Theory AC_in_L

theory AC_in_L
imports Separation
(*  Title:      ZF/Constructible/AC_in_L.thy
    Author:     Lawrence C Paulson, Cambridge University Computer Laboratory

header {* The Axiom of Choice Holds in L! *}

theory AC_in_L imports Formula Separation begin

subsection{*Extending a Wellordering over a List -- Lexicographic Power*}

text{*This could be moved into a library.*}

  rlist   :: "[i,i]=>i"

  domains "rlist(A,r)"  "list(A) * list(A)"
      "[| length(l') < length(l); l' ∈ list(A); l ∈ list(A) |]
       ==> <l', l> ∈ rlist(A,r)"

      "[| <l',l> ∈ rlist(A,r); a ∈ A |]
       ==> <Cons(a,l'), Cons(a,l)> ∈ rlist(A,r)"

      "[| length(l') = length(l); <a',a> ∈ r;
          l' ∈ list(A); l ∈ list(A); a' ∈ A; a ∈ A |]
       ==> <Cons(a',l'), Cons(a,l)> ∈ rlist(A,r)"
  type_intros list.intros

subsubsection{*Type checking*}

lemmas rlist_type = rlist.dom_subset

lemmas field_rlist = rlist_type [THEN field_rel_subset]


lemma rlist_Nil_Cons [intro]:
    "[|a ∈ A; l ∈ list(A)|] ==> <[], Cons(a,l)> ∈ rlist(A, r)"
by (simp add: shorterI)

lemma linear_rlist:
  assumes r: "linear(A,r)" shows "linear(list(A),rlist(A,r))"
proof -
  { fix xs ys
    have "xs ∈ list(A) ==> ys ∈ list(A) ==> ⟨xs,ys⟩ ∈ rlist(A,r) ∨ xs = ys ∨ ⟨ys,xs⟩ ∈ rlist(A, r) "
    proof (induct xs arbitrary: ys rule: list.induct)
      case Nil 
      thus ?case by (induct ys rule: list.induct) (auto simp add: shorterI)
      case (Cons x xs)
      { fix y ys
        assume "y ∈ A" and "ys ∈ list(A)"
        with Cons
        have "⟨Cons(x,xs),Cons(y,ys)⟩ ∈ rlist(A,r) ∨ x=y & xs = ys ∨ ⟨Cons(y,ys), Cons(x,xs)⟩ ∈ rlist(A,r)" 
          apply (rule_tac i = "length(xs)" and j = "length(ys)" in Ord_linear_lt)
          apply (simp_all add: shorterI)
          apply (rule linearE [OF r, of x y]) 
          apply (auto simp add: diffI intro: sameI) 
      note yConsCase = this
      show ?case using `ys ∈ list(A)`
        by (cases rule: list.cases) (simp_all add: Cons rlist_Nil_Cons yConsCase) 
  thus ?thesis by (simp add: linear_def) 


text{*Nothing preceeds Nil in this ordering.*}
inductive_cases rlist_NilE: " <l,[]> ∈ rlist(A,r)"

inductive_cases rlist_ConsE: " <l', Cons(x,l)> ∈ rlist(A,r)"

lemma not_rlist_Nil [simp]: " <l,[]> ∉ rlist(A,r)"
by (blast intro: elim: rlist_NilE)

lemma rlist_imp_length_le: "<l',l> ∈ rlist(A,r) ==> length(l') ≤ length(l)"
apply (erule rlist.induct)
apply (simp_all add: leI)

