/- Copyright (c) 2020 Bhavik Mehta. All rights reserved. Released under Apache 2.0 license as described in the file LICENSE. Authors: Bhavik Mehta, Yaël Dillies -/ import data.set.lattice import data.set_like.basic import order.galois_connection import order.hom.basic import tactic.monotonicity /-! # Closure operators between preorders We define (bundled) closure operators on a preorder as monotone (increasing), extensive (inflationary) and idempotent functions. We define closed elements for the operator as elements which are fixed by it. Lower adjoints to a function between preorders `u : β → α` allow to generalise closure operators to situations where the closure operator we are dealing with naturally decomposes as `u ∘ l` where `l` is a worthy function to have on its own. Typical examples include `l : set G → subgroup G := subgroup.closure`, `u : subgroup G → set G := coe`, where `G` is a group. This shows there is a close connection between closure operators, lower adjoints and Galois connections/insertions: every Galois connection induces a lower adjoint which itself induces a closure operator by composition (see `galois_connection.lower_adjoint` and `lower_adjoint.closure_operator`), and every closure operator on a partial order induces a Galois insertion from the set of closed elements to the underlying type (see `closure_operator.gi`). ## Main definitions * `closure_operator`: A closure operator is a monotone function `f : α → α` such that `∀ x, x ≤ f x` and `∀ x, f (f x) = f x`. * `lower_adjoint`: A lower adjoint to `u : β → α` is a function `l : α → β` such that `l` and `u` form a Galois connection. ## Implementation details Although `lower_adjoint` is technically a generalisation of `closure_operator` (by defining `to_fun := id`), it is desirable to have both as otherwise `id`s would be carried all over the place when using concrete closure operators such as `convex_hull`. `lower_adjoint` really is a semibundled `structure` version of `galois_connection`. ## References * https://en.wikipedia.org/wiki/Closure_operator#Closure_operators_on_partially_ordered_sets -/ universe u /-! ### Closure operator -/ variables (α : Type*) {ι : Sort*} {κ : ι → Sort*} /-- A closure operator on the preorder `α` is a monotone function which is extensive (every `x` is less than its closure) and idempotent. -/ structure closure_operator [preorder α] extends α →o α := (le_closure' : ∀ x, x ≤ to_fun x) (idempotent' : ∀ x, to_fun (to_fun x) = to_fun x) namespace closure_operator instance [preorder α] : has_coe_to_fun (closure_operator α) (λ _, α → α) := ⟨λ c, c.to_fun⟩ /-- See Note [custom simps projection] -/ def simps.apply [preorder α] (f : closure_operator α) : α → α := f initialize_simps_projections closure_operator (to_order_hom_to_fun → apply, -to_order_hom) section partial_order variable [partial_order α] /-- The identity function as a closure operator. -/ @[simps] def id : closure_operator α := { to_order_hom := order_hom.id, le_closure' := λ _, le_rfl, idempotent' := λ _, rfl } instance : inhabited (closure_operator α) := ⟨id α⟩ variables {α} (c : closure_operator α) @[ext] lemma ext : ∀ (c₁ c₂ : closure_operator α), (c₁ : α → α) = (c₂ : α → α) → c₁ = c₂ | ⟨⟨c₁, _⟩, _, _⟩ ⟨⟨c₂, _⟩, _, _⟩ h := by { congr, exact h } /-- Constructor for a closure operator using the weaker idempotency axiom: `f (f x) ≤ f x`. -/ @[simps] def mk' (f : α → α) (hf₁ : monotone f) (hf₂ : ∀ x, x ≤ f x) (hf₃ : ∀ x, f (f x) ≤ f x) : closure_operator α := { to_fun := f, monotone' := hf₁, le_closure' := hf₂, idempotent' := λ x, (hf₃ x).antisymm (hf₁ (hf₂ x)) } /-- Convenience constructor for a closure operator using the weaker minimality axiom: `x ≤ f y → f x ≤ f y`, which is sometimes easier to prove in practice. -/ @[simps] def mk₂ (f : α → α) (hf : ∀ x, x ≤ f x) (hmin : ∀ ⦃x y⦄, x ≤ f y → f x ≤ f y) : closure_operator α := { to_fun := f, monotone' := λ x y hxy, hmin (hxy.trans (hf y)), le_closure' := hf, idempotent' := λ x, (hmin le_rfl).antisymm (hf _) } /-- Expanded out version of `mk₂`. `p` implies being closed. This constructor should be used when you already know a sufficient condition for being closed and using `mem_mk₃_closed` will avoid you the (slight) hassle of having to prove it both inside and outside the constructor. -/ @[simps] def mk₃ (f : α → α) (p : α → Prop) (hf : ∀ x, x ≤ f x) (hfp : ∀ x, p (f x)) (hmin : ∀ ⦃x y⦄, x ≤ y → p y → f x ≤ y) : closure_operator α := mk₂ f hf (λ x y hxy, hmin hxy (hfp y)) /-- This lemma shows that the image of `x` of a closure operator built from the `mk₃` constructor respects `p`, the property that was fed into it. -/ lemma closure_mem_mk₃ {f : α → α} {p : α → Prop} {hf : ∀ x, x ≤ f x} {hfp : ∀ x, p (f x)} {hmin : ∀ ⦃x y⦄, x ≤ y → p y → f x ≤ y} (x : α) : p (mk₃ f p hf hfp hmin x) := hfp x /-- Analogue of `closure_le_closed_iff_le` but with the `p` that was fed into the `mk₃` constructor. -/ lemma closure_le_mk₃_iff {f : α → α} {p : α → Prop} {hf : ∀ x, x ≤ f x} {hfp : ∀ x, p (f x)} {hmin : ∀ ⦃x y⦄, x ≤ y → p y → f x ≤ y} {x y : α} (hxy : x ≤ y) (hy : p y) : mk₃ f p hf hfp hmin x ≤ y := hmin hxy hy @[mono] lemma monotone : monotone c := c.monotone' /-- Every element is less than its closure. This property is sometimes referred to as extensivity or inflationarity. -/ lemma le_closure (x : α) : x ≤ c x := c.le_closure' x @[simp] lemma idempotent (x : α) : c (c x) = c x := c.idempotent' x lemma le_closure_iff (x y : α) : x ≤ c y ↔ c x ≤ c y := ⟨λ h, c.idempotent y ▸ c.monotone h, λ h, (c.le_closure x).trans h⟩ /-- An element `x` is closed for the closure operator `c` if it is a fixed point for it. -/ def closed : set α := λ x, c x = x lemma mem_closed_iff (x : α) : x ∈ c.closed ↔ c x = x := iff.rfl lemma mem_closed_iff_closure_le (x : α) : x ∈ c.closed ↔ c x ≤ x := ⟨le_of_eq, λ h, h.antisymm (c.le_closure x)⟩ lemma closure_eq_self_of_mem_closed {x : α} (h : x ∈ c.closed) : c x = x := h @[simp] lemma closure_is_closed (x : α) : c x ∈ c.closed := c.idempotent x /-- The set of closed elements for `c` is exactly its range. -/ lemma closed_eq_range_close : c.closed = set.range c := set.ext $ λ x, ⟨λ h, ⟨x, h⟩, by { rintro ⟨y, rfl⟩, apply c.idempotent }⟩ /-- Send an `x` to an element of the set of closed elements (by taking the closure). -/ def to_closed (x : α) : c.closed := ⟨c x, c.closure_is_closed x⟩ @[simp] lemma closure_le_closed_iff_le (x : α) {y : α} (hy : c.closed y) : c x ≤ y ↔ x ≤ y := by rw [←c.closure_eq_self_of_mem_closed hy, ←le_closure_iff] /-- A