Theory PER

(*  Title:      HOL/ex/PER.thy
    Author:     Oscar Slotosch and Markus Wenzel, TU Muenchen
*)

section ‹Partial equivalence relations›

theory PER
imports Main
begin

text ‹
  Higher-order quotients are defined over partial equivalence
  relations (PERs) instead of total ones.  We provide axiomatic type
  classes equiv < partial_equiv› and a type constructor
  'a quot› with basic operations.  This development is based
  on:

  Oscar Slotosch: \emph{Higher Order Quotients and their
  Implementation in Isabelle HOL.}  Elsa L. Gunter and Amy Felty,
  editors, Theorem Proving in Higher Order Logics: TPHOLs '97,
  Springer LNCS 1275, 1997.
›


subsection ‹Partial equivalence›

text ‹
  Type class partial_equiv› models partial equivalence
  relations (PERs) using the polymorphic ∼ :: 'a ⇒ 'a ⇒
  bool› relation, which is required to be symmetric and transitive,
  but not necessarily reflexive.
›

class partial_equiv =
  fixes eqv :: "'a  'a  bool"    (infixl "" 50)
  assumes partial_equiv_sym [elim?]: "x  y  y  x"
  assumes partial_equiv_trans [trans]: "x  y  y  z  x  z"

text ‹
  \medskip The domain of a partial equivalence relation is the set of
  reflexive elements.  Due to symmetry and transitivity this
  characterizes exactly those elements that are connected with
  \emph{any} other one.
›

definition
  "domain" :: "'a::partial_equiv set" where
  "domain = {x. x  x}"

lemma domainI [intro]: "x  x  x  domain"
  unfolding domain_def by blast

lemma domainD [dest]: "x  domain  x  x"
  unfolding domain_def by blast

theorem domainI' [elim?]: "x  y  x  domain"
proof
  assume xy: "x  y"
  also from xy have "y  x" ..
  finally show "x  x" .
qed


subsection ‹Equivalence on function spaces›

text ‹
  The ∼› relation is lifted to function spaces.  It is
  important to note that this is \emph{not} the direct product, but a
  structural one corresponding to the congruence property.
›

instantiation "fun" :: (partial_equiv, partial_equiv) partial_equiv
begin

definition "f  g  (x  domain. y  domain. x  y  f x  g y)"

lemma partial_equiv_funI [intro?]:
    "(x y. x  domain  y  domain  x  y  f x  g y)  f  g"
  unfolding eqv_fun_def by blast

lemma partial_equiv_funD [dest?]:
    "f  g  x  domain  y  domain  x  y  f x  g y"
  unfolding eqv_fun_def by blast

text ‹
  The class of partial equivalence relations is closed under function
  spaces (in \emph{both} argument positions).
›

instance proof
  fix f g h :: "'a::partial_equiv  'b::partial_equiv"
  assume fg: "f  g"
  show "g  f"
  proof
    fix x y :: 'a
    assume x: "x  domain" and y: "y  domain"
    assume "x  y" then have "y  x" ..
    with fg y x have "f y  g x" ..
    then show "g x  f y" ..
  qed
  assume gh: "g  h"
  show "f  h"
  proof
    fix x y :: 'a
    assume x: "x  domain" and y: "y  domain" and "x  y"
    with fg have "f x  g y" ..
    also from y have "y  y" ..
    with gh y y have "g y  h y" ..
    finally show "f x  h y" .
  qed
qed

end


subsection ‹Total equivalence›

text ‹
  The class of total equivalence relations on top of PERs.  It
  coincides with the standard notion of equivalence, i.e.\ ∼
  :: 'a ⇒ 'a ⇒ bool› is required to be reflexive, transitive and
  symmetric.
›

class equiv =
  assumes eqv_refl [intro]: "x  x"

text ‹
  On total equivalences all elements are reflexive, and congruence
  holds unconditionally.
›

theorem equiv_domain [intro]: "(x::'a::equiv)  domain"
proof
  show "x  x" ..
qed

