module Cat.Instances.Sheaf.Omega {ℓ} {C : Precategory ℓ ℓ} (J : Coverage C ℓ) where
Closed sieves and the subobject classifier🔗
open is-generic-subobject open is-pullback-along open is-pullback open Subobject open Functor open Soc C open Cat C private module ΩPSh = Subobject-classifier PSh-omega module PSh = Cat (PSh ℓ C) open Coverage J using (Membership-covers ; Sem-covers)
The category of sheaves on a small site is a topos, which means that — in addition to finite limits and exponential objects — it has a subobject classifier, a sheaf which plays the role of the “universe of propositions”. We can construct the sheaf explicitly, as the sheaf of sieves.
A sieve is if it contains every morphism it covers. This notion is evidently closed under pullback, so the closed sieves form a presheaf on
is-closed : ∀ {U} → Sieve C U → Type ℓ is-closed {U} S = ∀ {V} (h : Hom V U) → J ∋ pullback h S → h ∈ S abstract is-closed-pullback : ∀ {U V} (f : Hom V U) (S : Sieve C U) → is-closed S → is-closed (pullback f S) is-closed-pullback f S c h p = c (f ∘ h) (subst (J ∋_) (sym pullback-∘) p)
instance Extensional-closed-sieve : ∀ {ℓr U} ⦃ _ : Extensional (Sieve C U) ℓr ⦄ → Extensional (Σ[ R ∈ Sieve C U ] is-closed R) ℓr Extensional-closed-sieve ⦃ e ⦄ = injection→extensional! Σ-prop-path! e
ΩJ : Sheaf J ℓ ΩJ .fst = pre where pre : Functor (C ^op) (Sets ℓ) pre .F₀ U = el! (Σ[ R ∈ Sieve C U ] is-closed R) pre .F₁ f (R , c) = pullback f R , is-closed-pullback f R c pre .F-id = funext λ _ → Σ-prop-path! pullback-id pre .F-∘ f g = funext λ _ → Σ-prop-path! pullback-∘
It remains to show that this is a sheaf. We start by showing that it is separated. Suppose we have two closed sieves which agree everywhere on some We want to show so fix some we’ll show iff
ΩJ .snd = from-is-separated sep mk where
It appears that we don’t know much about how and behave outside of their agreement on but knowing that they’re closed will be enough to show that they agree everywhere. First, let’s codify that they actually agree on their intersection with
sep : is-separated J (ΩJ .fst) sep {U} R {S , cS} {T , cT} α = ext λ {V} h → let rem₁ : S ∩S ⟦ R ⟧ ≡ T ∩S ⟦ R ⟧ rem₁ = ext λ {V} h → Ω-ua (λ (h∈S , h∈R) → cT h (subst (J ∋_) (ap fst (α h h∈R)) (max (S .closed h∈S id))) , h∈R) (λ (h∈T , h∈R) → cS h (subst (J ∋_) (ap fst (sym (α h h∈R))) (max (T .closed h∈T id))) , h∈R)
Then, assuming w.l.o.g. that we know that the pullback is a covering sieve. And since is a subsieve of we conclude that if then is and since is closed, this implies also that
rem₂ : h ∈ S → J ∋ pullback h (S ∩S ⟦ R ⟧) rem₂ h∈S = local (pull h (inc R)) λ f hf∈R → max ( S .closed h∈S (f ∘ id) , subst (_∈ R) (ap (h ∘_) (intror refl)) hf∈R ) rem₂' : h ∈ S → J ∋ pullback h T rem₂' h∈S = incl (λ _ → fst) (subst (J ∋_) (ap (pullback h) rem₁) (rem₂ h∈S))
We omit the symmetric converse for brevity.
rem₃ : h ∈ T → J ∋ pullback h S rem₃ ht = incl (λ _ → fst) (subst (J ∋_) (ap (pullback h) (sym rem₁)) (local (pull h (inc R)) λ f rfh → max (T .closed ht (f ∘ id) , subst (_∈ R) (ap (h ∘_) (intror refl)) rfh)))
in Ω-ua (λ h∈S → cT h (rem₂' h∈S)) (λ h∈T → cS h (rem₃ h∈T))
Now we have to show that a family of compatible closed sieves over a sieve can be uniquely patched to a closed sieve on This is the sieve which is defined to contain whenever, for all in the part is the maximal sieve.