closure operator is equal to the closure operator obtained by feeding `c.closed` into the `mk₃` constructor. -/ lemma eq_mk₃_closed (c : closure_operator α) : c = mk₃ c c.closed c.le_closure c.closure_is_closed (λ x y hxy hy, (c.closure_le_closed_iff_le x hy).2 hxy) := by { ext, refl } /-- The property `p` fed into the `mk₃` constructor implies being closed. -/ lemma mem_mk₃_closed {f : α → α} {p : α → Prop} {hf : ∀ x, x ≤ f x} {hfp : ∀ x, p (f x)} {hmin : ∀ ⦃x y⦄, x ≤ y → p y → f x ≤ y} {x : α} (hx : p x) : x ∈ (mk₃ f p hf hfp hmin).closed := (hmin le_rfl hx).antisymm (hf _) end partial_order variable {α} section order_top variables [partial_order α] [order_top α] (c : closure_operator α) @[simp] lemma closure_top : c ⊤ = ⊤ := le_top.antisymm (c.le_closure _) lemma top_mem_closed : ⊤ ∈ c.closed := c.closure_top end order_top lemma closure_inf_le [semilattice_inf α] (c : closure_operator α) (x y : α) : c (x ⊓ y) ≤ c x ⊓ c y := c.monotone.map_inf_le _ _ section semilattice_sup variables [semilattice_sup α] (c : closure_operator α) lemma closure_sup_closure_le (x y : α) : c x ⊔ c y ≤ c (x ⊔ y) := c.monotone.le_map_sup _ _ lemma closure_sup_closure_left (x y : α) : c (c x ⊔ y) = c (x ⊔ y) := ((c.le_closure_iff _ _).1 (sup_le (c.monotone le_sup_left) (le_sup_right.trans (c.le_closure _)))).antisymm (c.monotone (sup_le_sup_right (c.le_closure _) _)) lemma closure_sup_closure_right (x y : α) : c (x ⊔ c y) = c (x ⊔ y) := by rw [sup_comm, closure_sup_closure_left, sup_comm] lemma closure_sup_closure (x y : α) : c (c x ⊔ c y) = c (x ⊔ y) := by rw [closure_sup_closure_left, closure_sup_closure_right] end semilattice_sup section complete_lattice variables [complete_lattice α] (c : closure_operator α) @[simp] lemma closure_supr_closure (f : ι → α) : c (⨆ i, c (f i)) = c (⨆ i, f i) := le_antisymm ((c.le_closure_iff _ _).1 $ supr_le $ λ i, c.monotone $ le_supr f i) $ c.monotone $ supr_mono $ λ i, c.le_closure _ @[simp] lemma closure_supr₂_closure (f : Π i, κ i → α) : c (⨆ i j, c (f i j)) = c (⨆ i j, f i j) := le_antisymm ((c.le_closure_iff _ _).1 $ supr₂_le $ λ i j, c.monotone $ le_supr₂ i j) $ c.monotone $ supr₂_mono $ λ i j, c.le_closure _ end complete_lattice end closure_operator /-! ### Lower adjoint -/ variables {α} {β : Type*} /-- A lower adjoint of `u` on the preorder `α` is a function `l` such that `l` and `u` form a Galois connection. It allows us to define closure operators whose output does not match the input. In practice, `u` is often `coe : β → α`. -/ structure lower_adjoint [preorder α] [preorder β] (u : β → α) := (to_fun : α → β) (gc' : galois_connection to_fun u) namespace lower_adjoint variable (α) /-- The identity function as a lower adjoint to itself. -/ @[simps] protected def id [preorder α] : lower_adjoint (id : α → α) := { to_fun := λ x, x, gc' := galois_connection.id } variable {α} instance [preorder α] : inhabited (lower_adjoint (id : α → α)) := ⟨lower_adjoint.id α⟩ section preorder variables [preorder α] [preorder β] {u : β → α} (l : lower_adjoint u) instance : has_coe_to_fun (lower_adjoint u) (λ _, α → β) := { coe := to_fun } /-- See Note [custom simps projection] -/ def simps.apply : α → β := l lemma gc : galois_connection l u := l.gc' @[ext] lemma ext : ∀ (l₁ l₂ : lower_adjoint