roseking's picture
Sync from openai/math@faee2cf485a8 (part 24)
e3e0a0d verified
Raw History Blame Contribute Delete
28.3 kB
import Mathlib
import OAI.SetTheory.PartitionConsistency.Logic.Proofs
import OAI.SetTheory.PartitionConsistency.Logic.CountableModels
namespace OAI
namespace PartitionConsistency
/-! Countable Henkin completeness for the target proof calculus. -/
deriving instance Countable for Formula
namespace Formula
@[simp] theorem liftRen_id (n : ℕ) : liftRen (id : Fin n → Fin n) = id := by
funext i
refine Fin.lastCases ?_ (fun j => ?_) i <;> simp [liftRen]
@[simp] theorem liftRen_comp {n m k : ℕ} (r : Fin n → Fin m) (s : Fin m → Fin k) :
liftRen (s ∘ r) = liftRen s ∘ liftRen r := by
funext i
refine Fin.lastCases ?_ (fun j => ?_) i <;> simp [liftRen]
@[simp] theorem rename_id {n} (φ : Formula n) : φ.rename id = φ := by
induction φ with
| bot => rfl
| mem => rfl
| equal => rfl
| imp φ ψ hφ hψ => simp [rename, hφ, hψ]
| conj φ ψ hφ hψ => simp [rename, hφ, hψ]
| disj φ ψ hφ hψ => simp [rename, hφ, hψ]
| all φ hφ => simp [rename, hφ]
| ex φ hφ => simp [rename, hφ]
@[simp] theorem rename_comp {n m k} (φ : Formula n)
(r : Fin n → Fin m) (s : Fin m → Fin k) :
(φ.rename r).rename s = φ.rename (s ∘ r) := by
induction φ generalizing m k with
| bot => rfl
| mem => rfl
| equal => rfl
| imp φ ψ hφ hψ => simp [rename, hφ, hψ]
| conj φ ψ hφ hψ => simp [rename, hφ, hψ]
| disj φ ψ hφ hψ => simp [rename, hφ, hψ]
| all φ hφ => simp [rename, hφ, liftRen_comp]
| ex φ hφ => simp [rename, hφ, liftRen_comp]
/-- Instantiation after renaming the free variables extends that renaming by
one chosen variable; this is the capture-avoidance identity for witnesses. -/
@[simp] theorem instantiate_rename {n m} (φ : Formula (n + 1))
(r : Fin n → Fin m) (t : Fin m) :
(φ.rename (liftRen r)).instantiate t = φ.rename (Fin.snoc r t) := by
rw [instantiate, rename_comp]
congr 1
funext i
refine Fin.lastCases ?_ (fun j => ?_) i <;> simp [liftRen]
@[simp] theorem rename_snoc_castSucc_last {n} (φ : Formula (n + 1)) :
φ.rename (Fin.snoc Fin.castSucc (Fin.last n)) = φ := by
have h : Fin.snoc Fin.castSucc (Fin.last n) = (id : Fin (n+1) → Fin (n+1)) := by
funext i
refine Fin.lastCases ?_ (fun j => ?_) i <;> simp
rw [h, rename_id]
@[simp] theorem rename_instantiate {n m} (φ : Formula (n+1))
(r : Fin n → Fin m) (t : Fin n) :
(φ.instantiate t).rename r = (φ.rename (liftRen r)).instantiate (r t) := by
rw [instantiate_rename, instantiate, rename_comp]
congr 1
funext i
refine Fin.lastCases ?_ (fun j => ?_) i <;> simp
end Formula
namespace Derives
theorem eq_symm {n} {Γ : Set (Formula n)} {s t : Fin n} (h : Derives Γ (.equal s t)) :
Derives Γ (.equal t s) := by
have hp := Derives.eqSubst (.equal (Fin.last n) s.castSucc) h (by
simpa [Formula.instantiate, Formula.rename] using (Derives.eqRefl (Γ := Γ) s))
simpa [Formula.instantiate, Formula.rename] using hp
theorem eq_trans {n} {Γ : Set (Formula n)} {s t u : Fin n}
(h : Derives Γ (.equal s t)) (h' : Derives Γ (.equal t u)) :
Derives Γ (.equal s u) := by
have hp := Derives.eqSubst (.equal s.castSucc (Fin.last n)) h' (by
simpa [Formula.instantiate, Formula.rename] using h)
simpa [Formula.instantiate, Formula.rename] using hp
theorem mem_congr {n} {Γ : Set (Formula n)} {s t s' t' : Fin n}
(hs : Derives Γ (.equal s s')) (ht : Derives Γ (.equal t t'))
(hm : Derives Γ (.mem s t)) : Derives Γ (.mem s' t') := by
have h₁ : Derives Γ (.mem s' t) := by
have hp := Derives.eqSubst (.mem (Fin.last n) t.castSucc) hs (by
simpa [Formula.instantiate, Formula.rename] using hm)
simpa [Formula.instantiate, Formula.rename] using hp
have hp := Derives.eqSubst (.mem s'.castSucc (Fin.last n)) ht (by
simpa [Formula.instantiate, Formula.rename] using h₁)
simpa [Formula.instantiate, Formula.rename] using hp
end Derives
namespace Completeness
/-- Consistency of an open context, with its displayed finite variable supply. -/
def ContextConsistent {n} (Γ : Set (Formula n)) : Prop := ¬ Derives Γ .bot
/-- Enlarging the variable supply is conservative for a nonempty supply.