module _ {U : ⌞ C ⌟} (R : J ʻ U) (S : Patch (ΩJ .fst) ⟦ R ⟧) where S' : Sieve C U S' .arrows {V} g = elΩ $ ∀ {W} (f : Hom W V) (hf : f ∈ pullback g ⟦ R ⟧) → ∀ {V'} (i : Hom V' W) → i ∈ S .part (g ∘ f) hf .fst S' .closed = elim! λ α h → inc λ {W} g hf → subst (λ e → ∀ (h : e ∈ R) {V'} (i : Hom V' W) → i ∈ S .part e h .fst) (assoc _ _ _) (α (h ∘ g)) hf
module _ {V W W'} (f : Hom V U) (hf : f ∈ ⟦ R ⟧) (g : Hom W V) (hfg : f ∘ g ∈ ⟦ R ⟧) {h : Hom W' W} where lemma : h ∈ S .part (f ∘ g) hfg .fst ≡ (g ∘ h) ∈ S .part f hf .fst lemma = sym (ap (λ e → ⌞ e .fst .arrows h ⌟) (S .patch f hf g hfg)) module lemma = Equiv (path→equiv lemma)
The first thing we have to show is that this pulls back to This is, as usual, a proof of biimplication, though in this case both directions are painful — and entirely mechanical — calculations.
restrict : ∀ {V} (f : Hom V U) (hf : f ∈ R) → pullback f S' ≡ S .part f hf .fst restrict f hf = ext λ {V} h → Ω-ua (rec! λ α → let step₁ : id ∈ S .part (f ∘ h ∘ id) (⟦ R ⟧ .closed hf (h ∘ id)) .fst step₁ = subst₂ (λ e e' → id ∈ S .part e e' .fst) (pullr refl) (to-pathp⁻ refl) (α id _ id) step₂ : ((h ∘ id) ∘ id) ∈ S .part f hf .fst step₂ = lemma.to f hf (h ∘ id) (⟦ R ⟧ .closed hf (h ∘ id)) {id} step₁ in subst (λ e → ⌞ S .part f hf .fst .arrows e ⌟) (cancelr (idr _)) step₂) (λ hh → inc λ {W} g hg {V'} i → S .part ((f ∘ h) ∘ g) hg .snd i (max let s1 : i ∈ S .part (f ∘ h ∘ g) _ .fst s1 = lemma.from f hf (h ∘ g) _ (subst (_∈ S .part f hf .fst) (assoc _ _ _) (S .part f hf .fst .closed hh (g ∘ i))) q : PathP (λ i → assoc f h g i ∈ R) _ hg q = to-pathp⁻ refl in transport (λ j → ⌞ S .part (assoc f h g j) (q j) .fst .arrows (idr i (~ j)) ⌟) s1))
Finally, we can use this to show that is closed.
abstract S'-closed : is-closed S' S'-closed {V} h hb = inc λ {W} f hf {V'} i → S .part (h ∘ f) hf .snd i $ let p = pullback (f ∘ i) (pullback h S') ≡˘⟨ pullback-∘ ⟩≡˘ pullback (h ∘ f ∘ i) S' ≡⟨ restrict (h ∘ f ∘ i) (subst (_∈ R) (sym (assoc h f i)) (⟦ R ⟧ .closed hf i)) ⟩≡ S .part (h ∘ f ∘ i) _ .fst ≡⟨ ap₂ (λ e e' → S .part e e' .fst) (assoc h f i) (to-pathp⁻ refl) ⟩≡ S .part ((h ∘ f) ∘ i) _ .fst ≡˘⟨ ap fst (S .patch (h ∘ f) hf i (⟦ R ⟧ .closed hf i)) ⟩≡˘ pullback i (S .part (h ∘ f) hf .fst) ∎ in subst (J ∋_) p (pull (f ∘ i) hb)
mk : Section (ΩJ .fst) S mk .whole = S' , S'-closed mk .glues f hf = Σ-prop-path! (restrict f hf) open _=>_ ΩJ=>Ω : ΩJ .fst => Sieves {C = C} ΩJ=>Ω .η U = fst ΩJ=>Ω .is-natural x y f = refl
Closed subpresheaves🔗
To show that is the subobject classifier in we’ll start by showing that a subpresheaf of a sheaf is a sheaf precisely when the associated natural transformation into the presheaf of sieves on is valued in closed sieves.
