{-# OPTIONS --lossy-unification #-}
open import Cat.Monoidal.Instances.Cartesian
open import Cat.Displayed.Univalence.Thin
open import Cat.Instances.Sets.Complete
open import Cat.Displayed.Functor
open import Cat.Bi.Diagram.Monad
open import Cat.Displayed.Base
open import Cat.Displayed.Path
open import Cat.Monoidal.Base
open import Cat.Bi.Base
open import Cat.Prelude

import Algebra.Monoid.Category as Mon
import Algebra.Monoid as Mon

import Cat.Diagram.Monad as Mo
import Cat.Reasoning

module Cat.Monoidal.Diagram.Monoid where

module _ {o ℓ} {C : Precategory o ℓ} (M : Monoidal-category C) where
  private module C where
    open Cat.Reasoning C public
    open Monoidal-category M public

Monoids in a monoidal category🔗

Let $(\mathcal{C}, \otimes, 1)$ be a monoidal category you want to study. It can be, for instance, one of the endomorphism categories in a bicategory that you like. A monoid object in $\mathcal{C}$, generally just called a “monoid in $\mathcal{C}$”, is really a collection of diagrams in $\mathcal{C}$ centered around an object $M$, the monoid itself.

In addition to the object, we also require a “unit” map $\eta : 1 \to M$ and “multiplication” map $\mu : M \otimes M \to M$. Moreover, these maps should be compatible with the unitors and associator of the underlying monoidal category:

  record Monoid-on (M : C.Ob) : Type ℓ where
    no-eta-equality
    field
      η : C.Hom C.Unit M
      μ : C.Hom (M C.⊗ M) M

      μ-unitl : μ C.∘ (η C.⊗₁ C.id) ≡ C.λ←
      μ-unitr : μ C.∘ (C.id C.⊗₁ η) ≡ C.ρ←
      μ-assoc : μ C.∘ (C.id C.⊗₁ μ) ≡ μ C.∘ (μ C.⊗₁ C.id) C.∘ C.α← _ _ _

If we think of $\mathcal{C}$ as a bicategory with a single object $*$ — that is, we deloop it —, then a monoid in $\mathcal{C}$ is given by precisely the same data as a monad in $\mathbf{B}\mathcal{C}$, on the object $*$.

  private
    BC = Deloop M
    module BC = Prebicategory BC
  open Monoid-on

  Monoid-pathp
    : ∀ {P : I → C.Ob} {x : Monoid-on (P i0)} {y : Monoid-on (P i1)}
    → PathP (λ i → C.Hom C.Unit (P i)) (x .η) (y .η)
    → PathP (λ i → C.Hom (P i C.⊗ P i) (P i)) (x .μ) (y .μ)
    → PathP (λ i → Monoid-on (P i)) x y
  Monoid-pathp {x = x} {y} p q i .η = p i
  Monoid-pathp {x = x} {y} p q i .μ = q i
  Monoid-pathp {P = P} {x} {y} p q i .μ-unitl =
    is-prop→pathp
      (λ i → C.Hom-set _ (P i) (q i C.∘ (p i C.⊗₁ C.id)) C.λ←)
      (x .μ-unitl)
      (y .μ-unitl)
      i
  Monoid-pathp {P = P} {x} {y} p q i .μ-unitr =
    is-prop→pathp
      (λ i → C.Hom-set _ (P i) (q i C.∘ (C.id C.⊗₁ p i)) C.ρ←)
      (x .μ-unitr)
      (y .μ-unitr)
      i
  Monoid-pathp {P = P} {x} {y} p q i .μ-assoc =
    is-prop→pathp
      (λ i → C.Hom-set _ (P i)
        (q i C.∘ (C.id C.⊗₁ q i))
        (q i C.∘ (q i C.⊗₁ C.id) C.∘ C.α← _ _ _))
      (x .μ-assoc)
      (y .μ-assoc)
      i

  monad→monoid : (M : Monad BC tt) → Monoid-on (M .Monad.M)
  monad→monoid M = go where
    module M = Monad M
    go : Monoid-on M.M
    go .η = M.η
    go .μ = M.μ
    go .μ-unitl = M.μ-unitl
    go .μ-unitr = M.μ-unitr
    go .μ-assoc = M.μ-assoc

  monoid→monad : ∀ {M} → Monoid-on M → Monad BC tt
  monoid→monad M = go where
    module M = Monoid-on M
    go : Monad BC tt
    go .Monad.M = _
    go .Monad.μ = M.μ
    go .Monad.η = M.η
    go .Monad.μ-assoc = M.μ-assoc
    go .Monad.μ-unitr = M.μ-unitr
    go .Monad.μ-unitl = M.μ-unitl

