module Cat.Displayed.Instances.Slice {o ℓ} (B : Precategory o ℓ) where
open Cartesian-fibration
open Cartesian-lift
open Displayed
open is-cartesian
open is-weak-cartesian
open Functor
open CR B
open /-Obj
The canonical self-indexing🔗
There is a canonical way of viewing any category as displayed over itself, given fibrewise by taking slice categories. Following (Sterling and Angiuli 2022), we refer to this construction as the canonical self-indexing of and denote it Recall that the objects in the slice over are pairs consisting of an object and a map The core idea is that any morphism lets us view an object as being “structure over” an object the collection of all possible such structures, then, is the set of morphisms with domain allowed to vary.
Contrary to the maps in the slice category, the maps in the canonical self-indexing have an extra “adjustment” by a morphism of the base category. Where maps in the ordinary slice are given by commuting triangles, maps in the canonical self-indexing are given by commuting squares, of the form
where the primed objects and dotted arrows are displayed.
record
Slice-hom{x y} (f : Hom x y)
(px : /-Obj {C = B} x) (py : /-Obj {C = B} y)
: Type ℓ
where
constructor slice-hom
field
to : Hom (px .domain) (py .domain)
: f ∘ px .map ≡ py .map ∘ to
commute
open Slice-hom
private unquoteDecl eqv = declare-record-iso eqv (quote Slice-hom)
The intuitive idea for the canonical self-indexing is possibly best obtained by considering the canonical self-indexing of First, recall that an object is equivalently a family of sets, with the value of the family at each point being the fibre A function of sets then corresponds to a reindexing, which takes an of sets to a of sets (in a functorial way). A morphism in the canonical self-indexing of lying over a map is then a function between the families which commutes with the reindexing given by
module _ {x y} {f g : Hom x y} {px : /-Obj x} {py : /-Obj y}
{f' : Slice-hom f px py} {g' : Slice-hom g px py} where
: (p : f ≡ g) → (f' .to ≡ g' .to) → PathP (λ i → Slice-hom (p i) px py) f' g'
Slice-pathp .to = p' i
Slice-pathp p p' i .commute =
Slice-pathp p p' i
is-prop→pathp(λ i → Hom-set _ _ (p i ∘ px .map) (py .map ∘ (p' i)))
(f' .commute)
(g' .commute)
i
Slice-path: ∀ {x y} {f : Hom x y} {px : /-Obj x} {py : /-Obj y}
→ {f' g' : Slice-hom f px py}
→ (f' .to ≡ g' .to)
→ f' ≡ g'
= Slice-pathp refl
Slice-path
unquoteDecl H-Level-Slice-hom = declare-record-hlevel 2 H-Level-Slice-hom (quote Slice-hom)
It’s straightforward to piece together the objects of the (ordinary)
slice category and our displayed maps Slice-hom
into a category displayed over
: Displayed B (o ⊔ ℓ) ℓ
Slices .Ob[_] = /-Obj {C = B}
Slices .Hom[_] = Slice-hom
Slices .Hom[_]-set _ _ _ = hlevel 2
Slices .id' = slice-hom id id-comm-sym
Slices ._∘'_ {x = x} {y = y} {z = z} {f = f} {g = g} px py =
Slices (px .to ∘ py .to) $
slice-hom (f ∘ g) ∘ x .map ≡⟨ pullr (py .commute) ⟩
(y .map ∘ py .to) ≡⟨ extendl (px .commute) ⟩
f ∘ .map ∘ (px .to ∘ py .to) ∎
z .idr' {f = f} f' = Slice-pathp (idr f) (idr (f' .to))
Slices .idl' {f = f} f' = Slice-pathp (idl f) (idl (f' .to))
Slices .assoc' {f = f} {g = g} {h = h} f' g' h' =
Slices (assoc f g h) (assoc (f' .to) (g' .to) (h' .to)) Slice-pathp
It’s only slightly more annoying to show that a vertical map in the canonical self-indexing is a map in the ordinary slice category which, since the objects displayed over are defined to be those of the slice category gives an equivalence of categories between the fibre and the slice
: ∀ {x} → Functor (Fibre Slices x) (Slice B x)
Fibre→slice .F₀ x = x
Fibre→slice .F₁ f ./-Hom.map = f .to
Fibre→slice .F₁ f ./-Hom.commutes = sym (f .commute) ∙ eliml refl
Fibre→slice .F-id = trivial!
