TotallySeparatedReflectionOfOmega

Martin Escardo, 17-19 June 2026.

The totally separated reflection of the type Ω of propositions.

Any type X has a totally separated reflection, given by the image of
the evaluation map X → ((X → 𝟚) → 𝟚). Here we explore whether the
totally separated reflection of Ω has a more direct description.

First, we show, assuming propositional resizing, that the type

    T := (WEM → 𝟚)

has the universal property of the totally separated reflection of Ω,
where

    WEM := (p : Ω) → ¬ (p holds) + ¬¬ (p holds)

is the principle of weak excluded middle.

The unit η : Ω → T of the reflection sends a proposition p to the
function that, given a witness of WEM, gives ₀ or ₁ according to
whether ¬ p holds or ¬¬ p holds.

The universal property says that, for every totally separated type Y,
precomposition with η (the restriction map) is an equivalence

    (T → Y) ≃ (Ω → Y).

Resizing is used to define a section s : T → Ω of η by
s t = "the resized proposition that t is the constant function ₁".

Second, we ask whether this equivalence can be established without
assuming propositional resizing.

We don't know, but we explore this a bit here. In particular, we
establish the equivalence, without resizing, for types Y that are
retracts of powers of 𝟚.

TODO. Is every totally separated type a retract of a power of 𝟚,
without assuming resizing? No, because this excludes the empty type
(as pointed out to us by Jason Carr). But what can we say in this
direction?

A side-conclusion of this technical development is that we have an
equivalence

  (Ω → 𝟚) ≃ (𝟚 + WEM × 𝟚).

There are always two maps Ω → 𝟚, namely the constant ones, plus two
when WEM holds.

Moreover, we show that η is the universal map of Ω into a totally
separated type if and only if it is a surjection.


{-# OPTIONS --safe --without-K --lossy-unification #-}

open import UF.FunExt
open import UF.Subsingletons

open import MLTT.Spartan

module gist.TotallySeparatedReflectionOfOmega
        (fe : Fun-Ext)
        (pe : Prop-Ext)
        (𝓤 : Universe)
       where

open import MLTT.Two-Properties
open import TypeTopology.CompactTypes
open import TypeTopology.MicroTychonoff
open import TypeTopology.TotallySeparated
open import TypeTopology.SigmaTotallySeparated
open import UF.Base
open import UF.ClassicalLogic
             using (EM ; LEM ; EM-gives-LEM ; double-negation-of-decision)
open import UF.DiscreteAndSeparated
open import UF.Embeddings
open import UF.Equiv
open import UF.PropTrunc
open import UF.Retracts
open import UF.Sets
open import UF.Size
open import UF.SubtypeClassifier renaming (Ω to Ω-of)
open import UF.Subsingletons-FunExt

private
 fe' : FunExt
 fe' 𝓤 𝓥 = fe {𝓤} {𝓥}

 𝓤⁺ = 𝓤 ⁺

 Ω : 𝓤⁺ ̇
 Ω = Ω-of 𝓤

 WEM : 𝓤⁺ ̇
 WEM = (p : Ω) → is-decidable (¬ (p holds))

 WEM-is-prop : is-prop WEM
 WEM-is-prop = Π-is-prop fe
                (λ p → decidability-of-prop-is-prop fe (negations-are-props fe))

T : 𝓤⁺ ̇
T = WEM → 𝟚

T-is-totally-separated : is-totally-separated T
T-is-totally-separated = Π-is-totally-separated fe
                          (λ _ → 𝟚-is-totally-separated)

T-is-set : is-set T
T-is-set = totally-separated-types-are-sets fe T
            T-is-totally-separated

τ : 𝟚 → T
τ b = λ _ → b

τ₀ τ₁ : T
τ₀ = τ ₀
τ₁ = τ ₁


NB. The function δ needs --lossy-unification to avoid unsolved constraints.


δ : {p : Ω} → is-decidable (¬ (p holds)) → 𝟚
δ (inl _) = ₀
δ (inr _) = ₁


The unit of the reflection and its non-definitional "computation" rules.


η : Ω → T
η p w = δ (w p)

η₀ : (p : Ω) (w : WEM) → ¬ (p holds) → η p w ＝ ₀
η₀ p w ph = I (w p)
 where
  I : (d : is-decidable (¬ (p holds))) → δ d ＝ ₀
  I (inl _) = refl
  I (inr ν) = 𝟘-elim (ν ph)

η₁ : (p : Ω) (w : WEM) → ¬¬ (p holds) → η p w ＝ ₁
η₁ p w ν = I (w p)
 where
  I : (d : is-decidable (¬ (p holds))) → δ d ＝ ₁
  I (inl ph) = 𝟘-elim (ν ph)
  I (inr _)  = refl

η⊥ : η ⊥ ＝ τ₀
η⊥ = dfunext fe (λ w → η₀ ⊥ w ⊥-doesnt-hold)

η⊤ : η ⊤ ＝ τ₁
η⊤ = dfunext fe (λ w → η₁ ⊤ w (¬¬-intro ⊤-holds))

τ-lemma : (t : T) (w : WEM) → t ＝ τ (t w)
τ-lemma t w = dfunext fe (λ w' → ap t (WEM-is-prop w' w))


Sufficient condition for boolean-valued maps on Ω being constant.


lemma-⊥ : (f : Ω → 𝟚) (p : Ω) → ¬ (p holds) → f p ＝ f ⊥
lemma-⊥ f p ν = ap f (fails-gives-equal-⊥ pe fe p ν)

lemma-⊤ : (f : Ω → 𝟚) (p : Ω) → p holds → f p ＝ f ⊤
lemma-⊤ f p ph = ap f (holds-gives-equal-⊤ pe fe p ph)


Given f : Ω → 𝟚, we can decide whether f ⊥ ＝ f ⊤ or not.