lemma wf_on_rlist_n:
  "[| n ∈ nat; wf[A](r) |] ==> wf[{l ∈ list(A). length(l) = n}](rlist(A,r))"
apply (induct_tac n)
 apply (rule wf_onI2, simp)
apply (rule wf_onI2, clarify)
apply (erule_tac a=y in list.cases, clarify)
 apply (simp (no_asm_use))
apply clarify
apply (simp (no_asm_use))
apply (subgoal_tac "∀l2 ∈ list(A). length(l2) = x --> Cons(a,l2) ∈ B", blast)
apply (erule_tac a=a in wf_on_induct, assumption)
apply (rule ballI)
apply (rule impI)
apply (erule_tac a=l2 in wf_on_induct, blast, clarify)
apply (rename_tac a' l2 l')
apply (drule_tac x="Cons(a',l')" in bspec, typecheck)
apply simp
apply (erule mp, clarify)
apply (erule rlist_ConsE, auto)

lemma list_eq_UN_length: "list(A) = (\<Union>n∈nat. {l ∈ list(A). length(l) = n})"
by (blast intro: length_type)

lemma wf_on_rlist: "wf[A](r) ==> wf[list(A)](rlist(A,r))"
apply (subst list_eq_UN_length)
apply (rule wf_on_Union)
  apply (rule wf_imp_wf_on [OF wf_Memrel [of nat]])
 apply (simp add: wf_on_rlist_n)
apply (frule rlist_type [THEN subsetD])
apply (simp add: length_type)
apply (drule rlist_imp_length_le)
apply (erule leE)
apply (simp_all add: lt_def)

lemma wf_rlist: "wf(r) ==> wf(rlist(field(r),r))"
apply (simp add: wf_iff_wf_on_field)
apply (rule wf_on_subset_A [OF _ field_rlist])
apply (blast intro: wf_on_rlist)

lemma well_ord_rlist:
     "well_ord(A,r) ==> well_ord(list(A), rlist(A,r))"
apply (rule well_ordI)
apply (simp add: well_ord_def wf_on_rlist)
apply (simp add: well_ord_def tot_ord_def linear_rlist)

subsection{*An Injection from Formulas into the Natural Numbers*}

text{*There is a well-known bijection between @{term "nat*nat"} and @{term
nat} given by the expression f(m,n) = triangle(m+n) + m, where triangle(k)
enumerates the triangular numbers and can be defined by triangle(0)=0,
triangle(succ(k)) = succ(k + triangle(k)).  Some small amount of effort is
needed to show that f is a bijection.  We already know that such a bijection exists by the theorem @{text well_ord_InfCard_square_eq}:
@{thm[display] well_ord_InfCard_square_eq[no_vars]}

However, this result merely states that there is a bijection between the two
sets.  It provides no means of naming a specific bijection.  Therefore, we
conduct the proofs under the assumption that a bijection exists.  The simplest
way to organize this is to use a locale.*}

text{*Locale for any arbitrary injection between @{term "nat*nat"}
      and @{term nat}*}
locale Nat_Times_Nat =
  fixes fn
  assumes fn_inj: "fn ∈ inj(nat*nat, nat)"

consts   enum :: "[i,i]=>i"
  "enum(f, Member(x,y)) = f ` <0, f ` <x,y>>"
  "enum(f, Equal(x,y)) = f ` <1, f ` <x,y>>"
  "enum(f, Nand(p,q)) = f ` <2, f ` <enum(f,p), enum(f,q)>>"
  "enum(f, Forall(p)) = f ` <succ(2), enum(f,p)>"

lemma (in Nat_Times_Nat) fn_type [TC,simp]:
    "[|x ∈ nat; y ∈ nat|] ==> fn`<x,y> ∈ nat"
by (blast intro: inj_is_fun [OF fn_inj] apply_funtype)

lemma (in Nat_Times_Nat) fn_iff:
    "[|x ∈ nat; y ∈ nat; u ∈ nat; v ∈ nat|]
     ==> (fn`<x,y> = fn`<u,v>) <-> (x=u & y=v)"
by (blast dest: inj_apply_equality [OF fn_inj])

lemma (in Nat_Times_Nat) enum_type [TC,simp]:
    "p ∈ formula ==> enum(fn,p) ∈ nat"
by (induct_tac p, simp_all)

