/- Copyright (c) 2015 Microsoft Corporation. All rights reserved. Released under Apache 2.0 license as described in the file LICENSE. Author: Leonardo de Moura Formalization of Theorem 1 from the following paper: "The inconsistency of a Brouwerian continuity principle with the Curry–Howard interpretation" by Martín Escardó and Chuangjie Xu -/ import data.nat open nat sigma.ops /- Bounded equality: α and β agree in the first n positions. -/ definition beq (α β : nat → nat) (n : nat) : Prop := ∀ a, a < n → α a = β a notation α `=[`:50 n:50 `]` β:50 := beq α β n lemma pred_beq {α β : nat → nat} {n : nat} : α =[n+1] β → α =[n] β := λ h a altn, h a (lt.step altn) definition continuous (f : (nat → nat) → nat) : Type₁ := ∀ α, Σ n, ∀ β, α =[n] β → f α = f β definition zω : nat → nat := λ x, zero definition znkω (n : nat) (k : nat) : nat → nat := λ x, if x < n then 0 else k lemma znkω_succ (n : nat) (k : nat) : znkω (n+1) k 0 = 0 := rfl lemma znkω_bound (n : nat) (k : nat) : znkω n k n = k := if_neg !lt.irrefl lemma zω_eq_znkω (n : nat) (k : nat) : zω =[n] znkω n k := λ a altn, begin esimp [zω, znkω], rewrite [if_pos altn] end section hypothesis all_continuous : ∀ f, continuous f definition M (f : (nat → nat) → nat) : nat := (all_continuous f zω).1 lemma M_spec (f : (nat → nat) → nat) : ∀ β, zω =[M f] β → f zω = f β := (all_continuous f zω).2 definition m := M (λα, zero) definition f β := M (λα, β (α m)) lemma β0_eq (β : nat → nat) : ∀ α, zω =[f β] α → β 0 = β (α m) := λ α, M_spec (λα, β (α m)) α lemma not_all_continuous : false := let β := znkω (M f + 1) 1 in let α := znkω m (M f + 1) in assert βeq₁ : zω =[M f + 1] β, from λ (a : nat) (h : a < M f + 1), begin esimp [zω, znkω], rewrite [if_pos h] end, assert βeq₂ : zω =[M f] β, from pred_beq βeq₁, assert m_eq_fβ : m = f β, from M_spec f β βeq₂, assert aux : ∀ α, zω =[m] α → β 0 = β (α m), by rewrite m_eq_fβ at {1}; exact (β0_eq β), assert zero_eq_one : 0 = 1, from calc 0 = β 0 : by rewrite znkω_succ ... = β (α m) : aux α (zω_eq_znkω m (M f + 1)) ... = β (M f + 1) : by rewrite znkω_bound ... = 1 : by rewrite znkω_bound, nat.no_confusion zero_eq_one end /- Additional remarks: By using the slightly different definition of continuous ∀ α, ∃ n, ∀ β, α =[n] β → f α = f β i.e., using ∃ instead of Σ, we can assume the following axiom all_continuous : ∀ f, continuous f However, the system becomes inconsistent again if we also assume Hilbert's choice, because with Hilbert's choice we can convert ∃ into Σ -/