 * If so, then f is constant.
 * Otherwise, WEM follows.


constancy-lemma : (f : Ω → 𝟚) → f ⊥ ＝ f ⊤ → (p : Ω) → f p ＝ f ⊥
constancy-lemma f e p = 𝟚-is-¬¬-separated (f p) (f ⊥) I
 where
  I : ¬¬ (f p ＝ f ⊥)
  I ne = I₁ I₀
   where
    I₀ : ¬ (p holds)
    I₀ ph = ne (f p ＝⟨ lemma-⊤ f p ph ⟩
                f ⊤ ＝⟨ e ⁻¹ ⟩
                f ⊥ ∎)

    I₁ : ¬¬ (p holds)
    I₁ ν = ne (lemma-⊥ f p ν)

WEM-lemma : (f : Ω → 𝟚) → f ⊥ ≠ f ⊤ → WEM
WEM-lemma f ne p = I (𝟚-is-discrete (f p) (f ⊤))
 where
  I : is-decidable (f p ＝ f ⊤) → is-decidable (¬ (p holds))
  I (inl e)   = inr (λ ph → ne (f ⊥ ＝⟨ (lemma-⊥ f p ph)⁻¹ ⟩
                                f p ＝⟨ e ⟩
                                f ⊤ ∎))
  I (inr ne') = inl (λ ph → ne' (lemma-⊤ f p ph))


Some observations not needed for our development:


_ : (Y : 𝓥 ̇ ) → is-totally-separated Y → is-¬¬-separated Y
_ = totally-separated-types-are-¬¬-separated

_ : {Y : 𝓥 ̇ }
  → is-¬¬-separated Y
  → (f : Ω → Y) → f ⊥ ＝ f ⊤ → (p q : Ω) → f p ＝ f q
_ = ⊥-⊤-density' fe pe

¬WEM-observation : ¬ WEM
                 → {Y : 𝓥 ̇ }
                 → is-totally-separated Y
                 → (f : Ω → Y) (p q : Ω) → f p ＝ f q
¬WEM-observation nwem {Y} ts f = III
 where
  I : (g : Y → 𝟚) → g (f ⊥) ＝ g (f ⊤)
  I g = 𝟚-is-¬¬-separated (g (f ⊥)) (g (f ⊤)) I₀
   where
    I₀ : ¬¬ (g (f ⊥) ＝ g (f ⊤))
    I₀ ne = nwem (WEM-lemma (g ∘ f) ne)

  II : f ⊥ ＝ f ⊤
  II = ts I

  III : (p q : Ω) → f p ＝ f q
  III = ⊥-⊤-density' fe pe (totally-separated-types-are-¬¬-separated Y ts) f II

¬¬WEM-observation : (Y : 𝓥 ̇ )
                  → is-totally-separated Y
                  → (Σ f ꞉ (Ω → Y) , f ⊥ ≠ f ⊤)
                  → ¬¬ WEM
¬¬WEM-observation Y ts (f , ne) =
 contrapositive
  (λ (nwem : ¬ WEM) → ¬WEM-observation nwem ts f ⊥ ⊤)
  ne

τ₀＝τ₁-gives-¬WEM : τ₀ ＝ τ₁ → ¬ WEM
τ₀＝τ₁-gives-¬WEM e w = zero-is-not-one (happly e w)

¬¬WEM-gives-τ₀≠τ₁ : ¬¬ WEM → τ₀ ≠ τ₁
¬¬WEM-gives-τ₀≠τ₁ = contrapositive τ₀＝τ₁-gives-¬WEM

τ₀≠τ₁-gives-¬¬WEM : τ₀ ≠ τ₁ → ¬¬ WEM
τ₀≠τ₁-gives-¬¬WEM ne nw =
 ¬¬WEM-observation T T-is-totally-separated (η , I) nw
 where
  I : η ⊥ ≠ η ⊤
  I e = ne (τ₀  ＝⟨ η⊥ ⁻¹ ⟩
            η ⊥ ＝⟨ e ⟩
            η ⊤ ＝⟨ η⊤ ⟩
            τ₁  ∎)

¬WEM-gives-τ₀＝τ₁ : ¬ WEM → τ₀ ＝ τ₁
¬WEM-gives-τ₀＝τ₁ nw =
 totally-separated-types-are-¬¬-separated T T-is-totally-separated τ₀ τ₁ I
 where
  I : ¬¬ (τ₀ ＝ τ₁)
  I ne = τ₀≠τ₁-gives-¬¬WEM ne nw


Restriction along η:


ρ : (Y : 𝓥 ̇ ) → (T → Y) → (Ω → Y)
ρ Y g = g ∘ η


We now show that T is the totally separated reflection of Ω assuming
resizing, and after that we record everything we know about the
universal property of T without assuming resizing.


module T-is-ts-reflection-of-Ω-assuming-resizing
        (r : propositional-resizing 𝓤⁺ 𝓤)
       where

 being-equal-to-τ₁-is-prop : (t : T) → is-prop (t ＝ τ₁)
 being-equal-to-τ₁-is-prop t = T-is-set


We apply resizing to the proposition (t ＝ τ₁), to show that T is a
retract of Ω with a section s of η.


 s : T → Ω
 s t = resize r (t ＝ τ₁) (being-equal-to-τ₁-is-prop t) ,
       resize-is-prop r (t ＝ τ₁) (being-equal-to-τ₁-is-prop t)

 to-s-holds : (t : T) → (t ＝ τ₁) → s t holds
 to-s-holds t = to-resize r (t ＝ τ₁) (being-equal-to-τ₁-is-prop t)

 from-s-holds : (t : T) → s t holds → (t ＝ τ₁)
 from-s-holds t = from-resize r (t ＝ τ₁) (being-equal-to-τ₁-is-prop t)