A retraction substitutes an existing variable for every extra variable. -/
theorem lift_consistent {n m} {Γ : Set (Formula n)} (hn : 0 < n) (hnm : n ≤ m)
(hΓ : ContextConsistent Γ) :
ContextConsistent (Formula.rename (Fin.castLE hnm) '' Γ) := by
intro hd
let r : Fin m → Fin n := fun i => if h : i.val < n then ⟨i.val, h⟩ else ⟨0, hn⟩
have hr : r ∘ Fin.castLE hnm = id := by
funext i
simp [r]
apply hΓ
have h := hd.rename r
simpa [Formula.rename, Set.image_image, hr] using h
/-- One of the two decisions can always be added without contradiction. -/
theorem consistent_decision {n} {Γ : Set (Formula n)} (hΓ : ContextConsistent Γ)
(φ : Formula n) :
ContextConsistent (insert φ Γ) ∨ ContextConsistent (insert φ.neg Γ) := by
classical
by_cases hp : ContextConsistent (insert φ Γ)
· exact Or.inl hp
· right
intro hn
apply hΓ
exact Derives.impE (Derives.impI (not_not.mp hp)) (Derives.classical hn)
/-- Henkin conditions are genuine consistent first-order contexts. No
satisfiability or model-existence assertion occurs in their definition. -/
structure Condition where
size : ℕ
positive : 0 < size
context : Set (Formula size)
consistent : ContextConsistent context
namespace Condition
/-- A stronger condition may have more variables and more assumptions. -/
instance : Preorder Condition where
le q p := ∃ h : p.size ≤ q.size,
Formula.rename (Fin.castLE h) '' p.context ⊆ q.context
le_refl p := by
refine ⟨le_rfl, ?_⟩
rintro _ ⟨φ, hφ, rfl⟩
have h : (Fin.castLE (le_rfl : p.size ≤ p.size)) = id := rfl
simpa [h] using hφ
le_trans q p r hqp hpr := by
obtain ⟨hqp, hpq⟩ := hqp
obtain ⟨hpr, hrp⟩ := hpr
refine ⟨hpr.trans hqp, ?_⟩
rintro _ ⟨φ, hφ, rfl⟩
have hmiddle := hpq ⟨_, hrp ⟨φ, hφ, rfl⟩, rfl⟩
simpa only [Formula.rename_comp, Function.comp_def, Fin.castLE_castLE] using hmiddle
/-- The canonical embedding of the old variables into a stronger condition. -/
def emb {q p : Condition} (h : q ≤ p) : Fin p.size → Fin q.size := Fin.castLE h.choose
@[simp] theorem emb_val {q p : Condition} (h : q ≤ p) (i : Fin p.size) :
(emb h i).val = i.val := rfl
/-- Derivations are preserved by extension of conditions. -/
theorem transport {q p : Condition} (h : q ≤ p) {φ : Formula p.size}
(hd : Derives p.context φ) : Derives q.context (φ.rename (emb h)) :=
(hd.rename (emb h)).weaken h.choose_spec
def grow (p : Condition) (m : ℕ) (h : p.size ≤ m) : Condition :=
⟨m, p.positive.trans_le h, Formula.rename (Fin.castLE h) '' p.context,
lift_consistent p.positive h p.consistent⟩
theorem grow_le (p : Condition) (m : ℕ) (h : p.size ≤ m) : p.grow m h ≤ p :=
⟨h, Set.Subset.rfl⟩
def add (p : Condition) (φ : Formula p.size)
(h : ContextConsistent (insert φ p.context)) : Condition :=
⟨p.size, p.positive, insert φ p.context, h⟩
theorem add_le (p : Condition) (φ : Formula p.size)
(h : ContextConsistent (insert φ p.context)) : p.add φ h ≤ p := by
refine ⟨le_rfl, ?_⟩
rintro _ ⟨ψ, hψ, rfl⟩
have hid : (Fin.castLE (le_rfl : p.size ≤ p.size)) = id := rfl
change ψ.rename (Fin.castLE (le_rfl : p.size ≤ p.size)) ∈ insert φ p.context
rw [hid, Formula.rename_id]
exact Set.mem_insert_of_mem φ hψ
