open import Cat.Displayed.Cartesian
open import Cat.Instances.Functor
open import Cat.Displayed.Base
open import Cat.Prelude

import Cat.Displayed.Reasoning as Dr
import Cat.Displayed.Solver as Ds

module
  Cat.Displayed.Instances.Pullback
    {o ℓ o′ ℓ′ o′′ ℓ′′}
    {X : Precategory o ℓ} {B : Precategory o′ ℓ′}
    (F : Functor X B)
    (E : Displayed B o′′ ℓ′′)
  where

Pullback of a displayed category🔗

If we have a category $E$ displayed over $B$, then a functor $F : X \to B$ defines a (contravariant) “change-of-base” action, resulting in a category $F^*(E)$ displayed over $X$.

The reason for this contravariance is the following: A category displayed over $X$ must have a $X$-indexed space of objects; But our category $E$ has a $B$-indexed space of objects. Fortunately, $F$ gives us a map $x \mapsto F_0(x)$ which allows us to convert our $X$-indices to $B$-indices. The same goes for indexing the displayed $\mathbf{Hom}$-sets.

Change-of-base : Displayed X o′′ ℓ′′
Change-of-base .Ob[_] x      = E.Ob[ F₀ x ]
Change-of-base .Hom[_] f x y = E.Hom[ F₁ f ] x y
Change-of-base .Hom[_]-set f = E.Hom[ F₁ f ]-set
Change-of-base .id′          = hom[ sym F-id ] E.id′
Change-of-base ._∘′_ f′ g′   = hom[ sym (F-∘ _ _) ] (f′ E.∘′ g′)

Proving that the pullback $F^*(E)$ is indeed a displayed category is a bit trickier than it might seem — we must adjust the identity and composites in $E$ by the paths witnessing functoriality of $F$, and this really throws a spanner in the works — but the handy displayed category reasoning combinators are here to help.

Change-of-base .idr′ {f = f} f′ = to-pathp $
  hom[] (hom[ F-∘ _ _ ]⁻ (f′ E.∘′ hom[ F-id ]⁻ _)) ≡⟨ hom[]⟩⟨ smashr _ _ ⟩≡
  hom[] (hom[] (f′ E.∘′ E.id′))                    ≡⟨ Ds.disp! E ⟩≡
  f′                                               ∎

Change-of-base .idl′ f′ = to-pathp $
  hom[] (hom[ F-∘ _ _ ]⁻ (hom[ F-id ]⁻ _ E.∘′ f′)) ≡⟨ hom[]⟩⟨ smashl _ _ ⟩≡
  hom[] (hom[] (E.id′ E.∘′ f′))                    ≡⟨ Ds.disp! E ⟩≡
  f′                                               ∎

Change-of-base .assoc′ f′ g′ h′ = to-pathp $
  hom[ ap F₁ _ ] (hom[ F-∘ _ _ ]⁻ (f′ E.∘′ hom[ F-∘ _ _ ]⁻ (g′ E.∘′ h′)))   ≡⟨ hom[]⟩⟨ smashr _ _ ⟩≡
  hom[] (hom[] (f′ E.∘′ g′ E.∘′ h′))                                        ≡⟨ Ds.disp! E ⟩≡
  hom[ sym (F-∘ _ _) ] (hom[ sym (F-∘ _ _) ] (f′ E.∘′ g′) E.∘′ h′)          ∎

As a fibration🔗

If $\mathcal{E}$ is a cartesian fibration, then the base change of $\mathcal{E}$ along $F$ is also cartesian. To prove this, observe that the object and hom spaces of Change-of-base contain the same data as $\mathcal{E}$, just restricted to objects and morphisms in the image of $F$. This means that we still have cartesian lifts of all morphisms in $\mathcal{X}$: we just have to perform the lifting of $F f$.

Change-of-base-fibration : Cartesian-fibration E → Cartesian-fibration Change-of-base
Change-of-base-fibration fib .Cartesian-fibration.has-lift f FY′ = cart-lift
  where
    open Cartesian-lift (Cartesian-fibration.has-lift fib (F₁ f) FY′)

    cart-lift : Cartesian-lift Change-of-base f FY′
    cart-lift .Cartesian-lift.x′ = x′
    cart-lift .Cartesian-lift.lifting = lifting
    cart-lift .Cartesian-lift.cartesian .is-cartesian.universal m h′ =
      universal (F .Functor.F₁ m) (hom[ F-∘ f m ] h′)
    cart-lift .Cartesian-lift.cartesian .is-cartesian.commutes m h′ =
      hom[ F-∘ f m ]⁻ (lifting E.∘′ universal (F₁ m) (hom[ F-∘ f m ] h′)) ≡⟨ ap hom[ F-∘ f m ]⁻ (commutes _ _) ⟩≡
      hom[ F-∘ f m ]⁻ (hom[ F-∘ f m ] h′)                                 ≡⟨ Ds.disp! E ⟩≡
      h′                                                                  ∎
    cart-lift .Cartesian-lift.cartesian .is-cartesian.unique {m = m} {h′ = h′} m′ p =
      unique m′ $
        lifting E.∘′ m′                                    ≡⟨ Ds.disp! E ⟩≡
        hom[ F-∘ f m ] (hom[ F-∘ f m ]⁻ (lifting E.∘′ m′)) ≡⟨ ap hom[ F-∘ f m ] p ⟩≡
        hom[ F-∘ f m ] h′ ∎