 ηs : (t : T) → η (s t) ＝ t
 ηs t = dfunext fe (λ w → 𝟚-equality-cases (I w) (II w))
  where
   I : (w : WEM) → t w ＝ ₀ → η (s t) w ＝ t w
   I w e₀ = η (s t) w ＝⟨ η₀ (s t) w I₀ ⟩
            ₀         ＝⟨ e₀ ⁻¹ ⟩
            t w       ∎
    where
     I₀ : ¬ (s t holds)
     I₀ h = zero-is-not-one
             (₀     ＝⟨ e₀ ⁻¹ ⟩
              t  w  ＝⟨ happly (from-s-holds t h) w ⟩
              τ₁ w  ＝⟨ refl ⟩
              ₁     ∎)
   II : (w : WEM) → t w ＝ ₁ → η (s t) w ＝ t w
   II w e₁ = η (s t) w ＝⟨ η₁ (s t) w II₁ ⟩
             ₁         ＝⟨ e₁ ⁻¹ ⟩
             t w       ∎
    where
     II₀ : t ＝ τ₁
     II₀ = t       ＝⟨ τ-lemma t w ⟩
           τ (t w) ＝⟨ ap τ e₁ ⟩
           τ₁      ∎

     II₁ : ¬¬ (s t holds)
     II₁ ν = ν (to-s-holds t II₀)


Although s is not necessarily a retraction of η, any function Ω → 𝟚
believes it is, assuming WEM. But then this can be used to get the
same conclusion without assuming WEM.


 sη-with-WEM : WEM → (f : Ω → 𝟚) (p : Ω) → f (s (η p)) ＝ f p
 sη-with-WEM w f p = I (w p)
  where
   I : is-decidable (¬ (p holds)) → f (s (η p)) ＝ f p
   I (inl ν) = f (s (η p)) ＝⟨ ap f (fails-gives-equal-⊥ pe fe (s (η p)) I₀) ⟩
               f ⊥         ＝⟨ (lemma-⊥ f p ν)⁻¹ ⟩
               f p         ∎
    where
     I₀ : ¬ (s (η p) holds)
     I₀ sh = zero-is-not-one
              (₀      ＝⟨ (η₀ p w ν)⁻¹ ⟩
               η p w  ＝⟨ happly (from-s-holds (η p) sh) w ⟩
               τ₁ w   ＝⟨ refl ⟩
               ₁      ∎)

   I (inr νν) = f (s (η p)) ＝⟨ ap f (holds-gives-equal-⊤ pe fe (s (η p)) I₁) ⟩
                f ⊤         ＝⟨ I₂ ⁻¹ ⟩
                f p         ∎
    where
     I₀ : η p ＝ τ₁
     I₀ = dfunext fe (λ w → η₁ p w νν)

     I₁ : s (η p) holds
     I₁ = to-s-holds (η p) I₀

     I₂ : f p ＝ f ⊤
     I₂ = 𝟚-is-¬¬-separated (f p) (f ⊤)
           (λ (ne : f p ≠ f ⊤) → νν (λ (ph : p holds) → ne (lemma-⊤ f p ph)))

 sη : (f : Ω → 𝟚) (p : Ω) → f (s (η p)) ＝ f p
 sη f p = 𝟚-is-¬¬-separated (f (s (η p))) (f p) I
  where
   I : ¬¬ (f (s (η p)) ＝ f p)
   I ne = ne (f (s (η p)) ＝⟨ constancy-lemma f I₁ (s (η p)) ⟩
              f ⊥         ＝⟨ (constancy-lemma f I₁ p)⁻¹ ⟩
              f p         ∎)
    where
     I₀ : ¬ WEM
     I₀ w = ne (sη-with-WEM w f p)

     I₁ : f ⊥ ＝ f ⊤
     I₁ = 𝟚-is-¬¬-separated (f ⊥) (f ⊤) (λ ne' → I₀ (WEM-lemma f ne'))

 ρ-is-equiv : (Y : 𝓦 ̇ )
            → is-totally-separated Y
            → is-equiv (ρ Y)
 ρ-is-equiv Y ts = qinvs-are-equivs (ρ Y) (ρ⁻¹ , I , II)
  where
   ρ⁻¹ : (Ω → Y) → (T → Y)
   ρ⁻¹ f = f ∘ s

   I : (g : T → Y) → ρ⁻¹ (ρ Y g) ＝ g
   I g = dfunext fe (λ t → ap g (ηs t))

   II : (f : Ω → Y) → ρ Y (ρ⁻¹ f) ＝ f
   II f = dfunext fe (λ p → ts (λ h → sη (λ q → h (f q)) p))

 reflection : (Y : 𝓦 ̇ )
            → is-totally-separated Y
            → (T → Y) ≃ (Ω → Y)
 reflection Y ts = ρ Y , ρ-is-equiv Y ts

 module _ (pt : propositional-truncations-exist) where

  open import UF.ImageAndSurjection pt
  open PropositionalTruncation pt

  resizing-gives-ηsurjection : is-surjection η
  resizing-gives-ηsurjection t = ∣ s t , ηs t ∣


This is the end of the above module assuming resizing, and we now
record everything we know about the universal property of T without
assuming resizing.

We first show that the universal property holds when 𝟚 is the target type.


extension₂'-along-η : (f : Ω → 𝟚) → is-decidable (f ⊥ ＝ f ⊤) → T → 𝟚
extension₂'-along-η f (inl _)  t = f ⊥
extension₂'-along-η f (inr ne) t = 𝟚-cases (f ⊥) (f ⊤) (t (WEM-lemma f ne))

extension₂-along-η : (Ω → 𝟚) → (T → 𝟚)
extension₂-along-η f = extension₂'-along-η f (𝟚-is-discrete (f ⊥) (f ⊤))

extension₂'-property : (f : Ω → 𝟚) (d : is-decidable (f ⊥ ＝ f ⊤)) (p : Ω)
                     → extension₂'-along-η f d (η p) ＝ f p
extension₂'-property f (inl e)  p = (constancy-lemma f e p)⁻¹
extension₂'-property f (inr ne) p = I (WEM-lemma f ne p)
 where
  I : (d : is-decidable (¬ (p holds))) → 𝟚-cases (f ⊥) (f ⊤) (δ d) ＝ f p
  I (inl ν)  = (lemma-⊥ f p ν)⁻¹
  I (inr νν) = (𝟚-is-¬¬-separated (f p) (f ⊤)
                 (λ (ne : f p ≠ f ⊤) → νν (λ (ph : p holds) → ne (lemma-⊤ f p ph))))⁻¹

extension₂-property : (f : Ω → 𝟚) (p : Ω) → extension₂-along-η f (η p) ＝ f p
extension₂-property f p = extension₂'-property f (𝟚-is-discrete (f ⊥) (f ⊤)) p

