Library Coq.Logic.Diaconescu
Diaconescu showed that the Axiom of Choice entails Excluded-Middle
in topoi [Diaconescu75]. Lacas and Werner adapted the proof to show
that the axiom of choice in equivalence classes entails
Excluded-Middle in Type Theory [LacasWerner99].
Three variants of Diaconescu's result in type theory are shown below.
A. A proof that the relational form of the Axiom of Choice +
Extensionality for Predicates entails Excluded-Middle (by Hugo
Herbelin)
B. A proof that the relational form of the Axiom of Choice + Proof
Irrelevance entails Excluded-Middle for Equality Statements (by
Benjamin Werner)
C. A proof that extensional Hilbert epsilon's description operator
entails excluded-middle (taken from Bell [Bell93])
See also [Carlström04] for a discussion of the connection between the
Extensional Axiom of Choice and Excluded-Middle
[Diaconescu75] Radu Diaconescu, Axiom of Choice and Complementation,
in Proceedings of AMS, vol 51, pp 176-178, 1975.
[LacasWerner99] Samuel Lacas, Benjamin Werner, Which Choices imply
the excluded middle?, preprint, 1999.
[Bell93] John L. Bell, Hilbert's epsilon operator and classical
logic, Journal of Philosophical Logic, 22: 1-18, 1993
[Carlström04] Jesper Carlström, EM + Ext + AC_int is equivalent
to AC_ext, Mathematical Logic Quaterly, vol 50(3), pp 236-240, 2004.
The axiom of extensionality for predicates
Definition PredicateExtensionality :=
forall P Q:bool -> Prop, (forall b:bool, P b <-> Q b) -> P = Q.
From predicate extensionality we get propositional extensionality
hence proof-irrelevance
Import ClassicalFacts.
Variable pred_extensionality : PredicateExtensionality.
Lemma prop_ext : forall A B:Prop, (A <-> B) -> A = B.
Lemma proof_irrel : forall (A:Prop) (a1 a2:A), a1 = a2.
From proof-irrelevance and relational choice, we get guarded
relational choice
Import ChoiceFacts.
Variable rel_choice : RelationalChoice.
Lemma guarded_rel_choice : GuardedRelationalChoice.
The form of choice we need: there is a functional relation which chooses
an element in any non empty subset of bool
Import Bool.
Lemma AC_bool_subset_to_bool :
exists R : (bool -> Prop) -> bool -> Prop,
(forall P:bool -> Prop,
(exists b : bool, P b) ->
exists b : bool, P b /\ R P b /\ (forall b':bool, R P b' -> b = b')).
The proof of the excluded middle Remark: P could have been in Set or Type
Proof-Irrel. + Rel. Axiom of Choice -> Excl.-Middle for Equality
Section ProofIrrel_RelChoice_imp_EqEM.
Import ChoiceFacts.
Variable rel_choice : RelationalChoice.
Variable proof_irrelevance : forall P:Prop , forall x y:P, x=y.
Let a1 and a2 be two elements in some type A
We build the subset A' of A made of a1 and a2
Definition A' := @sigT A (fun x => x=a1 \/ x=a2).
Definition a1':A'.
Defined.
Definition a2':A'.
Defined.
By proof-irrelevance, projection is a retraction
But from the actual proofs of being in A', we can assert in the
proof-irrelevant world the existence of relevant boolean witnesses
Lemma decide : forall x:A', exists y:bool ,
(projT1 x = a1 /\ y = true ) \/ (projT1 x = a2 /\ y = false).
Thanks to the axiom of choice, the boolean witnesses move from the
propositional world to the relevant world
An alternative more concise proof can be done by directly using
the guarded relational choice
Extensional Hilbert's epsilon description operator -> Excluded-Middle
Local Notation inhabited A := A (only parsing).
Section ExtensionalEpsilon_imp_EM.
Variable epsilon : forall A : Type, inhabited A -> (A -> Prop) -> A.
Hypothesis epsilon_spec :
forall (A:Type) (i:inhabited A) (P:A->Prop),
(exists x, P x) -> P (epsilon A i P).
Hypothesis epsilon_extensionality :
forall (A:Type) (i:inhabited A) (P Q:A->Prop),
(forall a, P a <-> Q a) -> epsilon A i P = epsilon A i Q.
Local Notation eps := (epsilon bool true) (only parsing).
Theorem extensional_epsilon_imp_EM : forall P:Prop, P \/ ~ P.
End ExtensionalEpsilon_imp_EM.