/-- A condition can always accommodate a given finite tuple of variable codes. -/
theorem bound (p : Condition) {n} (v : Fin n → ℕ) :
∃ q : Condition, q ≤ p ∧ ∃ w : Fin n → Fin q.size, ∀ i, (w i).val = v i := by
let m := max p.size (Finset.univ.sup v + 1)
have hpm : p.size ≤ m := le_max_left _ _
let q := p.grow m hpm
have hv : ∀ i, v i < q.size := by
intro i
exact (Nat.lt_succ_of_le (Finset.le_sup (Finset.mem_univ i))).trans_le (le_max_right _ _)
exact ⟨q, p.grow_le m hpm, fun i => ⟨v i, hv i⟩, fun _ => rfl⟩
end Condition
/-- Extending the variable map by a new last variable is the binder lift. -/
@[simp] theorem snoc_castSucc {n m} (w : Fin n → Fin m) :
Fin.snoc (Fin.castSucc ∘ w) (Fin.last m) = Formula.liftRen w := by
funext i
refine Fin.lastCases ?_ (fun j => ?_) i <;> simp [Formula.liftRen]
namespace Condition
/-- Adding a fresh local witness is consistent when existence is derivable.
This is justified by the eigenvariable rule, not by a model assumption. -/
def exWitness (p : Condition) (φ : Formula (p.size + 1))
(hφ : Derives p.context (.ex φ)) : Condition where
size := p.size + 1
positive := Nat.zero_lt_succ _
context := insert φ (Formula.raise '' p.context)
consistent := fun hc => p.consistent (Derives.exE hφ hc)
theorem exWitness_le (p : Condition) (φ : Formula (p.size + 1))
(hφ : Derives p.context (.ex φ)) : p.exWitness φ hφ ≤ p := by
refine ⟨Nat.le_succ _, ?_⟩
rintro _ ⟨ψ, hψ, rfl⟩
exact Or.inr ⟨ψ, hψ, rfl⟩
/-- A negated universal has a consistent fresh counterexample context. -/
def allCounterexample (p : Condition) (φ : Formula (p.size + 1))
(hφ : Derives p.context (Formula.all φ).neg) : Condition where
size := p.size + 1
positive := Nat.zero_lt_succ _
context := insert φ.neg (Formula.raise '' p.context)
consistent := fun hc =>
p.consistent (Derives.impE hφ (Derives.allI (Derives.classical hc)))
theorem allCounterexample_le (p : Condition) (φ : Formula (p.size + 1))
(hφ : Derives p.context (Formula.all φ).neg) : p.allCounterexample φ hφ ≤ p := by
refine ⟨Nat.le_succ _, ?_⟩
rintro _ ⟨ψ, hψ, rfl⟩
exact Or.inr ⟨ψ, hψ, rfl⟩
end Condition
/-- All finite tuples are explicitly included, so this construction treats
all parameters, not just sentences or quantifier-free formulas. -/
def decisions {n} (φ : Formula n) (v : Fin n → ℕ) : Set Condition :=
{c | ∃ w : Fin n → Fin c.size, (∀ i, (w i).val = v i) ∧
(Derives c.context (φ.rename w) ∨ Derives c.context (φ.neg.rename w))}
def exWitnesses {n} (φ : Formula (n+1)) (v : Fin n → ℕ) : Set Condition :=
{c | ∃ w : Fin n → Fin c.size, (∀ i, (w i).val = v i) ∧
(Derives c.context ((Formula.ex φ).neg.rename w) ∨
∃ a : Fin c.size, Derives c.context (φ.rename (Fin.snoc w a)))}
def allCounterexamples {n} (φ : Formula (n+1)) (v : Fin n → ℕ) : Set Condition :=
{c | ∃ w : Fin n → Fin c.size, (∀ i, (w i).val = v i) ∧
(Derives c.context ((Formula.all φ).rename w) ∨
∃ a : Fin c.size, Derives c.context (φ.neg.rename (Fin.snoc w a)))}
theorem decisions_dense {n} (φ : Formula n) (v : Fin n → ℕ) :
ForcingQuotient.Dense (decisions φ v) := by
intro p
obtain ⟨q, hqp, w, hw⟩ := p.bound v
