_A, a) \coloneqq a$$

$$\rho(\mathrm{id}_A, a) \coloneqq a$$

$$(a =_{A, A}^{\mathrm{id}_A} b) \coloneqq (a =_A b)$$

Given an equivalence $R:A \simeq B$, the inverse equivalence of $R$ is an equivalence $R^{-1}:B \simeq A$ such that for all elements $a:A$ and $b:B$,

$$\rho(R^{-1}, a) \coloneqq \lambda(R, a)$$

$$\lambda(R^{-1}, b) \coloneqq \rho(R, b)$$

$$b =_{B, A}^{R^{-1}} a \coloneqq a =_{A, B}^R b$$

Given equivalences $R:A \simeq B$ and $S:B \simeq C$, the composite of $R$ and $S$ is an equivalence $S \circ R:A \simeq C$ such that for all elements $a:A$ and $c:C$,

$$\lambda(S \circ R, a) \coloneqq \lambda(R, \lambda(S, a))$$

$$\rho(S \circ R, c) \coloneqq \rho(S, \rho(R, c))$$

$$a =_{A, C}^{S \circ R} c \coloneqq \sum_{b:B} (a =_{A, B}^R b) \times (b =_{B, C}^R c)$$

Relation to universes and univalence

Given a Russell universe $U$, there are two ways to say that types $A:U$ and $B:U$ are equal: by the identity type $A =_U B$, and the equivalence type $A \simeq B$. The univalence axiom says that these two types $A =_U B$ and $A \simeq B$ are the same, which is represented by an equivalence between the two types

$$\mathrm{ua}(A, B):(A =_U B) \simeq (A \simeq B)$$

For Tarski universes $(U, \mathrm{El})$, one instead says that $A =_U B$ is the same as $\mathrm{El}(A) \simeq \mathrm{El}(B)$, represented as

$$\mathrm{ua}(A, B):(A =_U B) \simeq (\mathrm{El}(A) \simeq \mathrm{El}(B))$$

Action on equivalences

We introduce a modal operator $L$ to the type theory, which we assume in general not to be idempotent or monadic; this is given by the formation rule

$$\frac{\Gamma \vdash A \; \mathrm{type}}{\Gamma \vdash L(A) \; \mathrm{type}}$$

$L$ preserves equivalences: given types $A$ and $B$, there is a function $\mathrm{ae}_L:(A \simeq B) \to L(A) \simeq L(B)$, called the action on equivalences for $L$.

 Categorical semantics

The categorical semantics of an equivalence type is an object of isomorphisms.

 See also

• bijection set, object of isomorphisms

• equivalence in type theory

• identity type, heterogeneous identity type

• autoequivalence type

• function type, embedding type

• equivalence extensionality

 References

For the definition of the equivalence type as a dependent sum type, see:

• Univalent Foundations Project, §2.4 and §4 in: Homotopy Type Theory – Univalent Foundations of Mathematics (2013) [web, pdf]