This is the [Setoid.Overture.Inverses][] module of the agda-algebras library.
{-# OPTIONS --without-K --exact-split --safe #-} open import Relation.Binary using ( Setoid ) module Setoid.Overture.Inverses where -- {α ρᵃ β ρᵇ}{𝑨 : Setoid α ρᵃ}{𝑩 : Setoid β ρᵇ} -- Imports from Agda and the Agda Standard Library -------------------- open import Agda.Primitive using ( _⊔_ ; Level ) renaming ( Set to Type ) open import Function using ( id ) open import Function.Bundles using () renaming ( Func to _⟶_ ) open import Data.Product using ( _,_ ; Σ-syntax ) renaming ( proj₁ to fst ; proj₂ to snd ; _×_ to _∧_) open import Relation.Unary using ( Pred ; _∈_ ) open import Relation.Binary using ( _Preserves_⟶_ ) -- Imports from agda-algebras ----------------------------------------- open import Base.Overture.Preliminaries using ( ∣_∣ ; ∥_∥ ; ∃-syntax ) private variable α ρᵃ β ρᵇ : Level module _ {𝑨 : Setoid α ρᵃ}{𝑩 : Setoid β ρᵇ} where open Setoid 𝑨 using () renaming ( Carrier to A ; _≈_ to _≈₁_ ; refl to refl₁ ; sym to sym₁ ; trans to trans₁ ) open Setoid 𝑩 using () renaming ( Carrier to B ; _≈_ to _≈₂_ ; refl to refl₂ ; sym to sym₂ ; trans to trans₂ ) open _⟶_ {a = α}{ρᵃ}{β}{ρᵇ}{From = 𝑨}{To = 𝑩} renaming (f to _⟨$⟩_ )
We begin by defining two data types that represent the semantic concept of the image of a function. The first of these is for functions on bare types, while the second is for functions on setoids.
data Img_∋_ (f : A → B) : B → Type (α ⊔ β ⊔ ρᵇ) where eq : {b : B} → (a : A) → b ≈₂ (f a) → Img f ∋ b data Image_∋_ (F : 𝑨 ⟶ 𝑩) : B → Type (α ⊔ β ⊔ ρᵇ) where eq : {b : B} → (a : A) → b ≈₂ (F ⟨$⟩ a) → Image F ∋ b open Image_∋_ IsInRange : (𝑨 ⟶ 𝑩) → Pred B (α ⊔ ρᵇ) IsInRange F b = ∃[ a ∈ A ] (F ⟨$⟩ a) ≈₂ b Image⊆Range : ∀ {F b} → Image F ∋ b → b ∈ IsInRange F Image⊆Range (eq a x) = a , (sym₂ x) IsInRange→IsInImage : ∀ {F b} → b ∈ IsInRange F → Image F ∋ b IsInRange→IsInImage (a , x) = eq a (sym₂ x) Imagef∋f : ∀ {F a} → Image F ∋ (F ⟨$⟩ a) Imagef∋f = eq _ refl₂ -- Alternative representation of the range of a Func as a setoid -- the carrier _range : (𝑨 ⟶ 𝑩) → Type (α ⊔ β ⊔ ρᵇ) F range = Σ[ b ∈ B ] ∃[ a ∈ A ](F ⟨$⟩ a) ≈₂ b _image : (F : 𝑨 ⟶ 𝑩) → F range → B (F image) (b , (_ , _)) = b _preimage : (F : 𝑨 ⟶ 𝑩) → F range → A (F preimage) (_ , (a , _)) = a f∈range : ∀ {F} → A → F range f∈range {F} a = (F ⟨$⟩ a) , (a , refl₂) ⌜_⌝ : (F : 𝑨 ⟶ 𝑩) → A → F range ⌜ F ⌝ a = f∈range{F} a Ran : (𝑨 ⟶ 𝑩) → Setoid (α ⊔ β ⊔ ρᵇ) ρᵇ Ran F = record { Carrier = F