rcases consistent_decision q.consistent (φ.rename w) with hpos | hneg
· exact ⟨q.add (φ.rename w) hpos, (q.add_le _ _).trans hqp, w, hw,
Or.inl (Derives.hyp (Set.mem_insert _ _))⟩
· exact ⟨q.add (φ.rename w).neg hneg, (q.add_le _ _).trans hqp, w, hw,
Or.inr (Derives.hyp (Set.mem_insert _ _))⟩
theorem exWitnesses_dense {n} (φ : Formula (n+1)) (v : Fin n → ℕ) :
ForcingQuotient.Dense (exWitnesses φ v) := by
intro p
obtain ⟨q, hqp, w, hw⟩ := p.bound v
let ψ := φ.rename (Formula.liftRen w)
rcases consistent_decision q.consistent (.ex ψ) with hpos | hneg
· let r := q.add (.ex ψ) hpos
have hex : Derives r.context (.ex ψ) := Derives.hyp (Set.mem_insert _ _)
let s := r.exWitness ψ hex
refine ⟨s, (r.exWitness_le ψ hex).trans ((q.add_le _ _).trans hqp),
Fin.castSucc ∘ w, hw, Or.inr ⟨Fin.last r.size, ?_⟩⟩
dsimp only [s, r, Condition.exWitness, Condition.add]
rw [snoc_castSucc]
exact Derives.hyp (Set.mem_insert _ _)
· exact ⟨q.add (Formula.ex ψ).neg hneg, (q.add_le _ _).trans hqp, w, hw,
Or.inl (Derives.hyp (Set.mem_insert _ _))⟩
theorem allCounterexamples_dense {n} (φ : Formula (n+1)) (v : Fin n → ℕ) :
ForcingQuotient.Dense (allCounterexamples φ v) := by
intro p
obtain ⟨q, hqp, w, hw⟩ := p.bound v
let ψ := φ.rename (Formula.liftRen w)
rcases consistent_decision q.consistent (.all ψ) with hpos | hneg
· exact ⟨q.add (.all ψ) hpos, (q.add_le _ _).trans hqp, w, hw,
Or.inl (Derives.hyp (Set.mem_insert _ _))⟩
· let r := q.add (Formula.all ψ).neg hneg
have hn : Derives r.context (Formula.all ψ).neg := Derives.hyp (Set.mem_insert _ _)
let s := r.allCounterexample ψ hn
refine ⟨s, (r.allCounterexample_le ψ hn).trans ((q.add_le _ _).trans hqp),
Fin.castSucc ∘ w, hw, Or.inr ⟨Fin.last r.size, ?_⟩⟩
dsimp only [s, r, Condition.allCounterexample, Condition.add]
rw [snoc_castSucc]
exact Derives.hyp (Set.mem_insert _ _)
abbrev GenericFilter := ForcingQuotient.GenericFilter Condition
/-- A filter meeting every decision and every local quantifier-witness request. -/
def IsHenkin (G : GenericFilter) : Prop :=
(∀ n (φ : Formula n) (v : Fin n → ℕ), ∃ p ∈ G, p ∈ decisions φ v) ∧
(∀ n (φ : Formula (n+1)) (v : Fin n → ℕ), ∃ p ∈ G, p ∈ exWitnesses φ v) ∧
(∀ n (φ : Formula (n+1)) (v : Fin n → ℕ), ∃ p ∈ G, p ∈ allCounterexamples φ v)
/-- The countable Henkin construction, using only syntactic consistency. -/
theorem exists_henkin (p₀ : Condition) :
∃ G : GenericFilter, p₀ ∈ G ∧ IsHenkin G := by
let Request (j : ℕ) := Σ n, Formula (n+j) × (Fin n → ℕ)
let D : Request 0 ⊕ (Request 1 ⊕ Request 1) → Set Condition :=
fun i => match i with
| .inl ⟨_, φ, v⟩ => decisions φ v
| .inr (.inl ⟨_, φ, v⟩) => exWitnesses φ v
| .inr (.inr ⟨_, φ, v⟩) => allCounterexamples φ v
have hD : ∀ i, ForcingQuotient.Dense (D i) := by
rintro (⟨n, φ, v⟩ | ⟨n, φ, v⟩ | ⟨n, φ, v⟩)
· exact decisions_dense φ v
· exact exWitnesses_dense φ v
· exact allCounterexamples_dense φ v
obtain ⟨G, hp₀, hG⟩ := ForcingQuotient.exists_generic_through D hD p₀
exact ⟨G, hp₀, (fun n φ v => hG (.inl ⟨n, φ, v⟩)),
(fun n φ v => hG (.inr (.inl ⟨n, φ, v⟩))),
(fun n φ v => hG (.inr (.inr ⟨n, φ, v⟩)))⟩
/-- Eventual derivability in the directed filter, with all free variables