Fibre→slice .F-∘ f g = ext (transport-refl _)
Fibre→slice
: ∀ {x} → is-fully-faithful (Fibre→slice {x = x})
Fibre→slice-is-ff {_} {x} {y} = is-iso→is-equiv isom where
Fibre→slice-is-ff : is-iso (Fibre→slice .F₁)
isom .is-iso.inv hom =
isom (hom ./-Hom.map) (eliml refl ∙ sym (hom ./-Hom.commutes))
slice-hom .is-iso.rinv x = ext refl
isom .is-iso.linv x = Slice-pathp refl refl
isom
: ∀ {x} → is-equivalence (Fibre→slice {x})
Fibre→slice-is-equiv = is-precat-iso→is-equivalence $
Fibre→slice-is-equiv record { has-is-ff = Fibre→slice-is-ff
; has-is-iso = id-equiv
}
Cartesian maps🔗
A map over in the codomain fibration is cartesian if and only if it forms a pullback square as below:
This follows by a series of relatively straightforward computations, so we do not comment too heavily on the proof.
cartesian→pullback: ∀ {x y x' y'} {f : Hom x y} {f' : Slice-hom f x' y'}
→ is-cartesian Slices f f'
→ is-pullback B (x' .map) f (f' .to) (y' .map)
{x} {y} {x'} {y'} {f} {f'} cart = pb where
cartesian→pullback : is-pullback B (x' .map) f (f' .to) (y' .map)
pb .is-pullback.square = f' .commute
pb .is-pullback.universal p =
pb .universal _ (slice-hom _ (idr _ ∙ p)) .to
cart .is-pullback.p₁∘universal =
pb (cart .universal _ _ .commute) ∙ idr _
sym .is-pullback.p₂∘universal =
pb .to (cart .commutes _ _)
ap Slice-hom.is-pullback.unique p q =
pb .to (cart .unique (slice-hom _ (idr _ ∙ sym p)) (Slice-pathp refl q))
ap Slice-hom
pullback→cartesian: ∀ {x y x' y'} {f : Hom x y} {f' : Slice-hom f x' y'}
→ is-pullback B (x' .map) f (f' .to) (y' .map)
→ is-cartesian Slices f f'
{x} {y} {x'} {y'} {f} {f'} pb = cart where
pullback→cartesian module pb = is-pullback pb
: is-cartesian Slices f f'
cart .universal m h' .to = pb.universal (assoc _ _ _ ∙ h' .commute)
cart .universal m h' .commute = sym pb.p₁∘universal
cart .commutes m h' = Slice-pathp refl pb.p₂∘universal
cart .unique m' x = Slice-pathp refl $
cart .unique (sym (m' .commute)) (ap to x) pb
_ = weak-cartesian→cartesian
We can actually weaken the hypothesis of cartesian→pullback
so that pullback
squares also exactly characterise weakly
cartesian morphisms. While this is automatic if
has all pullbacks (since then cartesian and weakly cartesian morphisms
coincide
), it is sometimes
useful to have both characterisations if we do not want to make such an
assumption.
weak-cartesian→pullback: ∀ {x y x' y'} {f : Hom x y} {f' : Slice-hom f x' y'}
→ is-weak-cartesian Slices f f'
→ is-pullback B (x' .map) f (f' .to) (y' .map)
pullback→weak-cartesian: ∀ {x y x' y'} {f : Hom x y} {f' : Slice-hom f x' y'}
→ is-pullback B (x' .map) f (f' .to) (y' .map)
→ is-weak-cartesian Slices f f'
The computation is essentially the same.
