/- Copyright (c) 2014 Jakob von Raumer. All rights reserved. Released under Apache 2.0 license as described in the file LICENSE. Authors: Jakob von Raumer, Floris van Doorn Ported from Coq HoTT -/ prelude import .trunc .equiv .ua open eq is_trunc sigma function is_equiv equiv prod unit prod.ops lift /- We now prove that funext follows from a couple of weaker-looking forms of function extensionality. This proof is originally due to Voevodsky; it has since been simplified by Peter Lumsdaine and Michael Shulman. -/ definition funext.{l k} := Π ⦃A : Type.{l}⦄ {P : A → Type.{k}} (f g : Π x, P x), is_equiv (@apd10 A P f g) -- Naive funext is the simple assertion that pointwise equal functions are equal. definition naive_funext := Π ⦃A : Type⦄ {P : A → Type} (f g : Πx, P x), (f ~ g) → f = g -- Weak funext says that a product of contractible types is contractible. definition weak_funext := Π ⦃A : Type⦄ (P : A → Type) [H: Πx, is_contr (P x)], is_contr (Πx, P x) definition weak_funext_of_naive_funext : naive_funext → weak_funext := (λ nf A P (Pc : Πx, is_contr (P x)), let c := λx, center (P x) in is_contr.mk c (λ f, have eq' : (λx, center (P x)) ~ f, from (λx, center_eq (f x)), have eq : (λx, center (P x)) = f, from nf A P (λx, center (P x)) f eq', eq ) ) /- The less obvious direction is that weak_funext implies funext (and hence all three are logically equivalent). The point is that under weak funext, the space of "pointwise homotopies" has the same universal property as the space of paths. -/ section universe variables l k parameters [wf : weak_funext.{l k}] {A : Type.{l}} {B : A → Type.{k}} (f : Π x, B x) definition is_contr_sigma_homotopy : is_contr (Σ (g : Π x, B x), f ~ g) := is_contr.mk (sigma.mk f (homotopy.refl f)) (λ dp, sigma.rec_on dp (λ (g : Π x, B x) (h : f ~ g), let r := λ (k : Π x, Σ y, f x = y), @sigma.mk _ (λg, f ~ g) (λx, pr1 (k x)) (λx, pr2 (k x)) in let s := λ g h x, @sigma.mk _ (λy, f x = y) (g x) (h x) in have t1 : Πx, is_contr (Σ y, f x = y), from (λx, !is_contr_sigma_eq), have t2 : is_contr (Πx, Σ y, f x = y), from !wf, have t3 : (λ x, @sigma.mk _ (λ y, f x = y) (f x) idp) = s g h, from @eq_of_is_contr (Π x, Σ y, f x = y) t2 _ _, have t4 : r (λ x, sigma.mk (f x) idp) = r (s g h), from ap r t3, have endt : sigma.mk f (homotopy.refl f) = sigma.mk g h, from t4, endt ) ) local attribute is_contr_sigma_homotopy [instance] parameters (Q : Π g (h : f ~ g), Type) (d : Q f (homotopy.refl f)) definition homotopy_ind (g : Πx, B x) (h : f ~ g) : Q g h := @transport _ (λ gh, Q (pr1 gh) (pr2 gh)) (sigma.mk f (homotopy.refl f)) (sigma.mk g h) (@eq_of_is_contr _ is_contr_sigma_homotopy _ _) d local attribute weak_funext [reducible] local attribute homotopy_ind [reducible] definition homotopy_ind_comp : homotopy_ind f (homotopy.refl f) = d := (@prop_eq_of_is_contr _ _ _ _ !eq_of_is_contr idp)⁻¹ ▸ idp end /- Now the proof is fairly easy; we can just use the same induction principle on both sides. -/ section universe variables l k local attribute weak_funext [reducible] theorem funext_of_weak_funext (wf : weak_funext.{l k}) : funext.