interpreted as natural-number constant codes. -/
def Proven (G : GenericFilter) {n} (φ : Formula n) (v : Fin n → ℕ) : Prop :=
∃ p ∈ G, ∃ w : Fin n → Fin p.size,
(∀ i, (w i).val = v i) ∧ Derives p.context (φ.rename w)
namespace Proven
variable {G : GenericFilter} {n : ℕ} {φ ψ : Formula n} {v : Fin n → ℕ}
/-- Any existing representation of the tuple can be extended to one carrying
the given derivation. Directedness, not an assumed complete theory, is used. -/
theorem at_condition (h : Proven G φ v) {p : Condition} (hp : p ∈ G)
(w : Fin n → Fin p.size) (hw : ∀ i, (w i).val = v i) :
∃ q : Condition, q ∈ G ∧ ∃ hqp : q ≤ p,
Derives q.context (φ.rename (Condition.emb hqp ∘ w)) := by
obtain ⟨r, hr, u, hu, hd⟩ := h
obtain ⟨q, hq, hqr, hqp⟩ := G.directed _ hr _ hp
refine ⟨q, hq, hqp, ?_⟩
have he : Condition.emb hqr ∘ u = Condition.emb hqp ∘ w := by
funext i
apply Fin.ext
exact (hu i).trans (hw i).symm
simpa only [Formula.rename_comp, he] using Condition.transport hqr hd
/-- Two eventual derivations with the same tuple can be put into a single
consistent context with exactly the same variable map. -/
theorem joint (hφ : Proven G φ v) (hψ : Proven G ψ v) :
∃ p ∈ G, ∃ w : Fin n → Fin p.size, (∀ i, (w i).val = v i) ∧
Derives p.context (φ.rename w) ∧ Derives p.context (ψ.rename w) := by
obtain ⟨p, hp, w, hw, hdφ⟩ := hφ
obtain ⟨q, hq, hqp, hdψ⟩ := hψ.at_condition hp w hw
refine ⟨q, hq, Condition.emb hqp ∘ w, hw, ?_, hdψ⟩
simpa only [Formula.rename_comp] using Condition.transport hqp hdφ
theorem not_bot : ¬ Proven G (Formula.bot : Formula n) v := by
rintro ⟨p, _, _, _, hd⟩
exact p.consistent hd
theorem conjunction : Proven G (φ.conj ψ) v ↔ Proven G φ v ∧ Proven G ψ v := by
constructor
· rintro ⟨p, hp, w, hw, hd⟩
exact ⟨⟨p, hp, w, hw, hd.conjL⟩, ⟨p, hp, w, hw, hd.conjR⟩⟩
· rintro ⟨hφ, hψ⟩
obtain ⟨p, hp, w, hw, hdφ, hdψ⟩ := hφ.joint hψ
exact ⟨p, hp, w, hw, Derives.conjI hdφ hdψ⟩
theorem apply_imp (hpq : Proven G (φ.imp ψ) v) (hp : Proven G φ v) : Proven G ψ v := by
obtain ⟨p, hp, w, hw, hd₁, hd₂⟩ := hpq.joint hp
exact ⟨p, hp, w, hw, Derives.impE hd₁ hd₂⟩
theorem contradict (hn : Proven G φ.neg v) (hp : Proven G φ v) : False :=
not_bot (hn.apply_imp hp)
variable (hG : IsHenkin G)
include hG
/-- This also supplies a filter condition containing any prescribed finite
collection of constant codes, including codes not used by a formula. -/
theorem tuple (v : Fin n → ℕ) :
∃ p ∈ G, ∃ w : Fin n → Fin p.size, ∀ i, (w i).val = v i := by
obtain ⟨p, hp, w, hw, _⟩ := hG.1 n .bot v
exact ⟨p, hp, w, hw⟩
theorem decision (φ : Formula n) (v : Fin n → ℕ) :
Proven G φ v ∨ Proven G φ.neg v := by
obtain ⟨p, hp, w, hw, hd | hd⟩ := hG.1 n φ v
· exact Or.inl ⟨p, hp, w, hw, hd⟩
· exact Or.inr ⟨p, hp, w, hw, hd⟩
theorem negation : Proven G φ.neg v ↔ ¬ Proven G φ v := by
constructor
· exact contradict
· intro hn
exact (decision hG φ v).resolve_left hn
/-- Full substitution of finite variable tuples, in both directions. The
reverse implication includes variables unused in the substituted formula. -/
theorem rename_iff {m} (φ : Formula n) (r : Fin n → Fin m) (v : Fin m → ℕ) :