lemma (in Nat_Times_Nat) enum_inject [rule_format]:
    "p ∈ formula ==> ∀q∈formula. enum(fn,p) = enum(fn,q) --> p=q"
apply (induct_tac p, simp_all)
   apply (rule ballI)
   apply (erule formula.cases)
   apply (simp_all add: fn_iff)
  apply (rule ballI)
  apply (erule formula.cases)
  apply (simp_all add: fn_iff)
 apply (rule ballI)
 apply (erule_tac a=qa in formula.cases)
 apply (simp_all add: fn_iff)
 apply blast
apply (rule ballI)
apply (erule_tac a=q in formula.cases)
apply (simp_all add: fn_iff, blast)

lemma (in Nat_Times_Nat) inj_formula_nat:
    "(λp ∈ formula. enum(fn,p)) ∈ inj(formula, nat)"
apply (simp add: inj_def lam_type)
apply (blast intro: enum_inject)

lemma (in Nat_Times_Nat) well_ord_formula:
    "well_ord(formula, measure(formula, enum(fn)))"
apply (rule well_ord_measure, simp)
apply (blast intro: enum_inject)

lemmas nat_times_nat_lepoll_nat =
    InfCard_nat [THEN InfCard_square_eqpoll, THEN eqpoll_imp_lepoll]

text{*Not needed--but interesting?*}
theorem formula_lepoll_nat: "formula \<lesssim> nat"
apply (insert nat_times_nat_lepoll_nat)
apply (unfold lepoll_def)
apply (blast intro: Nat_Times_Nat.inj_formula_nat Nat_Times_Nat.intro)

subsection{*Defining the Wellordering on @{term "DPow(A)"}*}

text{*The objective is to build a wellordering on @{term "DPow(A)"} from a
given one on @{term A}.  We first introduce wellorderings for environments,
which are lists built over @{term "A"}.  We combine it with the enumeration of
formulas.  The order type of the resulting wellordering gives us a map from
(environment, formula) pairs into the ordinals.  For each member of @{term
"DPow(A)"}, we take the minimum such ordinal.*}

  env_form_r :: "[i,i,i]=>i" where
    --{*wellordering on (environment, formula) pairs*}
   "env_form_r(f,r,A) ==
      rmult(list(A), rlist(A, r),
            formula, measure(formula, enum(f)))"

  env_form_map :: "[i,i,i,i]=>i" where
    --{*map from (environment, formula) pairs to ordinals*}
      == ordermap(list(A) * formula, env_form_r(f,r,A)) ` z"

  DPow_ord :: "[i,i,i,i,i]=>o" where
    --{*predicate that holds if @{term k} is a valid index for @{term X}*}
   "DPow_ord(f,r,A,X,k) ==
           ∃env ∈ list(A). ∃p ∈ formula.
             arity(p) ≤ succ(length(env)) &
             X = {x∈A. sats(A, p, Cons(x,env))} &
             env_form_map(f,r,A,<env,p>) = k"

  DPow_least :: "[i,i,i,i]=>i" where
    --{*function yielding the smallest index for @{term X}*}
   "DPow_least(f,r,A,X) == μ k. DPow_ord(f,r,A,X,k)"

  DPow_r :: "[i,i,i]=>i" where
    --{*a wellordering on @{term "DPow(A)"}*}
   "DPow_r(f,r,A) == measure(DPow(A), DPow_least(f,r,A))"

lemma (in Nat_Times_Nat) well_ord_env_form_r:
     ==> well_ord(list(A) * formula, env_form_r(fn,r,A))"
by (simp add: env_form_r_def well_ord_rmult well_ord_rlist well_ord_formula)

lemma (in Nat_Times_Nat) Ord_env_form_map:
    "[|well_ord(A,r); z ∈ list(A) * formula|]
     ==> Ord(env_form_map(fn,r,A,z))"
by (simp add: env_form_map_def Ord_ordermap well_ord_env_form_r)