{l k} := λ A B f g, let eq_to_f := (λ g' x, f = g') in let sim2path := homotopy_ind f eq_to_f idp in have t1 : sim2path f (homotopy.refl f) = idp, proof homotopy_ind_comp f eq_to_f idp qed, have t2 : apd10 (sim2path f (homotopy.refl f)) = (homotopy.refl f), proof ap apd10 t1 qed, have left_inv : apd10 ∘ (sim2path g) ~ id, proof (homotopy_ind f (λ g' x, apd10 (sim2path g' x) = x) t2) g qed, have right_inv : (sim2path g) ∘ apd10 ~ id, from (λ h, eq.rec_on h (homotopy_ind_comp f _ idp)), is_equiv.adjointify apd10 (sim2path g) left_inv right_inv definition funext_from_naive_funext : naive_funext → funext := compose funext_of_weak_funext weak_funext_of_naive_funext end section universe variables l private theorem ua_isequiv_postcompose {A B : Type.{l}} {C : Type} {w : A → B} [H0 : is_equiv w] : is_equiv (@compose C A B w) := let w' := equiv.mk w H0 in let eqinv : A = B := ((@is_equiv.inv _ _ _ (univalence A B)) w') in let eq' := equiv_of_eq eqinv in is_equiv.adjointify (@compose C A B w) (@compose C B A (is_equiv.inv w)) (λ (x : C → B), have eqretr : eq' = w', from (@right_inv _ _ (@equiv_of_eq A B) (univalence A B) w'), have invs_eq : (to_fun eq')⁻¹ = (to_fun w')⁻¹, from inv_eq eq' w' eqretr, have eqfin : (to_fun eq') ∘ ((to_fun eq')⁻¹ ∘ x) = x, from (λ p, (@eq.rec_on Type.{l} A (λ B' p', Π (x' : C → B'), (to_fun (equiv_of_eq p')) ∘ ((to_fun (equiv_of_eq p'))⁻¹ ∘ x') = x') B p (λ x', idp)) ) eqinv x, have eqfin' : (to_fun w') ∘ ((to_fun eq')⁻¹ ∘ x) = x, from eqretr ▸ eqfin, have eqfin'' : (to_fun w') ∘ ((to_fun w')⁻¹ ∘ x) = x, from invs_eq ▸ eqfin', eqfin'' ) (λ (x : C → A), have eqretr : eq' = w', from (@right_inv _ _ (@equiv_of_eq A B) (univalence A B) w'), have invs_eq : (to_fun eq')⁻¹ = (to_fun w')⁻¹, from inv_eq eq' w' eqretr, have eqfin : (to_fun eq')⁻¹ ∘ ((to_fun eq') ∘ x) = x, from (λ p, eq.rec_on p idp) eqinv, have eqfin' : (to_fun eq')⁻¹ ∘ ((to_fun w') ∘ x) = x, from eqretr ▸ eqfin, have eqfin'' : (to_fun w')⁻¹ ∘ ((to_fun w') ∘ x) = x, from invs_eq ▸ eqfin', eqfin'' ) -- We are ready to prove functional extensionality, -- starting with the naive non-dependent version. private definition diagonal [reducible] (B : Type) : Type := Σ xy : B × B, pr₁ xy = pr₂ xy private definition isequiv_src_compose {A B : Type} : @is_equiv (A → diagonal B) (A → B) (compose (pr₁ ∘ pr1)) := @ua_isequiv_postcompose _ _ _ (pr₁ ∘ pr1) (is_equiv.adjointify (pr₁ ∘ pr1) (λ x, sigma.mk (x , x) idp) (λx, idp) (λ x, sigma.rec_on x (λ xy, prod.rec_on xy (λ b c p, eq.rec_on p idp)))) private definition isequiv_tgt_compose {A B : Type} : is_equiv (compose (pr₂ ∘ pr1) : (A → diagonal B) → (A → B)) := begin refine @ua_isequiv_postcompose _ _ _ (pr2 ∘ pr1) _, fapply adjointify, { intro b, exact ⟨(b, b), idp⟩}, { intro b, reflexivity}, { intro a, induction a with q p, induction q, esimp at *, induction p, reflexivity} end theorem nondep_funext_from_ua {A : Type} {B : Type} : Π {f g : A → B}, f ~ g → f = g := (λ (f g : A → B) (p : f ~ g), let d := λ (x : A), @sigma.mk (B × B) (λ (xy : B × B), xy.1 = xy.2) (f x , f x) (eq.refl (f x, f x).1) in let e := λ (x : A), @sigma.mk (B × B) (λ (xy : B × B), xy.1 = xy.2) (f x , g x) (p x) in let precomp1 := compose (pr₁ ∘ sigma.pr1) in have equiv1 : is_equiv precomp1, from @isequiv_src_compose A B, have equiv2 : Π (x y : A → diagonal B), is_equiv (ap precomp1), from is_equiv.is_equiv_ap precomp1, have H' : Π (x y : A → diagonal B), pr₁ ∘ pr1 ∘ x = pr₁ ∘ pr1 ∘ y → x = y, from (λ x y, is_equiv.inv (ap precomp1)), have eq2 : pr₁ ∘ pr1 ∘ d = pr₁ ∘ pr1 ∘ e, from idp, have eq0 : d = e, from H' d e eq2, have eq1 : (pr₂ ∘ pr1) ∘ d = (pr₂ ∘ pr1) ∘ e, from ap _ eq0, eq1 ) end -- Now we use this to prove weak funext, which as we know -- implies (with dependent eta) also the strong dependent funext. theorem weak_funext_of_ua : weak_funext := (λ (A : Type) (P : A → Type) allcontr, let U := (λ (x : A), lift unit) in have pequiv : Π (x : A), P x ≃ unit, from (λ x, @equiv_unit_of_is_contr (P x) (allcontr x)), have psim : Π (x : A), P x = U x, from (λ x, eq_of_equiv_lift (pequiv x)), have p : P = U, from @nondep_funext_from_ua A Type P U psim, have tU' : is_contr (A → lift unit), from is_contr.mk (λ x, up ⋆) (λ f, nondep_funext_from_ua (λa, by induction (f a) with u;induction u;reflexivity)), have tU : is_contr (Π x, U x), from tU', have tlast : is_contr (Πx, P x), from p⁻¹ ▸ tU, tlast) -- we have proven function extensionality from the univalence axiom definition funext_of_ua : funext := funext_of_weak_funext (@weak_funext_of_ua) /- We still take funext as an axiom, so that when you write "print axioms foo", you can see whether it uses only function extensionality, and not also univalence. -/ axiom function_extensionality : funext variables {A : Type} {P : A → Type} {f g : Π x, P x} namespace funext theorem is_equiv_apdt [instance] (f g : Π x, P x) : is_equiv (@apd10 A P f g) := function_extensionality f g end funext open funext definition eq_equiv_homotopy : (f = g) ≃ (f ~ g) := equiv.mk apd10 _ definition eq_of_homotopy [reducible] : f ~ g → f = g := (@apd10 A P f g)⁻¹ definition apd10_eq_of_homotopy (p : f ~ g) : apd10 (eq_of_homotopy p) = p := right_inv apd10 p definition eq_of_homotopy_apd10 (p : f = g) : eq_of_homotopy (apd10 p) = p := left_inv apd10 p definition eq_of_homotopy_idp (f : Π x, P x) : eq_of_homotopy (λx : A, idpath (f x)) = idpath f := is_equiv.left_inv apd10 idp definition naive_funext_of_ua : naive_funext := λ A P f g h, eq_of_homotopy h protected definition homotopy.rec_on [recursor] {Q : (f ~ g) → Type} (p : f ~ g) (H : Π(q : f = g), Q (apd10 q)) : Q p := right_inv apd10 p ▸ H (eq_of_homotopy p) protected definition homotopy.rec_on_idp [recursor] {Q : Π{g}, (f ~ g) → Type} {g : Π x, P x} (p : f ~ g) (H : Q (homotopy.refl f)) : Q p := homotopy.rec_on p (λq, eq.rec_on q H) definition eq_of_homotopy_inv {f g : Π x, P x} (H : f ~ g) : eq_of_homotopy (λx, (H x)⁻¹) = (eq_of_homotopy H)⁻¹ := begin apply homotopy.rec_on_idp H, rewrite [+eq_of_homotopy_idp] end definition eq_of_homotopy_con {f g h : Π x, P x} (H1 : f ~ g) (H2 : g ~ h) : eq_of_homotopy (λx, H1 x ⬝ H2 x) = eq_of_homotopy H1 ⬝ eq_of_homotopy H2 := begin apply homotopy.rec_on_idp H1, apply homotopy.rec_on_idp H2, rewrite [+eq_of_homotopy_idp] end