Proven G (φ.rename r) v ↔ Proven G φ (v ∘ r) := by
constructor
· rintro ⟨p, hp, w, hw, hd⟩
refine ⟨p, hp, w ∘ r, (fun i => hw (r i)), ?_⟩
simpa only [Formula.rename_comp] using hd
· intro hφ
obtain ⟨p, hp, w, hw⟩ := tuple hG v
obtain ⟨q, hq, hqp, hd⟩ := hφ.at_condition hp (w ∘ r) (fun i => hw (r i))
refine ⟨q, hq, Condition.emb hqp ∘ w, hw, ?_⟩
simpa only [Formula.rename_comp, Function.comp_assoc] using hd
theorem implication : Proven G (φ.imp ψ) v ↔ (Proven G φ v → Proven G ψ v) := by
constructor
· exact apply_imp
· intro h
classical
by_cases hφ : Proven G φ v
· obtain ⟨p, hp, w, hw, hd⟩ := h hφ
exact ⟨p, hp, w, hw, Derives.impI (hd.weaken (Set.subset_insert _ _))⟩
· obtain ⟨p, hp, w, hw, hd⟩ := (negation hG).2 hφ
refine ⟨p, hp, w, hw, Derives.impI ?_⟩
exact (Derives.impE (hd.weaken (Set.subset_insert _ _))
(Derives.hyp (Set.mem_insert _ _))).botE
theorem disjunction : Proven G (φ.disj ψ) v ↔ Proven G φ v ∨ Proven G ψ v := by
constructor
· intro h
classical
by_cases hφ : Proven G φ v
· exact Or.inl hφ
· right
obtain ⟨p, hp, w, hw, hd, hn⟩ := h.joint ((negation hG).2 hφ)
refine ⟨p, hp, w, hw, Derives.disjE hd ?_ ?_⟩
· exact (Derives.impE (hn.weaken (Set.subset_insert _ _))
(Derives.hyp (Set.mem_insert _ _))).botE
· exact Derives.hyp (Set.mem_insert _ _)
· rintro (⟨p, hp, w, hw, hd⟩ | ⟨p, hp, w, hw, hd⟩)
· exact ⟨p, hp, w, hw, hd.disjL⟩
· exact ⟨p, hp, w, hw, hd.disjR⟩
omit hG in
/-- Universal elimination is a derivable operation on the eventual theory. -/
theorem all_instance {φ : Formula (n+1)} (h : Proven G (.all φ) v) (t : Fin n) :
Proven G (φ.instantiate t) v := by
obtain ⟨p, hp, w, hw, hd⟩ := h
refine ⟨p, hp, w, hw, ?_⟩
rw [Formula.rename_instantiate]
exact Derives.allE (w t) hd
/-- The Henkin requests turn existence into an actual constant-code witness. -/
theorem exists_iff (φ : Formula (n+1)) (v : Fin n → ℕ) :
Proven G (.ex φ) v ↔ ∃ a : ℕ, Proven G φ (Fin.snoc v a) := by
constructor
· intro h
obtain ⟨p, hp, w, hw, hn | ⟨a, ha⟩⟩ := hG.2.1 n φ v
· exact (contradict ⟨p, hp, w, hw, hn⟩ h).elim
· refine ⟨a.val, p, hp, Fin.snoc w a, ?_, ha⟩
intro i
refine Fin.lastCases ?_ (fun j => ?_) i <;> simp [hw]
· rintro ⟨a, p, hp, w, hw, hd⟩
refine ⟨p, hp, Fin.init w, ?_, ?_⟩
· intro i
simpa only [Fin.init, Fin.snoc_castSucc] using hw i.castSucc
· apply Derives.exI (w (Fin.last n))
simpa only [Formula.instantiate_rename, Fin.snoc_init_self] using hd
/-- Universal truth is equivalent to truth for every constant code. The
counterexample requests are essential to the reverse implication. -/
theorem all_iff (φ : Formula (n+1)) (v : Fin n → ℕ) :
Proven G (.all φ) v ↔ ∀ a : ℕ, Proven G φ (Fin.snoc v a) := by
constructor
· intro h a
have hr : Proven G (Formula.all φ).raise (Fin.snoc v a) := by
apply (rename_iff hG (.all φ) Fin.castSucc _).2
simpa only [Fin.snoc_comp_castSucc] using h
have hi := all_instance hr (Fin.last n)
simpa only [Formula.raise, Formula.rename, Formula.instantiate_rename,
Formula.rename_snoc_castSucc_last] using hi
· intro h
obtain ⟨p, hp, w, hw, hall | ⟨a, ha⟩⟩ := hG.2.2 n φ v
· exact ⟨p, hp, w, hw, hall⟩