ρ₂ : (T → 𝟚) → Ω → 𝟚
ρ₂ = ρ 𝟚

restriction-of-extension₂ : (f : Ω → 𝟚) → ρ₂ (extension₂-along-η f) ＝ f
restriction-of-extension₂ f = dfunext fe (λ p → extension₂-property f p)


The points τ₀ and τ₁ are ¬¬-dense in T, which gives left-cancellability of ρ₂.


τ₀₁-density : (t : T) → ¬¬ ((t ＝ τ₀) + (t ＝ τ₁))
τ₀₁-density t ν = II (λ d → ν (I d))
 where
  I : is-decidable WEM → (t ＝ τ₀) + (t ＝ τ₁)
  I (inl w) = 𝟚-equality-cases
               (λ e → inl (t       ＝⟨ τ-lemma t w ⟩
                           τ (t w) ＝⟨ ap τ e ⟩
                           τ₀      ∎))
               (λ e → inr (t       ＝⟨ τ-lemma t w ⟩
                           τ (t w) ＝⟨ ap τ e ⟩
                           τ₁      ∎))
  I (inr nw) = inl (dfunext fe (λ w → 𝟘-elim (nw w)))

  II : ¬¬ (is-decidable WEM)
  II = double-negation-of-decision

ρ₂-lc : (g g' : T → 𝟚) → ρ₂ g ＝ ρ₂ g' → g ＝ g'
ρ₂-lc g g' e = dfunext fe (λ t → 𝟚-is-¬¬-separated (g t) (g' t) (III t))
 where
  I : g τ₀ ＝ g' τ₀
  I = g τ₀     ＝⟨ ap g (η⊥ ⁻¹) ⟩
      g (η ⊥)  ＝⟨ happly e ⊥ ⟩
      g' (η ⊥) ＝⟨ ap g' η⊥ ⟩
      g' τ₀    ∎

  II : g τ₁ ＝ g' τ₁
  II = g τ₁     ＝⟨ ap g (η⊤ ⁻¹) ⟩
       g (η ⊤)  ＝⟨ happly e ⊤ ⟩
       g' (η ⊤) ＝⟨ ap g' η⊤ ⟩
       g' τ₁    ∎

  III : (t : T) → ¬¬ (g t ＝ g' t)
  III t ne = τ₀₁-density t III₀
   where
    III₀ : ¬ ((t ＝ τ₀) + (t ＝ τ₁))
    III₀ (inl e₀) = ne (g  t  ＝⟨ ap g e₀ ⟩
                        g  τ₀ ＝⟨ I ⟩
                        g' τ₀ ＝⟨ ap g' (e₀ ⁻¹) ⟩
                        g' t  ∎)

    III₀ (inr e₁) = ne (g  t  ＝⟨ ap g e₁ ⟩
                        g  τ₁ ＝⟨ II ⟩
                        g' τ₁ ＝⟨ ap g' (e₁ ⁻¹) ⟩
                        g' t  ∎)

extension₂-of-restriction : (g : T → 𝟚) → extension₂-along-η (ρ₂ g) ＝ g
extension₂-of-restriction g = ρ₂-lc (extension₂-along-η (ρ₂ g)) g
                               (restriction-of-extension₂ (ρ₂ g))

ρ₂-is-equiv : is-equiv ρ₂
ρ₂-is-equiv = qinvs-are-equivs ρ₂
               (extension₂-along-η ,
                extension₂-of-restriction ,
                restriction-of-extension₂)


TODO. Actually, we can replace 𝟚 by any discrete type to get the same
conclusion.

We now prove the universal property when the target type is a power of 𝟚,
pointwise. More generally, if ρ Y is an equivalence then so is ρ (J → Y)
for any J.


module _
        {𝓥 𝓙 : Universe}
        (Y : 𝓥 ̇ )
        {J : 𝓙 ̇ }
        (ρY-is-equiv : is-equiv (ρ Y))
       where

 private
  𝕣 : (T → Y) ≃ (Ω → Y)
  𝕣 = (ρ Y , ρY-is-equiv)

 extension-power : (Ω → (J → Y)) → (T → (J → Y))
 extension-power f t j = ⌜ 𝕣 ⌝⁻¹ (λ p → f p j) t

 restriction-of-extension-power : (f : Ω → (J → Y))
                                → ρ (J → Y) (extension-power f) ＝ f
 restriction-of-extension-power f = dfunext fe (λ p →
                                    dfunext fe (λ j → happly
                                                       (inverses-are-sections' 𝕣
                                                         (λ (q : Ω) → f q j))
                                                       p))

 extension-of-restriction-power : (g : T → (J → Y))
                                → extension-power (ρ (J → Y) g) ＝ g
 extension-of-restriction-power g =
  dfunext fe (λ t → dfunext fe (λ j →
   happly (inverses-are-retractions' 𝕣 (λ t' → g t' j)) t))

 ρ-of-power-is-equiv : is-equiv (ρ (J → Y))
 ρ-of-power-is-equiv =
  qinvs-are-equivs (ρ (J → Y))
   (extension-power ,
    extension-of-restriction-power ,
    restriction-of-extension-power)

ρ-of-power-of-𝟚-is-equiv : {𝓘 : Universe} {J : 𝓘 ̇ } → is-equiv (ρ (J → 𝟚))
ρ-of-power-of-𝟚-is-equiv {𝓘} {J} = ρ-of-power-is-equiv 𝟚 ρ₂-is-equiv


Retracts of targets that satisfy the universal property also satisfy
the universal property of totally separated reflection.