First, suppose that the domain of is indeed a sheaf, and note that since the inclusion is a right adjoint, any subobject in is also a subobject in Since subobjects in are componentwise embeddings, the same is thus true of a subobject in
sheaf-name : {A : Sheaf J _} (A' : Subobject (Sheaves J _) A) → A .fst => ΩJ .fst sheaf-name {A} A' = nm module sheaf-name where sub' : Subobject (PSh ℓ C) (A .fst) sub' .domain = A' .domain .fst sub' .map = A' .map sub' .monic = right-adjoint→is-monic _ (free-objects→left-adjoint (Small.make-free-sheaf J)) {A' .domain} {A} (λ {C} → A' .monic {C}) emb : ∀ {i} → is-embedding (A' .map .η i) emb = is-monic→is-embedding-at ℓ C (sub' .monic) nm' : A .fst => Sieves {C = C} nm' = psh-name sub'
We can then compute, in the name of resulting in a natural transformation It remains to show that this transformation is actually valued in i.e. that each of the resulting sieves is closed. To this end, suppose we have some and that the sieve is We must show that contains
By the definition of it suffices to exhibit a fibre of over and since is a sheaf, we can construct this fibre by gluing over the sieve At each we are given some fibre of over but since is an embedding everywhere, we can extract an actual element of from this mere fibre. By the usual functoriality work, this extends to a patch over
name-is-closed : {U : ⌞ C ⌟} (x : A ʻ U) → is-closed (nm' .η U x) name-is-closed {U} x {V} h pb = let it = sat.split (A' .domain .snd) pb record { part = λ {W} f hf → □-out (emb _) hf .fst ; patch = λ f hf g hgf → ap fst $ emb _ (_ , (A' .map .η _ (A' .domain .fst ⟪ g ⟫ (□-out (emb _) hf .fst)) ≡⟨ A' .map .is-natural _ _ _ $ₚ _ ⟩≡ (A .fst ⟪ g ⟫ (A' .map .η _ (□-out (emb _) hf .fst))) ≡⟨ ap (A .fst .F₁ g) (□-out (emb _) hf .snd) ⟩≡ (A .fst ⟪ g ⟫ (A .fst ⟪ h ∘ f ⟫ x)) ≡⟨ sym (A .fst .F-∘ _ _ $ₚ _) ⟩≡ (A .fst ⟪ (h ∘ f) ∘ g ⟫ x) ∎)) (_ , □-out (emb _) hgf .snd ∙ ap₂ (A .fst .F₁) (assoc _ _ _) refl) }
We must then show that this is indeed a fibre of over But since being a sheaf, is separated, it suffices to do so locally along at which point we know that this is true by construction.
prf = sat.separate (A .snd) pb λ f hf → A .fst ⟪ f ⟫ A' .map .η V (it .whole) ≡˘⟨ A' .map .is-natural _ _ _ $ₚ _ ⟩≡˘ A' .map .η _ (A' .domain .fst ⟪ f ⟫ it .whole) ≡⟨ ap (A' .map .η _) (it .glues f hf) ⟩≡ A' .map .η _ (□-out (emb _) hf .fst) ≡⟨ □-out (emb _) hf .snd ⟩≡ A .fst ⟪ h ∘ f ⟫ x ≡⟨ A .fst .F-∘ _ _ $ₚ _ ⟩≡ A .fst ⟪ f ⟫ (A .fst ⟪ h ⟫ x) ∎ in inc (it .whole , prf) nm : A .fst => ΩJ .fst nm .η U x = record { fst = nm' .η U x ; snd = name-is-closed x } nm .is-natural x y f = funext λ a → Σ-prop-path! (happly (nm' .is-natural x y f) a)
This “if” direction turns out to be sufficient to prove that is a subobject classifier. First, the maximal sieve is obviously closed, because it contains every morphism:
Sh[]-true : Subobject (Sheaves J ℓ) ΩJ Sh[]-true .domain = Terminal.top (Sh[]-terminal J) Sh[]-true .map .η _ _ = maximal' , (λ _ _ → tt) Sh[]-true .map .is-natural x y f = trivial! Sh[]-true .monic g h x = trivial!
Now the universal property follows essentially by arguing in Since the proofs boil down to shuffling whether a given square is a pullback in or in we do not comment on them further.