u), (l₁ : α → β) = (l₂ : α → β) → l₁ = l₂ | ⟨l₁, _⟩ ⟨l₂, _⟩ h := by { congr, exact h } @[mono] lemma monotone : monotone (u ∘ l) := l.gc.monotone_u.comp l.gc.monotone_l /-- Every element is less than its closure. This property is sometimes referred to as extensivity or inflationarity. -/ lemma le_closure (x : α) : x ≤ u (l x) := l.gc.le_u_l _ end preorder section partial_order variables [partial_order α] [preorder β] {u : β → α} (l : lower_adjoint u) /-- Every lower adjoint induces a closure operator given by the composition. This is the partial order version of the statement that every adjunction induces a monad. -/ @[simps] def closure_operator : closure_operator α := { to_fun := λ x, u (l x), monotone' := l.monotone, le_closure' := l.le_closure, idempotent' := λ x, l.gc.u_l_u_eq_u (l x) } lemma idempotent (x : α) : u (l (u (l x))) = u (l x) := l.closure_operator.idempotent _ lemma le_closure_iff (x y : α) : x ≤ u (l y) ↔ u (l x) ≤ u (l y) := l.closure_operator.le_closure_iff _ _ end partial_order section preorder variables [preorder α] [preorder β] {u : β → α} (l : lower_adjoint u) /-- An element `x` is closed for `l : lower_adjoint u` if it is a fixed point: `u (l x) = x` -/ def closed : set α := λ x, u (l x) = x lemma mem_closed_iff (x : α) : x ∈ l.closed ↔ u (l x) = x := iff.rfl lemma closure_eq_self_of_mem_closed {x : α} (h : x ∈ l.closed) : u (l x) = x := h end preorder section partial_order variables [partial_order α] [partial_order β] {u : β → α} (l : lower_adjoint u) lemma mem_closed_iff_closure_le (x : α) : x ∈ l.closed ↔ u (l x) ≤ x := l.closure_operator.mem_closed_iff_closure_le _ @[simp] lemma closure_is_closed (x : α) : u (l x) ∈ l.closed := l.idempotent x /-- The set of closed elements for `l` is the range of `u ∘ l`. -/ lemma closed_eq_range_close : l.closed = set.range (u ∘ l) := l.closure_operator.closed_eq_range_close /-- Send an `x` to an element of the set of closed elements (by taking the closure). -/ def to_closed (x : α) : l.closed := ⟨u (l x), l.closure_is_closed x⟩ @[simp] lemma closure_le_closed_iff_le (x : α) {y : α} (hy : l.closed y) : u (l x) ≤ y ↔ x ≤ y := l.closure_operator.closure_le_closed_iff_le x hy end partial_order lemma closure_top [partial_order α] [order_top α] [preorder β] {u : β → α} (l : lower_adjoint u) : u (l ⊤) = ⊤ := l.closure_operator.closure_top lemma closure_inf_le [semilattice_inf α] [preorder β] {u : β → α} (l : lower_adjoint u) (x y : α) : u (l (x ⊓ y)) ≤ u (l x) ⊓ u (l y) := l.closure_operator.closure_inf_le x y section semilattice_sup variables [semilattice_sup α] [preorder β] {u : β → α} (l : lower_adjoint u) lemma closure_sup_closure_le (x y : α) : u (l x) ⊔ u (l y) ≤ u (l (x ⊔ y)) := l.closure_operator.closure_sup_closure_le x y lemma closure_sup_closure_left (x y : α) : u (l (u (l x) ⊔ y)) = u (l (x ⊔ y)) := l.closure_operator.closure_sup_closure_left x y lemma closure_sup_closure_right (x y : α) : u (l (x ⊔ u (l y))) = u (l (x ⊔ y)) := l.closure_operator.closure_sup_closure_right x y lemma closure_sup_closure (x y : α) : u (l (u (l x) ⊔ u (l y))) = u (l (x ⊔ y)) := l.closure_operator.closure_sup_closure x y end semilattice_sup section complete_lattice variables [complete_lattice α] [preorder β] {u : β → α} (l : lower_adjoint