ρ-of-retract-is-equiv : {Y : 𝓦 ̇ } {Z : 𝓣 ̇ }
                      → retract Y of Z
                      → is-equiv (ρ Z)
                      → is-equiv (ρ Y)
ρ-of-retract-is-equiv {𝓦} {𝓣} {Y} {Z} (r , s , rs) ez =
 qinvs-are-equivs (ρ Y) (ρY⁻¹ , III , IV)
 where
  ρZ⁻¹ : (Ω → Z) → (T → Z)
  ρZ⁻¹ = inverse (ρ Z) ez

  I : (φ : Ω → Z) → ρ Z (ρZ⁻¹ φ) ＝ φ
  I = inverses-are-sections (ρ Z) ez

  II : (ψ : T → Z) → ρZ⁻¹ (ρ Z ψ) ＝ ψ
  II = inverses-are-retractions (ρ Z) ez

  ρY⁻¹ : (Ω → Y) → (T → Y)
  ρY⁻¹ f = r ∘ ρZ⁻¹ (s ∘ f)

  III : (g : T → Y) → ρY⁻¹ (ρ Y g) ＝ g
  III g = ρY⁻¹ (ρ Y g)  ＝⟨ ap (λ - → r ∘ -) (II (s ∘ g)) ⟩
          r ∘ (s ∘ g)   ＝⟨ dfunext fe (λ t → rs (g t)) ⟩
          g             ∎

  IV : (f : Ω → Y) → ρ Y (ρY⁻¹ f) ＝ f
  IV f = ρ Y (ρY⁻¹ f) ＝⟨ ap (λ - → r ∘ -) (I (s ∘ f)) ⟩
         r ∘ (s ∘ f)  ＝⟨ dfunext fe (λ p → rs (f p)) ⟩
         f            ∎


The universal property for retracts of powers of 𝟚.


ρ-of-retract-of-power-of-𝟚-is-equiv
 : {𝓘 : Universe} {Y : 𝓦 ̇ } {J : 𝓘 ̇ }
 → retract Y of (J → 𝟚)
 → is-equiv (ρ Y)
ρ-of-retract-of-power-of-𝟚-is-equiv ret =
 ρ-of-retract-is-equiv ret ρ-of-power-of-𝟚-is-equiv

reflection-for-retract-of-power-of-𝟚
 : {𝓘 : Universe} {Y : 𝓦 ̇ } {J : 𝓘 ̇ }
 → retract Y of (J → 𝟚)
 → (T → Y) ≃ (Ω → Y)
reflection-for-retract-of-power-of-𝟚 r =
 ρ _ , ρ-of-retract-of-power-of-𝟚-is-equiv r


The remainder of this file has a number of observations, eventually
culminating in the fact that η : Ω → T is the universal map from Ω to
a totally separated type if and only if it is a surjection.

We first connect this to the investigation of 𝟚-injective types from
the file gist.2-injective-types.


open import gist.2-injective-types fe'

T-is-𝟚-injective : {𝓥 𝓦 : Universe} → 𝟚-injective T 𝓥 𝓦
T-is-𝟚-injective = first-dual-is-𝟚-injective

η-is-𝟚-injecting : is-𝟚-injecting η
η-is-𝟚-injecting f = extension₂-along-η f , happly (restriction-of-extension₂ f)

ρ-of-𝟚-injective-is-equiv : {Y : 𝓦 ̇ }
                          → 𝟚-injective Y 𝓦 𝓦
                          → is-equiv (ρ Y)
ρ-of-𝟚-injective-is-equiv i =
 ρ-of-retract-of-power-of-𝟚-is-equiv (𝟚-injectives-are-K-retracts i)


There is at most one extension for a totally separated target. The
following generalizes and uses ρ₂-lc.


ρ₂-of-ts-is-lc : (Y : 𝓦 ̇ )
               → is-totally-separated Y
               → (g g' : T → Y) → ρ Y g ＝ ρ Y g' → g ＝ g'
ρ₂-of-ts-is-lc Y ts g g' e =
 dfunext fe (λ t → ts (λ q → happly
                              (ρ₂-lc
                                (λ t' → q (g t'))
                                (λ t' → q (g' t'))
                                (ap (λ - → q ∘ -) e)) t))


The notion of compactness is defined in TypeTopology.CompactTypes,
where it is proved that Ω is-compact.


T-is-compact∙ : is-compact∙ T
T-is-compact∙ = micro-tychonoff fe WEM-is-prop (λ _ → 𝟚-is-compact∙)

EM-gives-Ω-discrete : EM 𝓤 → is-discrete Ω
EM-gives-Ω-discrete em p q = II (I p) (I q)
 where
  I : LEM 𝓤
  I = EM-gives-LEM em

  II : is-decidable (p holds) → is-decidable (q holds) → is-decidable (p ＝ q)
  II (inl ph) (inl qh)  = inl (p ＝⟨ holds-gives-equal-⊤ pe fe p ph ⟩
                               ⊤ ＝⟨ (holds-gives-equal-⊤ pe fe q qh)⁻¹ ⟩
                               q ∎)

  II (inl ph) (inr nq) = inr (λ e → nq (transport _holds e ph))
  II (inr np) (inl qh) = inr (λ e → np (transport _holds (e ⁻¹) qh))
  II (inr np) (inr nq) = inl (p ＝⟨ fails-gives-equal-⊥ pe fe p np ⟩
                              ⊥ ＝⟨ (fails-gives-equal-⊥ pe fe q nq)⁻¹ ⟩
                              q ∎)