Put another way, a monad is just a monoid in the category of endofunctors endo-1-cells, what’s the problem?

The category Mon(C)🔗

The monoid objects in $\mathcal{C}$ can be made into a category, much like the monoids in the category of sets. In fact, we shall see later that when $\mathbf{Sets}$ is equipped with its Cartesian monoidal structure, $\mathrm{Mon}(\mathbf{Sets}) \cong \mathrm{Mon}$. Rather than defining $\mathrm{Mon}(\mathcal{C})$ directly as a category, we instead define it as a category $\mathrm{Mon}(\mathcal{C}) \mathrel{\htmlClass{liesover}{\hspace{1.366em}}} \mathcal{C}$ displayed over $\mathcal{C}$, which fits naturally with the way we have defined Monoid-object-on.

module _ {o ℓ} {C : Precategory o ℓ} (M : Monoidal-category C) where
  private module C where
    open Cat.Reasoning C public
    open Monoidal-category M public

Therefore, rather than defining a type of monoid homomorphisms, we define a predicate on maps $f : m \to n$ expressing the condition of being a monoid homomorphism. This is the familiar condition from algebra, but expressed in a point-free way:

  record
    is-monoid-hom {m n} (f : C.Hom m n)
     (mo : Monoid-on M m) (no : Monoid-on M n) : Type ℓ where

    private
      module m = Monoid-on mo
      module n = Monoid-on no

    field
      pres-η : f C.∘ m.η ≡ n.η
      pres-μ : f C.∘ m.μ ≡ n.μ C.∘ (f C.⊗₁ f)

Since being a monoid homomorphism is a pair of propositions, the overall condition is a proposition as well. This means that we will not need to concern ourselves with proving displayed identity and associativity laws, a great simplification.

  private unquoteDecl eqv = declare-record-iso eqv (quote is-monoid-hom)

  is-monoid-hom-is-prop : ∀ {m n} {f : C.Hom m n} {mo no} → is-prop (is-monoid-hom f mo no)
  is-monoid-hom-is-prop = is-hlevel≃  (Iso→Equiv eqv) hlevel!

  open Displayed
  open Functor
  open is-monoid-hom

  Mon[_] : Displayed C ℓ ℓ
  Mon[_] .Ob[_]  = Monoid-on M
  Mon[_] .Hom[_] = is-monoid-hom
  Mon[_] .Hom[_]-set f x y = is-prop→is-set is-monoid-hom-is-prop

The most complicated step of putting together the displayed category of monoid objects is proving that monoid homomorphisms are closed under composition. However, even in the point-free setting of an arbitrary category $\mathcal{C}$, the reasoning isn’t that painful:

  Mon[ .id′ ] .pres-η = C.idl _
  Mon[ .id′ ] .pres-μ = C.idl _ ∙ C.intror (C.-⊗- .F-id)

  Mon[_] ._∘′_ fh gh .pres-η = C.pullr (gh .pres-η) ∙ fh .pres-η
  Mon[_] ._∘′_ {x = x} {y} {z} {f} {g} fh gh .pres-μ =
    (f C.∘ g) C.∘ x .Monoid-on.μ                ≡⟨ C.pullr (gh .pres-μ) ⟩≡
    f C.∘ y .Monoid-on.μ C.∘ (g C.⊗₁ g)         ≡⟨ C.extendl (fh .pres-μ) ⟩≡
    Monoid-on.μ z C.∘ (f C.⊗₁ f) C.∘ (g C.⊗₁ g) ≡˘⟨ C.refl⟩∘⟨ C.-⊗- .F-∘ _ _ ⟩≡˘
    Monoid-on.μ z C.∘ (f C.∘ g C.⊗₁ f C.∘ g)    ∎