lemma DPow_imp_ex_DPow_ord:
    "X ∈ DPow(A) ==> ∃k. DPow_ord(fn,r,A,X,k)"
apply (simp add: DPow_ord_def)
apply (blast dest!: DPowD)

lemma (in Nat_Times_Nat) DPow_ord_imp_Ord:
     "[|DPow_ord(fn,r,A,X,k); well_ord(A,r)|] ==> Ord(k)"
apply (simp add: DPow_ord_def, clarify)
apply (simp add: Ord_env_form_map)

lemma (in Nat_Times_Nat) DPow_imp_DPow_least:
    "[|X ∈ DPow(A); well_ord(A,r)|]
     ==> DPow_ord(fn, r, A, X, DPow_least(fn,r,A,X))"
apply (simp add: DPow_least_def)
apply (blast dest: DPow_imp_ex_DPow_ord intro: DPow_ord_imp_Ord LeastI)

lemma (in Nat_Times_Nat) env_form_map_inject:
    "[|env_form_map(fn,r,A,u) = env_form_map(fn,r,A,v); well_ord(A,r);
       u ∈ list(A) * formula;  v ∈ list(A) * formula|]
     ==> u=v"
apply (simp add: env_form_map_def)
apply (rule inj_apply_equality [OF bij_is_inj, OF ordermap_bij,
                                OF well_ord_env_form_r], assumption+)

lemma (in Nat_Times_Nat) DPow_ord_unique:
    "[|DPow_ord(fn,r,A,X,k); DPow_ord(fn,r,A,Y,k); well_ord(A,r)|]
     ==> X=Y"
apply (simp add: DPow_ord_def, clarify)
apply (drule env_form_map_inject, auto)

lemma (in Nat_Times_Nat) well_ord_DPow_r:
    "well_ord(A,r) ==> well_ord(DPow(A), DPow_r(fn,r,A))"
apply (simp add: DPow_r_def)
apply (rule well_ord_measure)
 apply (simp add: DPow_least_def Ord_Least)
apply (drule DPow_imp_DPow_least, assumption)+
apply simp
apply (blast intro: DPow_ord_unique)

lemma (in Nat_Times_Nat) DPow_r_type:
    "DPow_r(fn,r,A) ⊆ DPow(A) * DPow(A)"
by (simp add: DPow_r_def measure_def, blast)

subsection{*Limit Construction for Well-Orderings*}

text{*Now we work towards the transfinite definition of wellorderings for
@{term "Lset(i)"}.  We assume as an inductive hypothesis that there is a family
of wellorderings for smaller ordinals.*}

  rlimit :: "[i,i=>i]=>i" where
  --{*Expresses the wellordering at limit ordinals.  The conditional
      lets us remove the premise @{term "Limit(i)"} from some theorems.*}
    "rlimit(i,r) ==
       if Limit(i) then 
         {z: Lset(i) * Lset(i).
          ∃x' x. z = <x',x> &
                 (lrank(x') < lrank(x) |
                  (lrank(x') = lrank(x) & <x',x> ∈ r(succ(lrank(x)))))}
       else 0"

  Lset_new :: "i=>i" where
  --{*This constant denotes the set of elements introduced at level
      @{term "succ(i)"}*}
    "Lset_new(i) == {x ∈ Lset(succ(i)). lrank(x) = i}"

lemma Limit_Lset_eq2:
    "Limit(i) ==> Lset(i) = (\<Union>j∈i. Lset_new(j))"
apply (simp add: Limit_Lset_eq)
apply (rule equalityI)
 apply safe
 apply (subgoal_tac "Ord(y)")
  prefer 2 apply (blast intro: Ord_in_Ord Limit_is_Ord)
 apply (simp_all add: Limit_is_Ord Lset_iff_lrank_lt Lset_new_def
 apply (blast intro: lt_trans)
apply (rule_tac x = "succ(lrank(x))" in bexI)
 apply (simp add: Lset_succ_lrank_iff)
apply (blast intro: Limit_has_succ ltD)