EM-gives-Ω-totally-separated : EM 𝓤 → is-totally-separated Ω
EM-gives-Ω-totally-separated em = discrete-types-are-totally-separated
                                   (EM-gives-Ω-discrete em)

extension₂-along-η-under-WEM : (f : Ω → 𝟚) (w : WEM) (t : T)
                             → extension₂-along-η f t ＝ 𝟚-cases (f ⊥) (f ⊤) (t w)
extension₂-along-η-under-WEM f w t = I (𝟚-is-discrete (f ⊥) (f ⊤))
 where
  I : (d : is-decidable (f ⊥ ＝ f ⊤))
    → extension₂'-along-η f d t ＝ 𝟚-cases (f ⊥) (f ⊤) (t w)
  I (inl e)  = 𝟚-equality-cases
                (λ e' → f ⊥                       ＝⟨ I₀ e' ⟩
                        𝟚-cases (f ⊥) (f ⊤) (t w) ∎)
                (λ e' → f ⊥                       ＝⟨ e ⟩
                        f ⊤                       ＝⟨ I₁ e' ⟩
                        𝟚-cases (f ⊥) (f ⊤) (t w) ∎)
               where
                I₀ = λ e' → ap (𝟚-cases (f ⊥) (f ⊤)) (e' ⁻¹)
                I₁ = λ e' → ap (𝟚-cases (f ⊥) (f ⊤)) (e' ⁻¹)

  I (inr ne) = ap (𝟚-cases (f ⊥) (f ⊤))
                  (ap t (WEM-is-prop (WEM-lemma f ne) w))

extension₂-along-η-under-¬WEM : (f : Ω → 𝟚) (t : T)
                              → ¬ WEM
                              → extension₂-along-η f t ＝ f ⊥
extension₂-along-η-under-¬WEM f t nw = I (𝟚-is-discrete (f ⊥) (f ⊤))
 where
  I : (d : is-decidable (f ⊥ ＝ f ⊤)) → extension₂'-along-η f d t ＝ f ⊥
  I (inl e)  = refl
  I (inr ne) = 𝟘-elim (nw (WEM-lemma f ne))


We now assume propositional truncations.


module comparison (pt : propositional-truncations-exist) where

 open import UF.ImageAndSurjection pt

 ι : image η → T
 ι = restriction η

 ι-is-embedding : is-embedding ι
 ι-is-embedding = restrictions-are-embeddings η

 ι-image-is-ts : is-totally-separated (image η)
 ι-image-is-ts = subtype-is-totally-separated' ι
                  T-is-totally-separated
                  ι-is-embedding

 ηc : Ω → image η
 ηc = corestriction η

 section-of-ι-gives-η-surjection : (𝓼 : T → image η)
                                 → ι ∘ 𝓼 ＝ id
                                 → is-surjection η
 section-of-ι-gives-η-surjection 𝓼 e =
  ∘-is-surjection
   (corestrictions-are-surjections η)
   (equivs-are-surjections
     (embeddings-with-sections-are-equivs ι ι-is-embedding (𝓼 , happly e)))

 ρ-equiv-gives-η-surjection
  : ({𝓥 : Universe} (Y : 𝓥 ̇ ) → is-totally-separated Y → is-equiv (ρ Y))
  → is-surjection η
 ρ-equiv-gives-η-surjection up = section-of-ι-gives-η-surjection 𝓼 III
  where
   I : is-equiv (ρ (image η))
   I = up (image η) ι-image-is-ts

   𝓼 : T → image η
   𝓼 = inverse (ρ (image η)) I ηc

   II : ρ (image η) 𝓼 ＝ ηc
   II = inverses-are-sections (ρ (image η)) I ηc

   III : ι ∘ 𝓼 ＝ id
   III = ρ₂-of-ts-is-lc T T-is-totally-separated (ι ∘ 𝓼) id
          (ap (λ - → ι ∘ -) II)

 𝟚-injective-image-gives-η-surjection : 𝟚-injective (image η) 𝓤⁺ 𝓤⁺
                                      → is-surjection η
 𝟚-injective-image-gives-η-surjection i = section-of-ι-gives-η-surjection 𝓼 III
  where
   I : Σ 𝓼 ꞉ (T → image η) , 𝓼 ∘ η ∼ ηc
   I = i η η-is-𝟚-injecting ηc

   𝓼 : T → image η
   𝓼 = pr₁ I

   II : 𝓼 ∘ η ∼ ηc
   II = pr₂ I

   III : ι ∘ 𝓼 ＝ id
   III = ρ₂-of-ts-is-lc T T-is-totally-separated (ι ∘ 𝓼) id
          (dfunext fe (λ p → ap ι (II p)))


We now relate T to the general construction of the totally separated reflection
of any type X as the image of the evaluation map X → ((X → 𝟚) → 𝟚).


 open totally-separated-reflection fe' pt


The comparison map 𝓬.


 𝓬 : 𝕋 Ω → T
 𝓬 = ∃!-witness (totally-separated-reflection T-is-totally-separated η)

 𝓬-triangle : 𝓬 ∘ ηᵀ ＝ η
 𝓬-triangle = ∃!-is-witness
                        (totally-separated-reflection T-is-totally-separated η)

 reflection-gives-𝕋-equivalence
  : ({𝓥 : Universe} (Y : 𝓥 ̇ ) → is-totally-separated Y → is-equiv (ρ Y))
  → is-equiv 𝓬
 reflection-gives-𝕋-equivalence up
  = qinvs-are-equivs 𝓬 (𝓬⁻¹ , III , IV)
  where
   I : is-equiv (ρ (𝕋 Ω))
   I = up (𝕋 Ω) 𝕋-is-totally-separated

   𝓬⁻¹ : T → 𝕋 Ω
   𝓬⁻¹ = inverse (ρ (𝕋 Ω)) I ηᵀ

   II : ρ (𝕋 Ω) 𝓬⁻¹ ＝ ηᵀ
   II = inverses-are-sections (ρ (𝕋 Ω)) I ηᵀ

   III : 𝓬⁻¹ ∘ 𝓬 ∼ id
   III = happly VI
    where
     V : (𝓬⁻¹ ∘ 𝓬) ∘ ηᵀ ＝ ηᵀ
     V = (𝓬⁻¹ ∘ 𝓬) ∘ ηᵀ ＝⟨ ap (λ - → 𝓬⁻¹ ∘ -) 𝓬-triangle ⟩
         𝓬⁻¹ ∘ η                 ＝⟨ II ⟩
         ηᵀ                    ∎
     VI : 𝓬⁻¹ ∘ 𝓬 ＝ id
     VI = witness-uniqueness _
            (totally-separated-reflection 𝕋-is-totally-separated ηᵀ)
            (𝓬⁻¹ ∘ 𝓬) id V refl