{x} {y} {x'} {y'} {f} {f'} cart = pb where
weak-cartesian→pullback : is-pullback B (x' .map) f (f' .to) (y' .map)
pb .is-pullback.square = f' .commute
pb .is-pullback.universal p =
pb .universal (slice-hom _ p) .to
cart .is-pullback.p₁∘universal =
pb (cart .universal _ .commute) ∙ idl _
sym .is-pullback.p₂∘universal =
pb (λ _ → Slice-hom.to) (cart .commutes _)
apd .is-pullback.unique p q =
pb .to (cart .unique (slice-hom _ (idl _ ∙ sym p)) (Slice-pathp (idr _) q))
ap Slice-hom
= cartesian→weak-cartesian _ (pullback→cartesian pb) pullback→weak-cartesian pb
As a fibration🔗
If (and only if) has all pullbacks, then its self-indexing is a Cartesian fibration. This is almost by definition, and is in fact where the “Cartesian” in “Cartesian fibration” comes from (recall that another term for “pullback square” is “cartesian square”). Since the total space is equivalently the arrow category of with the projection functor corresponding under this equivalence to the codomain functor, we refer to regarded as a Cartesian fibration as the codomain fibration.
Codomain-fibration: (∀ {x y z} (f : Hom x y) (g : Hom z y) → Pullback B f g)
→ Cartesian-fibration Slices
.has-lift f y' = lift-f where
Codomain-fibration pullbacks module pb = Pullback (pullbacks f (y' .map))
: Cartesian-lift Slices f y'
lift-f .x' = cut pb.p₁
lift-f .lifting .to = pb.p₂
lift-f .lifting .commute = pb.square
lift-f .cartesian = pullback→cartesian pb.has-is-pb lift-f
Since the proof that Slices
is a
cartesian fibration is given by essentially rearranging the data of
pullbacks in
we also have the converse implication: If
is a Cartesian fibration, then
has all pullbacks.
Codomain-fibration→pullbacks: ∀ {x y z} (f : Hom x y) (g : Hom z y)
→ Cartesian-fibration Slices
→ Pullback B f g
= pb where
Codomain-fibration→pullbacks f g lifts open Pullback
open is-pullback
module the-lift = Cartesian-lift (lifts .has-lift f (cut g))
: Pullback B f g
pb .apex = the-lift.x' .domain
pb .p₁ = the-lift.x' .map
pb .p₂ = the-lift.lifting .to
pb .has-is-pb .square = the-lift.lifting .commute
pb .has-is-pb .universal {p₁' = p₁'} {p₂'} p =
pb .cartesian .universal {u' = cut id}
the-lift(slice-hom p₂' (pullr (idr _) ∙ p)) .to
p₁' .has-is-pb .p₁∘universal =
pb (the-lift.universal _ _ .commute) ∙ idr _
sym .has-is-pb .p₂∘universal = ap to (the-lift.cartesian .commutes _ _)
pb .has-is-pb .unique p q = ap to $ the-lift.cartesian .unique
pb (slice-hom _ (idr _ ∙ sym p)) (Slice-pathp refl q)
Since the fibres of the codomain fibration are given by slice categories, then the interpretation of Cartesian fibrations as “displayed categories whose fibres vary functorially” leads us to reinterpret the above results as, essentially, giving the pullback functors between slice categories.
As an opfibration🔗
The canonical self-indexing is always an opfibration, where opreindexing is given by postcomposition. If we think about slices as families, then opreindexing along extends a family over to a family over by adding in empty fibres for all elements of that do not lie in the image of
: Cocartesian-fibration Slices
Codomain-opfibration .Cocartesian-fibration.has-lift f x' = lift-f where
Codomain-opfibration
: Cocartesian-lift Slices f x'
lift-f .Cocartesian-lift.y' = cut (f ∘ x' .map)
lift-f .Cocartesian-lift.lifting = slice-hom id (sym (idr _))
lift-f .Cocartesian-lift.cocartesian .is-cocartesian.universal m h' =
lift-f (h' .to) (assoc _ _ _ ∙ h' .commute)
slice-hom .Cocartesian-lift.cocartesian .is-cocartesian.commutes m h' =
lift-f (idr _)
Slice-pathp refl .Cocartesian-lift.cocartesian .is-cocartesian.unique m' p =
lift-f (sym (idr _) ∙ ap to p) Slice-pathp refl
References
- Sterling, Jonathan, and Carlo Angiuli. 2022. “Foundations of Relative Category Theory.” https://www.jonmsterling.com/frct-003I.xml.