Sh[]-true-is-generic : is-generic-subobject (Sheaves J ℓ) Sh[]-true Sh[]-true-is-generic .name = sheaf-name Sh[]-true-is-generic .classifies m = record { has-is-pb = pb' } where rem : is-pullback-along (PSh ℓ C) (m .map) (ΩJ=>Ω ∘nt sheaf-name m) (ΩJ=>Ω ∘nt Sh[]-true .map) rem = record { has-is-pb = subst-is-pullback refl trivial! refl trivial! (ΩPSh.classifies (sheaf-name.sub' m) .has-is-pb) } pb' : is-pullback (Sheaves J ℓ) (m .map) (sheaf-name m) (rem .top) (Sh[]-true .map) pb' .square = ext λ i h {V} x → unext (rem .square) i h x pb' .universal α = rem .universal (PSh.extendr α) pb' .p₁∘universal = rem .p₁∘universal pb' .p₂∘universal = rem .p₂∘universal pb' .unique = rem .unique Sh[]-true-is-generic .unique {m = m} {nm} α = ext λ i x {V} h → unext path i x h where pb : is-pullback (PSh ℓ C) (m .map) nm (α .top) (Sh[]-true .map) pb = right-adjoint→is-pullback (free-objects→left-adjoint (Small.make-free-sheaf J)) (α .has-is-pb) pb' : is-pullback (PSh ℓ C) (m .map) (ΩJ=>Ω ∘nt nm) (α .top) (ΩPSh.true .map) pb' .square = ext λ i h {V} x → unext (pb .square) i h x pb' .universal {p₁' = p₁'} {p₂'} α = pb .universal {p₁' = p₁'} {p₂'} (ext λ i x {V} h → unext α i x h) pb' .p₁∘universal = pb .p₁∘universal pb' .p₂∘universal = pb .p₂∘universal pb' .unique = pb .unique path = ΩPSh.unique {m = sheaf-name.sub' m} record { has-is-pb = pb' }
Finally, we can show the converse direction to the result we claimed in the start of this section: a subpresheaf with name valued in sieves is itself a sheaf. First, we show that it is separated, because it embeds into a sheaf:
name-is-closed→is-sheaf : {A : Sheaf J _} (A' : Subobject (PSh ℓ C) (A .fst)) → (∀ {U : ⌞ C ⌟} (x : A ʻ U) → is-closed (psh-name A' .η U x)) → is-sheaf J (A' .domain) name-is-closed→is-sheaf {A = A} A' cl = from-is-sheaf₁ λ {U} j p → from-is-separated₁ _ (sep j) (s j p) where emb : ∀ {i} → is-embedding (A' .map .η i) emb = is-monic→is-embedding-at ℓ C (A' .monic) sep : ∀ {U} (j : J .covers U) → is-separated₁ (A' .domain) (J .cover j) sep j {x} {y} h = ap fst $ emb _ (_ , refl) $ _ , A .snd .separate j λ f hf → A .fst ⟪ f ⟫ A' .map .η _ y ≡˘⟨ A' .map .is-natural _ _ _ $ₚ _ ⟩≡˘ A' .map .η _ ⌜ A' .domain ⟪ f ⟫ y ⌝ ≡⟨ ap! (sym (h f hf)) ⟩≡ A' .map .η _ (A' .domain ⟪ f ⟫ x) ≡⟨ A' .map .is-natural _ _ _ $ₚ _ ⟩≡ A .fst ⟪ f ⟫ A' .map .η _ x ∎
Now, given a sieve of some object and a patch along over we must put together a section in First, we lift to a patch of the sheaf, over the same sieve; and this glues to a section
module _ {U} (j : J .covers U) (p : Patch (A' .domain) (J .cover j)) where p' : Patch (A .fst) (J .cover j) p' = record { part = λ {V} f hf → A' .map .η V (p .part f hf) ; patch = λ {V} {W} f hf g hgf → sym (A' .map .is-natural _ _ _ $ₚ _) ∙ ap (A' .map .η _) (p .patch f hf g hgf) } s' : Section (A .fst) p' s' = is-sheaf.split (A .snd) p'
We must show that actually comes from a section This follows from having a closed name: by definition, this means that if we can show that is a sieve, then the fibre of over is contractible. To show this, it suffices to show that pulls back to a locally at every in some other sieve — we choose the we were already working against. Then we’re left with a simple computation:
abstract has : J ∋ pullback id (psh-name A' .η U (s' .whole)) has = local (inc j) λ {V} f hf → let α : pullback f (pullback id (psh-name A' .η U (s' .whole))) ≡ psh-name A' .η V (p' .part f hf) α = pullback f ⌜ pullback id (psh-name A' .η U (s' .whole)) ⌝ ≡⟨ ap! pullback-id ⟩≡ pullback f (psh-name A' .η U (s' .whole)) ≡⟨ sym (psh-name A' .is-natural _ _ _ $ₚ _) ⟩≡ psh-name A' .η V (A .fst ⟪ f ⟫ s' .whole) ≡⟨ ap (psh-name A' .η _) (s' .glues f hf) ⟩≡ psh-name A' .η V (p' .part f hf) ∎ in subst (J ∋_) (sym α) (max (inc (p .part f hf , sym (A .fst .F-id $ₚ _)))) it : fibre (map A' .η U) (A .fst ⟪ id ⟫ s' .whole) it = □-out (emb _) (cl (s' .whole) id has) s : Section (A' .domain) p s = record { whole = it .fst ; glues = λ f hf → ap fst (emb (A .fst ⟪ f ⟫ s' .whole) (_ , A' .map .is-natural _ _ _ $ₚ _ ∙ ap (A .fst .F₁ f) (it .snd ∙ A .fst .F-id $ₚ _)) (_ , sym (s' .glues f hf))) }