u) lemma closure_supr_closure (f : ι → α) : u (l (⨆ i, u (l (f i)))) = u (l (⨆ i, f i)) := l.closure_operator.closure_supr_closure _ lemma closure_supr₂_closure (f : Π i, κ i → α) : u (l $ ⨆ i j, u (l $ f i j)) = u (l $ ⨆ i j, f i j) := l.closure_operator.closure_supr₂_closure _ end complete_lattice /- Lemmas for `lower_adjoint (coe : α → set β)`, where `set_like α β` -/ section coe_to_set variables [set_like α β] (l : lower_adjoint (coe : α → set β)) lemma subset_closure (s : set β) : s ⊆ l s := l.le_closure s lemma not_mem_of_not_mem_closure {s : set β} {P : β} (hP : P ∉ l s) : P ∉ s := λ h, hP (subset_closure _ s h) lemma le_iff_subset (s : set β) (S : α) : l s ≤ S ↔ s ⊆ S := l.gc s S lemma mem_iff (s : set β) (x : β) : x ∈ l s ↔ ∀ S : α, s ⊆ S → x ∈ S := by { simp_rw [←set_like.mem_coe, ←set.singleton_subset_iff, ←l.le_iff_subset], exact ⟨λ h S, h.trans, λ h, h _ le_rfl⟩ } lemma eq_of_le {s : set β} {S : α} (h₁ : s ⊆ S) (h₂ : S ≤ l s) : l s = S := ((l.le_iff_subset _ _).2 h₁).antisymm h₂ lemma closure_union_closure_subset (x y : α) : (l x : set β) ∪ (l y) ⊆ l (x ∪ y) := l.closure_sup_closure_le x y @[simp] lemma closure_union_closure_left (x y : α) : (l ((l x) ∪ y) : set β) = l (x ∪ y) := l.closure_sup_closure_left x y @[simp] lemma closure_union_closure_right (x y : α) : l (x ∪ (l y)) = l (x ∪ y) := set_like.coe_injective (l.closure_sup_closure_right x y) @[simp] lemma closure_union_closure (x y : α) : l ((l x) ∪ (l y)) = l (x ∪ y) := set_like.coe_injective (l.closure_operator.closure_sup_closure x y) @[simp] lemma closure_Union_closure (f : ι → α) : l (⋃ i, l (f i)) = l (⋃ i, f i) := set_like.coe_injective $ l.closure_supr_closure _ @[simp] lemma closure_Union₂_closure (f : Π i, κ i → α) : l (⋃ i j, l (f i j)) = l (⋃ i j, f i j) := set_like.coe_injective $ l.closure_supr₂_closure _ end coe_to_set end lower_adjoint /-! ### Translations between `galois_connection`, `lower_adjoint`, `closure_operator` -/ variable {α} /-- Every Galois connection induces a lower adjoint. -/ @[simps] def galois_connection.lower_adjoint [preorder α] [preorder β] {l : α → β} {u : β → α} (gc : galois_connection l u) : lower_adjoint u := { to_fun := l, gc' := gc } /-- Every Galois connection induces a closure operator given by the composition. This is the partial order version of the statement that every adjunction induces a monad. -/ @[simps] def galois_connection.closure_operator [partial_order α] [preorder β] {l : α → β} {u : β → α} (gc : galois_connection l u) : closure_operator α := gc.lower_adjoint.closure_operator /-- The set of closed elements has a Galois insertion to the underlying type. -/ def closure_operator.gi [partial_order α] (c : closure_operator α) : galois_insertion c.to_closed coe := { choice := λ x hx, ⟨x, hx.antisymm (c.le_closure x)⟩, gc := λ x y, (c.closure_le_closed_iff_le _ y.2), le_l_u := λ x, c.le_closure _, choice_eq := λ x hx, le_antisymm (c.le_closure x) hx } /-- The Galois insertion associated to a closure operator can be used to reconstruct the closure operator. Note that the inverse in the opposite direction does not hold in general. -/ @[simp] lemma closure_operator_gi_self [partial_order α] (c : closure_operator α) : c.gi.gc.closure_operator = c := by { ext x, refl }