   IV : 𝓬 ∘ 𝓬⁻¹ ∼ id
   IV = happly VII
    where
     VII : 𝓬 ∘ 𝓬⁻¹ ＝ id
     VII = ρ₂-of-ts-is-lc T T-is-totally-separated (𝓬 ∘ 𝓬⁻¹) id
            (ρ T (𝓬 ∘ 𝓬⁻¹) ＝⟨ ap (λ - → 𝓬 ∘ -) II ⟩
             𝓬 ∘ ηᵀ           ＝⟨ 𝓬-triangle ⟩
             η                         ∎)


The above development gives the equivalence

    (Ω → 𝟚) ≃ (𝟚 + WEM × 𝟚)

more or less directly.


ψ' : (f : Ω → 𝟚) → is-decidable (f ⊥ ＝ f ⊤) → 𝟚 + WEM × 𝟚
ψ' f (inl _)  = inl (f ⊥)
ψ' f (inr ne) = inr (WEM-lemma f ne , f ⊥)

ψ : (Ω → 𝟚) → 𝟚 + WEM × 𝟚
ψ f = ψ' f (𝟚-is-discrete (f ⊥) (f ⊤))

ψ⁻¹ : 𝟚 + WEM × 𝟚 → (Ω → 𝟚)
ψ⁻¹ (inl b)       _ = b
ψ⁻¹ (inr (w , b)) p = 𝟚-cases b (complement b) (δ (w p))

ψη : ψ⁻¹ ∘ ψ ∼ id
ψη f = I (𝟚-is-discrete (f ⊥) (f ⊤))
 where
  I : (d : is-decidable (f ⊥ ＝ f ⊤)) → ψ⁻¹ (ψ' f d) ＝ f
  I (inl e)  = dfunext fe (λ p → (constancy-lemma f e p)⁻¹)
  I (inr ne) = dfunext fe II
   where
    w : WEM
    w = WEM-lemma f ne

    II : (p : Ω) → 𝟚-cases (f ⊥) (complement (f ⊥)) (δ (w p)) ＝ f p
    II p = III (w p)
     where
      III : (d : is-decidable (¬ (p holds)))
          → 𝟚-cases (f ⊥) (complement (f ⊥)) (δ d) ＝ f p
      III (inl ν)  = (lemma-⊥ f p ν)⁻¹
      III (inr νν) = complement (f ⊥) ＝⟨ (complement-of-different-booleans ne)⁻¹ ⟩
                     f ⊤              ＝⟨ IV ⁻¹ ⟩
                     f p              ∎
       where
        IV = 𝟚-is-¬¬-separated (f p) (f ⊤)
             (λ ν → νν (λ ph → ν (lemma-⊤ f p ph)))

ψε : ψ ∘ ψ⁻¹ ∼ id
ψε (inl b) = I (𝟚-is-discrete b b)
 where
  I : (d : is-decidable (b ＝ b)) → ψ' (ψ⁻¹ (inl b)) d ＝ inl b
  I (inl _)  = refl
  I (inr ne) = 𝟘-elim (ne refl)
ψε (inr (w , b)) = IV (𝟚-is-discrete (f ⊥) (f ⊤))
 where
  f : Ω → 𝟚
  f = ψ⁻¹ (inr (w , b))

  I : f ⊥ ＝ b
  I = ap (𝟚-cases b (complement b)) (η₀ ⊥ w ⊥-doesnt-hold)

  II : f ⊤ ＝ complement b
  II = ap (𝟚-cases b (complement b)) (η₁ ⊤ w (¬¬-intro ⊤-holds))

  III : f ⊥ ≠ f ⊤
  III e = complement-no-fp b
           (b            ＝⟨ I ⁻¹ ⟩
            f ⊥          ＝⟨ e ⟩
            f ⊤          ＝⟨ II ⟩
            complement b ∎)

  IV : (d : is-decidable (f ⊥ ＝ f ⊤)) → ψ' f d ＝ inr (w , b)
  IV (inl e)  = 𝟘-elim (III e)
  IV (inr ne) = ap inr (to-×-＝ (WEM-is-prop (WEM-lemma f ne) w) I)

Ψ : (Ω → 𝟚) ≃ (𝟚 + WEM × 𝟚)
Ψ = ψ , qinvs-are-equivs ψ (ψ⁻¹ , ψη , ψε)


We now show that η : Ω → T is the universal map from Ω into a totally
separated type if and only if it is a surjection.


module _ (pt : propositional-truncations-exist) where

 open import UF.ImageAndSurjection pt
 open PropositionalTruncation pt
 open totally-separated-reflection fe' pt
 open comparison pt

 universal-property : 𝓤ω
 universal-property = {𝓥 : Universe} (Y : 𝓥 ̇ )
                    → is-totally-separated Y
                    → is-equiv (ρ Y)

 universal-property-gives-η-surjection : universal-property
                                       → is-surjection η
 universal-property-gives-η-surjection = ρ-equiv-gives-η-surjection

 η-surjection-gives-universal-property : is-surjection η
                                       → universal-property
 η-surjection-gives-universal-property η-surj Y ts = ρ-is-equiv
  where
   _ : type-of (eval Y) ＝ (Y → ((Y → 𝟚) → 𝟚))
   _ = refl

   _ : eval Y ＝ (λ (y : Y) (g : Y → 𝟚) → g y)
   _ = refl

   eval-is-embedding : is-embedding (eval Y)
   eval-is-embedding = totally-separated-gives-totally-separated₂ fe ts