· have hn : Proven G φ.neg (Fin.snoc v a.val) := by
refine ⟨p, hp, Fin.snoc w a, ?_, ha⟩
intro i
refine Fin.lastCases ?_ (fun j => ?_) i <;> simp [hw]
exact (contradict hn (h a.val)).elim
end Proven
/-- Equality and membership of constant codes in the canonical term model. -/
def EqRel (G : GenericFilter) (a b : ℕ) : Prop :=
Proven G (.equal (0 : Fin 2) 1) ![a,b]
def MemRel (G : GenericFilter) (a b : ℕ) : Prop :=
Proven G (.mem (0 : Fin 2) 1) ![a,b]
variable {G : GenericFilter} (hG : IsHenkin G)
include hG
theorem equal_iff {n} (i j : Fin n) (v : Fin n → ℕ) :
Proven G (.equal i j) v ↔ EqRel G (v i) (v j) := by
have h := Proven.rename_iff hG (.equal (0 : Fin 2) 1) ![i,j] v
have hv : v ∘ ![i,j] = ![v i, v j] := by
funext k
fin_cases k <;> rfl
rw [hv] at h
exact h
theorem mem_iff {n} (i j : Fin n) (v : Fin n → ℕ) :
Proven G (.mem i j) v ↔ MemRel G (v i) (v j) := by
have h := Proven.rename_iff hG (.mem (0 : Fin 2) 1) ![i,j] v
have hv : v ∘ ![i,j] = ![v i, v j] := by
funext k
fin_cases k <;> rfl
rw [hv] at h
exact h
theorem eqRel_refl (a : ℕ) : EqRel G a a := by
obtain ⟨p, hp, w, hw⟩ := Proven.tuple hG ![a,a]
refine ⟨p, hp, w, hw, ?_⟩
have he : w 0 = w 1 := Fin.ext ((hw 0).trans (hw 1).symm)
change Derives p.context (.equal (w 0) (w 1))
rw [he]
exact Derives.eqRefl _
omit hG in
theorem eqRel_symm {a b : ℕ} (h : EqRel G a b) : EqRel G b a := by
obtain ⟨p, hp, w, hw, hd⟩ := h
refine ⟨p, hp, ![w 1, w 0], ?_, hd.eq_symm⟩
intro i
fin_cases i <;> simp [hw]
theorem eqRel_trans {a b c : ℕ} (hab : EqRel G a b) (hbc : EqRel G b c) :
EqRel G a c := by
have h₁ := (equal_iff hG (0 : Fin 3) 1 ![a,b,c]).2 hab
have h₂ := (equal_iff hG (1 : Fin 3) 2 ![a,b,c]).2 hbc
obtain ⟨p, hp, w, hw, hd₁, hd₂⟩ := h₁.joint h₂
have hc : Proven G (.equal (0 : Fin 3) 2) ![a,b,c] :=
⟨p, hp, w, hw, hd₁.eq_trans hd₂⟩
exact (equal_iff hG (0 : Fin 3) 2 ![a,b,c]).1 hc
theorem memRel_congr {a b a' b' : ℕ} (ha : EqRel G a a') (hb : EqRel G b b') :
MemRel G a b ↔ MemRel G a' b' := by
have transport : ∀ {a b a' b' : ℕ}, EqRel G a a' → EqRel G b b' →
MemRel G a b → MemRel G a' b' := by
intro a b a' b' ha hb hm
have h₁ := (equal_iff hG (0 : Fin 4) 2 ![a,b,a',b']).2 ha
have h₂ := (equal_iff hG (1 : Fin 4) 3 ![a,b,a',b']).2 hb
have h₃ := (mem_iff hG (0 : Fin 4) 1 ![a,b,a',b']).2 hm
obtain ⟨p, hp, w, hw, hd₁₂, hd₃⟩ := ((Proven.conjunction).2 ⟨h₁,h₂⟩).joint h₃
have h' : Proven G (.mem (2 : Fin 4) 3) ![a,b,a',b'] :=
⟨p, hp, w, hw, Derives.mem_congr hd₁₂.conjL hd₁₂.conjR hd₃⟩
exact (mem_iff hG (2 : Fin 4) 3 ![a,b,a',b']).1 h'
exact ⟨transport ha hb, transport (eqRel_symm ha) (eqRel_symm hb)⟩
/-- Quotient of the natural-number constant codes by their syntactically
provable equality. -/
def termSetoid (G : GenericFilter) (hG : IsHenkin G) : Setoid ℕ where
r := EqRel G
iseqv := ⟨eqRel_refl hG, eqRel_symm, eqRel_trans hG⟩
def TermModel (G : GenericFilter) (hG : IsHenkin G) : Type := Quotient (termSetoid G hG)
def termClass (G : GenericFilter) (hG : IsHenkin G) (a : ℕ) : TermModel G hG :=
Quotient.mk _ a
instance termModel_nonempty : Nonempty (TermModel G hG) := ⟨termClass G hG 0⟩