  Mon[_] .idr′ f = is-prop→pathp (λ i → is-monoid-hom-is-prop) _ _
  Mon[_] .idl′ f = is-prop→pathp (λ i → is-monoid-hom-is-prop) _ _
  Mon[_] .assoc′ f g h = is-prop→pathp (λ i → is-monoid-hom-is-prop) _ _

private
  Setsₓ : ∀ {ℓ} → Monoidal-category (Sets ℓ)
  Setsₓ = Cartesian-monoidal Sets-products Sets-terminal
  Mon : ∀ {ℓ} → Displayed (Sets ℓ) _ _
  Mon = Thin-structure-over (Mon.Monoid-structure _)

Constructing this displayed category for the Cartesian monoidal structure on the category of sets, we find that it is but a few renamings away from the ordinary category of monoids-on-sets. The only thing out of the ordinary about the proof below is that we can establish the displayed categories themselves are identical, so it is a trivial step to show they induce identical¹ total categories.

Mon[Sets]≡Mon : ∀ {ℓ} → Mon[ Setsₓ ] ≡ Mon {ℓ}
Mon[Sets]≡Mon {ℓ} = Displayed-path F (λ a → is-iso→is-equiv (fiso a)) ff where
  open Displayed-functor
  open Monoid-on
  open is-monoid-hom

  open Mon.Monoid-hom
  open Mon.Monoid-on

The construction proceeds in three steps: First, put together a functor (displayed over the identity) $\mathrm{Mon}(\mathcal{C}) \to \mathbf{Mon}$; Then, prove that its action on objects (“step 2”) and action on morphisms (“step 3”) are independently equivalences of types. The characterisation of paths of displayed categories will take care of turning this data into an identification.

  F : Displayed-functor Mon[ Setsₓ ] Mon Id
  F .F₀′ o .identity = o .η (lift tt)
  F .F₀′ o ._⋆_ x y = o .μ (x , y)
  F .F₀′ o .has-is-monoid .Mon.has-is-semigroup =
    record { has-is-magma = record { has-is-set = hlevel! }
           ; associative  = o .μ-assoc $ₚ _
           }
  F .F₀′ o .has-is-monoid .Mon.idl = o .μ-unitl $ₚ _
  F .F₀′ o .has-is-monoid .Mon.idr = o .μ-unitr $ₚ _
  F .F₁′ wit .pres-id = wit .pres-η $ₚ _
  F .F₁′ wit .pres-⋆ x y = wit .pres-μ $ₚ _
  F .F-id′ = prop!
  F .F-∘′ = prop!

  open is-iso

  fiso : ∀ a → is-iso (F .F₀′ {a})
  fiso T .inv m .η _ = m .identity
  fiso T .inv m .μ (a , b) = m ._⋆_ a b
  fiso T .inv m .μ-unitl = funext λ _ → m .idl
  fiso T .inv m .μ-unitr = funext λ _ → m .idr
  fiso T .inv m .μ-assoc = funext λ _ → m .associative
  fiso T .rinv x = Mon.Monoid-structure _ .id-hom-unique
    (record { pres-id = refl ; pres-⋆ = λ _ _ → refl })
    (record { pres-id = refl ; pres-⋆ = λ _ _ → refl })
  fiso T .linv m = Monoid-pathp Setsₓ refl refl

  ff : ∀ {a b : Set _} {f : ∣ a ∣ → ∣ b ∣} {a′ b′}
     → is-equiv (F₁′ F {a} {b} {f} {a′} {b′})
  ff {a} {b} {f} {a′} {b′} =
    prop-ext (is-monoid-hom-is-prop Setsₓ) (hlevel )
             (λ z → F₁′ F z) invs .snd
    where
      invs : Mon.Monoid-hom (F .F₀′ a′) (F .F₀′ b′) f
           → is-monoid-hom Setsₓ f a′ b′
      invs m .pres-η = funext λ _ → m .pres-id
      invs m .pres-μ = funext λ _ → m .pres-⋆ _ _