Library Coq.Logic.Berardi

This file formalizes Berardi's paradox which says that in the calculus of constructions, excluded middle (EM) and axiom of choice (AC) imply proof irrelevance (PI). Here, the axiom of choice is not necessary because of the use of inductive types.
@article{Barbanera-Berardi:JFP96,
   author    = {F. Barbanera and S. Berardi},
   title     = {Proof-irrelevance out of Excluded-middle and Choice
                in the Calculus of Constructions},
   journal   = {Journal of Functional Programming},
   year      = {1996},
   volume    = {6},
   number    = {3},
   pages     = {519-525}
}

Set Implicit Arguments.

Section Berardis_paradox.

Excluded middle
Hypothesis EM : forall P:Prop, P \/ ~ P.

Conditional on any proposition.
Definition IFProp (P B:Prop) (e1 e2:P) :=
  match EM B with
  | or_introl _ => e1
  | or_intror _ => e2
  end.

Axiom of choice applied to disjunction. Provable in Coq because of dependent elimination.
Lemma AC_IF :
 forall (P B:Prop) (e1 e2:P) (Q:P -> Prop),
   (B -> Q e1) -> (~ B -> Q e2) -> Q (IFProp B e1 e2).
Proof.
intros P B e1 e2 Q p1 p2.
unfold IFProp in |- *.
case (EM B); assumption.
Qed.

We assume a type with two elements. They play the role of booleans. The main theorem under the current assumptions is that T=F
Variable Bool : Prop.
Variable T : Bool.
Variable F : Bool.

The powerset operator
Definition pow (P:Prop) := P -> Bool.

A piece of theory about retracts
Section Retracts.

Variables A B : Prop.

Record retract : Prop :=
  {i : A -> B; j : B -> A; inv : forall a:A, j (i a) = a}.

Record retract_cond : Prop :=
  {i2 : A -> B; j2 : B -> A; inv2 : retract -> forall a:A, j2 (i2 a) = a}.

The dependent elimination above implies the axiom of choice:
Lemma AC : forall r:retract_cond, retract -> forall a:A, j2 r (i2 r a) = a.
Proof.
intros r.
case r; simpl in |- *.
trivial.
Qed.

End Retracts.

This lemma is basically a commutation of implication and existential quantification: (EX x | A -> P(x)) <=> (A -> EX x | P(x)) which is provable in classical logic ( => is already provable in intuitionnistic logic).

Lemma L1 : forall A B:Prop, retract_cond (pow A) (pow B).
Proof.
intros A B.
destruct (EM (retract (pow A) (pow B))) as [(f0,g0,e) | hf].
  exists f0 g0; trivial.
  exists (fun (x:pow A) (y:B) => F) (fun (x:pow B) (y:A) => F); intros;
    destruct hf; auto.
Qed.

The paradoxical set
Definition U := forall P:Prop, pow P.

Bijection between U and (pow U)
Definition f (u:U) : pow U := u U.

Definition g (h:pow U) : U :=
  fun X => let lX := j2 (L1 X U) in let rU := i2 (L1 U U) in lX (rU h).

We deduce that the powerset of U is a retract of U. This lemma is stated in Berardi's article, but is not used afterwards.
Lemma retract_pow_U_U : retract (pow U) U.
Proof.
exists g f.
intro a.
unfold f, g in |- *; simpl in |- *.
apply AC.
exists (fun x:pow U => x) (fun x:pow U => x).
trivial.
Qed.

Encoding of Russel's paradox

The boolean negation.
Definition Not_b (b:Bool) := IFProp (b = T) F T.

the set of elements not belonging to itself
Definition R : U := g (fun u:U => Not_b (u U u)).

Lemma not_has_fixpoint : R R = Not_b (R R).
Proof.
unfold R at 1 in |- *.
unfold g in |- *.
rewrite AC with (r := L1 U U) (a := fun u:U => Not_b (u U u)).
trivial.
exists (fun x:pow U => x) (fun x:pow U => x); trivial.
Qed.

Theorem classical_proof_irrelevence : T = F.
Proof.
generalize not_has_fixpoint.
unfold Not_b in |- *.
apply AC_IF.
intros is_true is_false.
elim is_true; elim is_false; trivial.

intros not_true is_true.
elim not_true; trivial.
Qed.

End Berardis_paradox.