instance termModel_countable : Countable (TermModel G hG) :=
inferInstanceAs (Countable (Quotient (termSetoid G hG)))
/-- Membership is independent of representatives because first-order equality
has its actual substitution rule. -/
def termMem (G : GenericFilter) (hG : IsHenkin G) :
TermModel G hG → TermModel G hG → Prop :=
Quotient.lift₂ (MemRel G) (fun _ _ _ _ ha hb => propext (memRel_congr hG ha hb))
/-- Truth in the countable term model, for every formula and every tuple of
constant codes. This is syntactic completeness, not compactness assumed as a
premise or semantic consistency used as a definition. -/
theorem term_truth {n} (φ : Formula n) (v : Fin n → ℕ) :
Formula.Realize (termMem G hG) (termClass G hG ∘ v) φ ↔ Proven G φ v := by
induction φ with
| bot => exact ⟨False.elim, Proven.not_bot⟩
| mem i j => exact (mem_iff hG i j v).symm
| equal i j => exact Quotient.eq.trans (equal_iff hG i j v).symm
| conj φ ψ hφ hψ =>
simp only [Formula.Realize, hφ v, hψ v, Proven.conjunction]
| imp φ ψ hφ hψ =>
simp only [Formula.Realize, hφ v, hψ v, Proven.implication hG]
| disj φ ψ hφ hψ =>
simp only [Formula.Realize, hφ v, hψ v, Proven.disjunction hG]
| all φ hφ =>
rw [Proven.all_iff hG]
constructor
· intro h a
apply (hφ (Fin.snoc v a)).1
simpa only [Fin.comp_snoc, termClass] using h (termClass G hG a)
· intro h x
obtain ⟨a, rfl⟩ := Quotient.exists_rep x
simpa only [Fin.comp_snoc, termClass] using (hφ (Fin.snoc v a)).2 (h a)
| ex φ hφ =>
rw [Proven.exists_iff hG]
constructor
· rintro ⟨x, hx⟩
obtain ⟨a, rfl⟩ := Quotient.exists_rep x
refine ⟨a, (hφ (Fin.snoc v a)).1 ?_⟩
simpa only [Fin.comp_snoc, termClass] using hx
· rintro ⟨a, ha⟩
refine ⟨termClass G hG a, ?_⟩
simpa only [Fin.comp_snoc, termClass] using (hφ (Fin.snoc v a)).2 ha
omit hG in
/-- Countable-model completeness for the actual finite-derivation consistency
predicate used in the target. No transitive or externally well-founded model
is asserted. The domain is concretely a quotient of `ℕ`. -/
theorem exists_countable_model {T : Theory} (hT : Consistent T) :
∃ M : Type, Nonempty M ∧ Countable M ∧ ∃ E : M → M → Prop, Models E T := by
let p₀ : Condition := ⟨1, Nat.zero_lt_succ 0, T.at 1, hT 1⟩
obtain ⟨G, hp₀, hG⟩ := exists_henkin p₀
refine ⟨TermModel G hG, inferInstance, inferInstance, termMem G hG, ?_⟩
intro φ hφ
have ht := term_truth hG φ Fin.elim0
have hv : termClass G hG ∘ Fin.elim0 = (Fin.elim0 : Fin 0 → TermModel G hG) :=
Subsingleton.elim _ _
rw [hv] at ht
apply ht.2
refine ⟨p₀, hp₀, Fin.elim0, (fun i => Fin.elim0 i), Derives.hyp ?_⟩
exact ⟨φ, hφ, rfl⟩
omit hG in
/-- Equivalence of syntactic consistency and countable nonempty models for
this first-order language. This does not perform the constructible or
symmetric-forcing interpretations required in `main_consistency`. -/
theorem consistent_iff_has_countable_model (T : Theory) :
Consistent T ↔
∃ M : Type, Nonempty M ∧ Countable M ∧ ∃ E : M → M → Prop, Models E T := by
constructor
· exact exists_countable_model
· rintro ⟨M, hM, _, E, hE⟩
let _ := hM
exact consistent_of_model E hE
end Completeness
end PartitionConsistency
end OAI