Tope disjuction elimination along identity paths¶
\(\mathsf{rec}_{\lor}^{\psi,\phi}(a_\psi, a_\phi)\) (written recOR(psi, phi, a_psi, a_phi)
in the code)
is well-typed when \(a_\psi\) and \(a_\phi\) are definitionally equal on \(\psi \land \phi\).
Sometimes this is too strong since many terms are not definitionally equal, but only equal up to a path.
Luckily, assuming relative function extensionality, we can define a weaker version of \(rec_{\lor}\) (recOR
), which we call recId
, that can work in presence of a witness of type \(\prod_{t : I \mid \psi \land \phi} a_\psi = a_\phi\).
Prerequisites¶
We begin by introducing common HoTT definitions:
#lang rzk-1
-- A is contractible there exists x : A such that for any y : A we have x = y.
#define iscontr (A : U) : U
:= ∑ (a : A), (x : A) -> a =_{A} x
-- A is a proposition if for any x, y : A we have x = y
#define isaprop (A : U) : U
:= (x : A) -> (y : A) -> x =_{A} y
-- A is a set if for any x, y : A the type x =_{A} y is a proposition
#define isaset (A : U) : U
:= (x : A) -> (y : A) -> isaprop (x =_{A} y)
-- Non-dependent product of A and B
#define prod (A : U) (B : U) : U
:= ∑ (x : A), B
-- A function f : A -> B is an equivalence
-- if there exists g : B -> A
-- such that for all x : A we have g (f x) = x
-- and for all y : B we have f (g y) = y
#define isweq (A : U) (B : U) (f : A -> B) : U
:= ∑ (g : B -> A), prod ((x : A) -> g (f x) =_{A} x) ((y : B) -> f (g y) =_{B} y)
-- Equivalence of types A and B
#define weq (A : U) (B : U) : U
:= ∑ (f : A -> B), isweq A B f
-- Transport along a path
#define transport
(A : U)
(C : A -> U)
(x y : A)
(p : x =_{A} y)
: C x -> C y
:= \cx -> idJ(A, x, (\z q -> C z), cx, y, p)
Relative function extensionality¶
We can now define relative function extensionality. There are several formulations, we provide two, following Riehl and Shulman:
-- [RS17, Axiom 4.6] Relative function extensionality.
#define relfunext : U
:= (I : CUBE)
-> (psi : I -> TOPE)
-> (phi : psi -> TOPE)
-> (A : psi -> U)
-> ((t : psi) -> iscontr (A t))
-> (a : (t : phi) -> A t)
-> (t : psi) -> A t [ phi t |-> a t]
-- [RS17, Proposition 4.8] A (weaker) formulation of function extensionality.
#define relfunext2 : U
:= (I : CUBE)
-> (psi : I -> TOPE)
-> (phi : psi -> TOPE)
-> (A : psi -> U)
-> (a : (t : phi) -> A t)
-> (f : (t : psi) -> A t [ phi t |-> a t ])
-> (g : (t : psi) -> A t [ phi t |-> a t ])
-> weq (f = g)
((t : psi) -> (f t =_{A t} g t) [ phi t |-> refl ])
Construction of recId
¶
The idea is straightforward. We ask for a proof that a = b
for all points in psi /\ phi
. Then, by relative function extensionality (relfunext2
), we can show that restrictions of a
and b
to psi /\ phi
are equal. If we reformulate a
as extension of its restriction, then we can transport
such reformulation along the path connecting two restrictions and apply recOR
.
First, we define how to restrict an extension type to a subshape:
#section construction-of-recId
#variable r : relfunext2
#variable I : CUBE
#variables psi phi : I -> TOPE
#variable A : {t : I | psi t \/ phi t} -> U
-- Restrict extension type to a subshape.
#define restrict_phi
(a : {t : I | phi t} -> A t)
: {t : I | psi t /\ phi t} -> A t
:= \t -> a t
-- Restrict extension type to a subshape.
#define restrict_psi
(a : {t : I | psi t} -> A t)
: {t : I | psi t /\ phi t} -> A t
:= \t -> a t
Then, how to reformulate an a
(or b
) as an extension of its restriction:
-- Reformulate extension type as an extension of a restriction.
#define ext-of-restrict_psi
(a : {t : I | psi t} -> A t)
: (t : psi) -> A t [ psi t /\ phi t |-> restrict_psi a t ]
:= a -- type is coerced automatically here
-- Reformulate extension type as an extension of a restriction.
#define ext-of-restrict_phi
(a : {t : I | phi t} -> A t)
: (t : phi) -> A t [ psi t /\ phi t |-> restrict_phi a t ]
:= a -- type is coerced automatically here
Now, assuming relative function extensionality, we construct a path between restrictions:
-- Transform extension of an identity into an identity of restrictions.
#define restricts-path
(a_psi : (t : psi) -> A t)
(a_phi : (t : phi) -> A t)
(e : {t : I | psi t /\ phi t} -> a_psi t = a_phi t)
: restrict_psi a_psi = restrict_phi a_phi
:= (first (second (r I
(\t -> psi t /\ phi t)
(\t -> BOT)
(\t -> A t)
(\t -> recBOT)
(\t -> a_psi t)
(\t -> a_phi t)))) e
Finally, we bring everything together into recId
:
-- A weaker version of recOR, demanding only a path between a and b:
-- recOR(psi, phi, a, b) demands that for psi /\ phi we have a == b (definitionally)
-- (recId psi phi a b e) demands that e is the proof that a = b (intensionally) for psi /\ phi
#define recId uses (r) -- we declare that recId is using r on purpose
(a_psi : (t : psi) -> A t)
(a_phi : (t : phi) -> A t)
(e : {t : I | psi t /\ phi t} -> a_psi t = a_phi t)
: {t : I | psi t \/ phi t} -> A t
:= \t -> recOR(
psi t |-> transport
({s : I | psi s /\ phi s} -> A s)
(\ra -> (s : psi) -> A s [ psi s /\ phi s |-> ra s ])
(restrict_psi a_psi)
(restrict_phi a_phi)
(restricts-path a_psi a_phi e)
(ext-of-restrict_psi a_psi)
t,
phi t |-> ext-of-restrict_phi a_phi t
)
#end construction-of-recId
Gluing extension types¶
An application of of recId
is gluing together extension types,
whenever we can show that they are equal on the intersection of shapes:
-- If two extension types are equal along two subshapes,
-- then they are also equal along their union.
#define id-along-border
(r : relfunext2)
(I : CUBE)
(psi : I -> TOPE)
(phi : I -> TOPE)
(A : {t : I | psi t \/ phi t} -> U)
(a b : {t : I | psi t \/ phi t} -> A t)
(e_psi : (t : psi) -> a t = b t)
(e_phi : (t : phi) -> a t = b t)
(border-is-a-set : {t : I | psi t /\ phi t} -> isaset (A t))
: {t : I | psi t \/ phi t} -> a t = b t
:= recId r I psi phi
(\t -> a t = b t)
e_psi e_phi
(\t -> border-is-a-set t (a t) (b t) (e_psi t) (e_phi t))