theorem equiv_cong [dest?]: "f  g  x  y  f x  g (y::'a::equiv)"
proof -
  assume "f  g"
  moreover have "x  domain" ..
  moreover have "y  domain" ..
  moreover assume "x  y"
  ultimately show ?thesis ..
qed


subsection ‹Quotient types›

text ‹
  The quotient type 'a quot› consists of all
  \emph{equivalence classes} over elements of the base type typ'a.
›

definition "quot = {{x. a  x}| a::'a::partial_equiv. True}"

typedef (overloaded) 'a quot = "quot :: 'a::partial_equiv set set"
  unfolding quot_def by blast

lemma quotI [intro]: "{x. a  x}  quot"
  unfolding quot_def by blast

lemma quotE [elim]: "R  quot  (a. R = {x. a  x}  C)  C"
  unfolding quot_def by blast

text ‹
  \medskip Abstracted equivalence classes are the canonical
  representation of elements of a quotient type.
›

definition eqv_class :: "('a::partial_equiv)  'a quot"  ("_")
  where "a = Abs_quot {x. a  x}"

theorem quot_rep: "a. A = a"
proof (cases A)
  fix R assume R: "A = Abs_quot R"
  assume "R  quot" then have "a. R = {x. a  x}" by blast
  with R have "a. A = Abs_quot {x. a  x}" by blast
  then show ?thesis by (unfold eqv_class_def)
qed

lemma quot_cases [cases type: quot]:
  obtains (rep) a where "A = a"
  using quot_rep by blast


subsection ‹Equality on quotients›

text ‹
  Equality of canonical quotient elements corresponds to the original
  relation as follows.
›

theorem eqv_class_eqI [intro]: "a  b  a = b"
proof -
  assume ab: "a  b"
  have "{x. a  x} = {x. b  x}"
  proof (rule Collect_cong)
    fix x show "a  x  b  x"
    proof
      from ab have "b  a" ..
      also assume "a  x"
      finally show "b  x" .
    next
      note ab
      also assume "b  x"
      finally show "a  x" .
    qed
  qed
  then show ?thesis by (simp only: eqv_class_def)
qed

theorem eqv_class_eqD' [dest?]: "a = b  a  domain  a  b"
proof (unfold eqv_class_def)
  assume "Abs_quot {x. a  x} = Abs_quot {x. b  x}"
  then have "{x. a  x} = {x. b  x}" by (simp only: Abs_quot_inject quotI)
  moreover assume "a  domain" then have "a  a" ..
  ultimately have "a  {x. b  x}" by blast
  then have "b  a" by blast
  then show "a  b" ..
qed

theorem eqv_class_eqD [dest?]: "a = b  a  (b::'a::equiv)"
proof (rule eqv_class_eqD')
  show "a  domain" ..
qed

lemma eqv_class_eq' [simp]: "a  domain  a = b  a  b"
  using eqv_class_eqI eqv_class_eqD' by (blast del: eqv_refl)

lemma eqv_class_eq [simp]: "a = b  a  (b::'a::equiv)"
  using eqv_class_eqI eqv_class_eqD by blast


subsection ‹Picking representing elements›

definition pick :: "'a::partial_equiv quot  'a"
  where "pick A = (SOME a. A = a)"

theorem pick_eqv' [intro?, simp]: "a  domain  pick a  a"
proof (unfold pick_def)
  assume a: "a  domain"
  show "(SOME x. a = x)  a"
  proof (rule someI2)
    show "a = a" ..
    fix x assume "a = x"
    from this and a have "a  x" ..
    then show "x  a" ..
  qed
qed

theorem pick_eqv [intro, simp]: "pick a  (a::'a::equiv)"
proof (rule pick_eqv')
  show "a  domain" ..
qed

theorem pick_inverse: "pick A = (A::'a::equiv quot)"
proof (cases A)
  fix a assume a: "A = a"
  then have "pick A  a" by simp
  then have "pick A = a" by simp
  with a show ?thesis by simp
qed

end