open import Cat.Instances.Comma
open import Cat.Functor.Base
open import Cat.Univalent
open import Cat.Prelude

import Cat.Functor.Reasoning as Func
import Cat.Reasoning

module
  Cat.Instances.Comma.Univalent
  {xo xℓ yo yℓ zo zℓ}
  {X : Precategory xo xℓ} {Y : Precategory yo yℓ} {Z : Precategory zo zℓ}
  (F : Functor Y X) (G : Functor Z X)
  (xuniv : is-category X)
  (yuniv : is-category Y)
  (zuniv : is-category Z)
  where

private
  module X = Cat.Reasoning X
  module Y = Cat.Reasoning Y
  module Z = Cat.Reasoning Z
  module F = Func F
  module G = Func G
  module F↓G = Cat.Reasoning (F ↓ G)

open ↓Obj
open ↓Hom

Comma categories preserve univalence🔗

Theorem. Let $\mathcal{Y} \xrightarrow{F} \mathcal{X} \xleftarrow{G} \mathcal{Z}$ be a cospan of functors between three univalent categories. Then the comma category $F \downarrow G$ is also univalent.

It suffices to establish that, given an isomorphism $f : o \cong o'$ in $F \downarrow G$, one gets an identification $\^f : o \equiv o'$, over which $f$ is the identity map. Since $\mathcal{Y}$ and $\mathcal{Z}$ are both univalent categories, we get (from the components $f_\alpha$, $f_\beta$ of $f$) identifications $\^f_\alpha : o_x \equiv o'_x$ and $\^f_\beta : o_y \equiv o'_y$.

Comma-is-category : is-category (F ↓ G)
Comma-is-category = record { to-path = objs ; to-path-over = maps } where
  module _ {ob ob′} (isom : F↓G.Isomorphism ob ob′) where
    module isom = F↓G._≅_ isom

    x-is-x : ob .x Y.≅ ob′ .x
    y-is-y : ob .y Z.≅ ob′ .y

    x-is-x = Y.make-iso (isom.to .α) (isom.from .α) (ap α isom.invl) (ap α isom.invr)
    y-is-y = Z.make-iso (isom.to .β) (isom.from .β) (ap β isom.invl) (ap β isom.invr)

Observe that, over $\^f_\alpha$ and $\^f_\beta$, the map components $o_m$ and $o'_m$ are equal (we say “equal” rather than “identified” because Hom-sets are sets). This follows by standard abstract nonsense: in particular, functors between univalent categories respect isomorphisms in a strong sense (F-map-path).

This lets us reduce statements about $F$ and $G$’s object-part action on paths to statements about their morphism parts $F_1$, $G_1$ on the components of the isomorphisms $f_\alpha$ and $f_\beta$. But then we have

$$\begin{split} G_1(f_\beta) o_m F_1(f_\alpha^{-1}) &= o'_m F_1(f_\alpha) F_1(f_\alpha^{-1}) \\ &= o'_m F_1(f_\alpha f_\alpha^{-1}) \\ &= o'_m\text{,} \end{split}$$

so over these isomorphisms the map parts become equal, thus establishing an identification $o \equiv o'$.

    objs : ob ≡ ob′
    objs i .x = yuniv .to-path x-is-x i
    objs i .y = zuniv .to-path y-is-y i
    objs i .map = lemma′ i where
      lemma′ : PathP (λ i → X.Hom (F.₀ (objs i .x)) (G.₀ (objs i .y)))
                (ob .map) (ob′ .map)
      lemma′ = transport
        (λ i → PathP (λ j → X.Hom (F-map-path F x-is-x yuniv xuniv (~ i) j)
                                  (F-map-path G y-is-y zuniv xuniv (~ i) j))
                    (ob .map) (ob′ .map)) $
        Univalent.Hom-pathp-iso xuniv $
          X.pulll   (sym (isom.to .sq)) ∙
          X.cancelr (F.annihilate (ap α isom.invl))

It still remains to show that, over this identification, the isomorphism $f$ is equal to the identity function. But this is simply a matter of pushing the identifications down to reach the “leaf” morphisms.

    maps : PathP (λ i → ob F↓G.≅ objs i) _ isom
    maps = F↓G.≅-pathp _ _
      (↓Hom-pathp _ _ (Univalent.Hom-pathp-reflr-iso yuniv (Y.idr _))
                      (Univalent.Hom-pathp-reflr-iso zuniv (Z.idr _)))