   ε : (Ω → Y) → (T → ((Y → 𝟚) → 𝟚))
   ε f t g = extension₂-along-η (g ∘ f) t


In the next step we show that

                η
      Ω ───────────────────→ T
      │                      │
      │                      │
    f │                      │ ε f
      │                      │
      ↓                      ↓
      Y ─────────────→ ((Y → 𝟚) → 𝟚)
              eval Y


   ε-square : (f : Ω → Y) → ε f ∘ η ∼ eval Y ∘ f
   ε-square f p = dfunext fe (λ g → extension₂-property (g ∘ f) p)


It is in the following step that the surjectivity of η is used:


   φ : (f : Ω → Y) (t : T) → fiber (eval Y) (ε f t)
   φ f t = ∥∥-rec (eval-is-embedding (ε f t)) I (η-surj t)
    where
     I : (Σ p ꞉ Ω , η p ＝ t) → fiber (eval Y) (ε f t)
     I (p , e) = f p , (eval Y (f p) ＝⟨ (ε-square f p)⁻¹ ⟩
                        ε f (η p)    ＝⟨ ap (ε f) e ⟩
                        ε f t        ∎)

   σ : (Ω → Y) → (T → Y)
   σ f t = fiber-point (φ f t)


Next we show that

                  T
                 ╱│
                ╱ │
               ╱  │
              ╱   │
             ╱    │
       σ f  ╱     │ ε f
           ╱      │
          ╱       │
         ╱        │
        ↙         ↓
       Y ─────→ ((Y → 𝟚) → 𝟚)
          eval Y


   σ-triangle : (f : Ω → Y) → eval Y ∘ σ f ∼ ε f
   σ-triangle f t = fiber-identification (φ f t)


Pasting these two diagrams we get that σ is section of ρ.


   ρσ : ρ Y ∘ σ ∼ id
   ρσ f = dfunext fe
           (λ p → embeddings-are-lc (eval Y) eval-is-embedding
                   (eval Y (σ f (η p)) ＝⟨ σ-triangle f (η p) ⟩
                    ε f (η p)          ＝⟨ ε-square f p ⟩
                    eval Y (f p)       ∎))


And that σ is a retraction of ρ follows from from this and the total
separatedness of Y.


   σρ : σ ∘ ρ Y ∼ id
   σρ g = σ (ρ Y g) ＝⟨ ρ₂-of-ts-is-lc Y ts (σ (ρ Y g)) g I ⟩
          g         ∎
          where
           I : ρ Y (σ (ρ Y g)) ＝ ρ Y g
           I = ρσ (ρ Y g)

   ρ-is-equiv : is-equiv (ρ Y)
   ρ-is-equiv = qinvs-are-equivs (ρ Y) (σ , σρ , ρσ)


So the main question reduces to whether the map η : Ω → T is a
surjection in the absense of propositional resizing, or whether its
surjectivity implies an unprovable form of resizing.

Next we record the immediate fact that η : Ω → T is a surjection if
and only if the comparison map 𝓬 : 𝕋 Ω → T is an equivalence.


 ρᵀ-lc : (g g' : 𝕋 Ω → 𝟚) → g ∘ ηᵀ ＝ g' ∘ ηᵀ → g ＝ g'
 ρᵀ-lc g g' e = dfunext fe
                 (surjection-induction ηᵀ ηᵀ-is-surjection
                   (λ t → g t ＝ g' t)
                   (λ t → 𝟚-is-set)
                   (happly e))

 𝓬-lc : left-cancellable 𝓬
 𝓬-lc {x} {y} c = 𝕋-is-totally-separated I
  where
   I : (g : 𝕋 Ω → 𝟚) → g x ＝ g y
   I g = g x     ＝⟨ (happly I₁ x)⁻¹ ⟩
         r (𝓬 x) ＝⟨ ap r c ⟩
         r (𝓬 y) ＝⟨ happly I₁ y ⟩
         g y     ∎
    where
     r : T → 𝟚
     r = inverse ρ₂ ρ₂-is-equiv (g ∘ ηᵀ)

     I₀ : r ∘ η ＝ g ∘ ηᵀ
     I₀ = inverses-are-sections ρ₂ ρ₂-is-equiv (g ∘ ηᵀ)

     I₁ : r ∘ 𝓬 ＝ g
     I₁ = ρᵀ-lc (r ∘ 𝓬) g
           ((r ∘ 𝓬) ∘ ηᵀ ＝⟨ ap (λ - → r ∘ -) 𝓬-triangle ⟩
            r ∘ η        ＝⟨ I₀ ⟩
            g ∘ ηᵀ       ∎)

 𝓬-is-embedding : is-embedding 𝓬
 𝓬-is-embedding = lc-maps-into-sets-are-embeddings 𝓬 𝓬-lc T-is-set

 𝓬-is-surjection : is-surjection η → is-surjection 𝓬
 𝓬-is-surjection σ t = ∥∥-functor f (σ t)
  where
   f : (Σ p ꞉ Ω , η p ＝ t) → (Σ z ꞉ 𝕋 Ω , 𝓬 z ＝ t)
   f (p , e) = ηᵀ p , (𝓬 (ηᵀ p) ＝⟨ happly 𝓬-triangle p ⟩
                       η p      ＝⟨ e ⟩
                       t        ∎)

 η-surjection-gives-𝓬-is-equiv : is-surjection η → is-equiv 𝓬
 η-surjection-gives-𝓬-is-equiv σ = surjective-embeddings-are-equivs 𝓬
                                    𝓬-is-embedding
                                    (𝓬-is-surjection σ)

 𝓬-is-equiv-gives-η-surjection : is-equiv 𝓬 → is-surjection η
 𝓬-is-equiv-gives-η-surjection e = transport is-surjection 𝓬-triangle
                                    (∘-is-surjection
                                      ηᵀ-is-surjection
                                      (equivs-are-surjections e))


It is worth comparing the development in this file to the earlier
development in Taboos.P2.