range ; _≈_ = λ x y → ((F image) x) ≈₂ ((F image) y) ; isEquivalence = record { refl = refl₂ ; sym = sym₂ ; trans = trans₂ } } RRan : (𝑨 ⟶ 𝑩) → Setoid (α ⊔ β ⊔ ρᵇ) (ρᵃ ⊔ ρᵇ) RRan F = record { Carrier = F range ; _≈_ = λ x y → ((F preimage) x) ≈₁ ((F preimage) y) ∧ ((F image) x) ≈₂ ((F image) y) ; isEquivalence = record { refl = refl₁ , refl₂ ; sym = λ x → (sym₁ ∣ x ∣) , (sym₂ ∥ x ∥) ; trans = λ x y → (trans₁ ∣ x ∣ ∣ y ∣) , (trans₂ ∥ x ∥ ∥ y ∥) } } _preimage≈image : ∀ F r → F ⟨$⟩ (F preimage) r ≈₂ (F image) r (F preimage≈image) (_ , (_ , p)) = p Dom : (𝑨 ⟶ 𝑩) → Setoid α ρᵇ Dom F = record { Carrier = A ; _≈_ = λ x y → F ⟨$⟩ x ≈₂ F ⟨$⟩ y ; isEquivalence = record { refl = refl₂ ; sym = sym₂ ; trans = trans₂ } }
An inhabitant of Image f ∋ b is a dependent pair (a , p), where a : A and p : b ≡ f a is a proof that f maps a to b. Since the proof that b belongs to the image of f is always accompanied by a witness a : A, we can actually compute a (pseudo)inverse of f. For convenience, we define this inverse function, which we call Inv, and which takes an arbitrary b : B and a (witness, proof)-pair, (a , p) : Image f ∋ b, and returns the witness a.
inv : (f : A → B){b : B} → Img f ∋ b → A inv _ (eq a _) = a Inv : (F : 𝑨 ⟶ 𝑩){b : B} → Image F ∋ b → A Inv _ (eq a _) = a Inv' : (F : 𝑨 ⟶ 𝑩){b : B} → b ∈ IsInRange F → A Inv' _ (a , _) = a [_]⁻¹ : (F : 𝑨 ⟶ 𝑩) → F range → A [ F ]⁻¹ = F preimage ⟦_⟧⁻¹ : (F : 𝑨 ⟶ 𝑩) → Ran F ⟶ Dom F ⟦ F ⟧⁻¹ = record { f = F preimage ; cong = λ {x}{y} → c{x}{y} } where c : (F preimage) Preserves (Setoid._≈_ (Ran F)) ⟶ (Setoid._≈_ (Dom F)) c {x}{y} ix≈iy = Goal where Goal : F ⟨$⟩ ((F preimage) x) ≈₂ F ⟨$⟩ ((F preimage) y) Goal = trans₂ ((F preimage≈image) x) (trans₂ ix≈iy (sym₂ ((F preimage≈image) y)))
We can prove that Inv f is the range-restricted right-inverse of f, as follows.
invIsInvʳ : {f : A → B}{b : B}(q : Img f ∋ b) → (f (inv f q)) ≈₂ b invIsInvʳ (eq _ p) = sym₂ p InvIsInverseʳ : {F : 𝑨 ⟶ 𝑩}{b : B}(q : Image F ∋ b) → (F ⟨$⟩ (Inv F q)) ≈₂ b InvIsInverseʳ (eq _ p) = sym₂ p ⁻¹IsInverseʳ : {F : 𝑨 ⟶ 𝑩}{bap : F range} → (F ⟨$⟩ ([ F ]⁻¹ bap )) ≈₂ ∣ bap ∣ ⁻¹IsInverseʳ {bap = (_ , (_ , p))} = p
Of course, the “range-restricted” qualifier is needed because Inf f is not defined outside the range of f.
In a certain sense, Inv f is also a (range-restricted) left-inverse.
InvIsInverseˡ : ∀ {F a} → Inv F {b = F ⟨$⟩ a} Imagef∋f ≈₁ a InvIsInverseˡ = refl₁ ⁻¹IsInverseˡ : ∀ {F a} → [ F ]⁻¹ (f∈range{F} a) ≈₁ a ⁻¹IsInverseˡ = refl₁