lemma wf_on_Lset:
    "wf[Lset(succ(j))](r(succ(j))) ==> wf[Lset_new(j)](rlimit(i,r))"
apply (simp add: wf_on_def Lset_new_def)
apply (erule wf_subset)
apply (simp add: rlimit_def, force)

lemma wf_on_rlimit:
    "(∀j<i. wf[Lset(j)](r(j))) ==> wf[Lset(i)](rlimit(i,r))"
apply (case_tac "Limit(i)") 
 prefer 2
 apply (simp add: rlimit_def wf_on_any_0)
apply (simp add: Limit_Lset_eq2)
apply (rule wf_on_Union)
  apply (rule wf_imp_wf_on [OF wf_Memrel [of i]])
 apply (blast intro: wf_on_Lset Limit_has_succ Limit_is_Ord ltI)
apply (force simp add: rlimit_def Limit_is_Ord Lset_iff_lrank_lt Lset_new_def

lemma linear_rlimit:
    "[|Limit(i); ∀j<i. linear(Lset(j), r(j)) |]
     ==> linear(Lset(i), rlimit(i,r))"
apply (frule Limit_is_Ord)
apply (simp add: Limit_Lset_eq2 Lset_new_def)
apply (simp add: linear_def rlimit_def Ball_def lt_Ord Lset_iff_lrank_lt)
apply (simp add: ltI, clarify)
apply (rename_tac u v)
apply (rule_tac i="lrank(u)" and j="lrank(v)" in Ord_linear_lt, simp_all) 
apply (drule_tac x="succ(lrank(u) ∪ lrank(v))" in ospec)
 apply (simp add: ltI)
apply (drule_tac x=u in spec, simp)
apply (drule_tac x=v in spec, simp)

lemma well_ord_rlimit:
    "[|Limit(i); ∀j<i. well_ord(Lset(j), r(j)) |]
     ==> well_ord(Lset(i), rlimit(i,r))"
by (blast intro: well_ordI wf_on_rlimit well_ord_is_wf
                           linear_rlimit well_ord_is_linear)

lemma rlimit_cong:
     "(!!j. j<i ==> r'(j) = r(j)) ==> rlimit(i,r) = rlimit(i,r')"
apply (simp add: rlimit_def, clarify) 
apply (rule refl iff_refl Collect_cong ex_cong conj_cong)+
apply (simp add: Limit_is_Ord Lset_lrank_lt)

subsection{*Transfinite Definition of the Wellordering on @{term "L"}*}

  L_r :: "[i, i] => i" where
  "L_r(f) == %i.
      transrec3(i, 0, λx r. DPow_r(f, r, Lset(x)), 
                λx r. rlimit(x, λy. r`y))"

subsubsection{*The Corresponding Recursion Equations*}
lemma [simp]: "L_r(f,0) = 0"
by (simp add: L_r_def)

lemma [simp]: "L_r(f, succ(i)) = DPow_r(f, L_r(f,i), Lset(i))"
by (simp add: L_r_def)

text{*The limit case is non-trivial because of the distinction between
object-level and meta-level abstraction.*}
lemma [simp]: "Limit(i) ==> L_r(f,i) = rlimit(i, L_r(f))"
by (simp cong: rlimit_cong add: transrec3_Limit L_r_def ltD)

lemma (in Nat_Times_Nat) L_r_type:
    "Ord(i) ==> L_r(fn,i) ⊆ Lset(i) * Lset(i)"
apply (induct i rule: trans_induct3)
  apply (simp_all add: Lset_succ DPow_r_type well_ord_DPow_r rlimit_def
                       Transset_subset_DPow [OF Transset_Lset], blast)

lemma (in Nat_Times_Nat) well_ord_L_r:
    "Ord(i) ==> well_ord(Lset(i), L_r(fn,i))"
apply (induct i rule: trans_induct3)
apply (simp_all add: well_ord0 Lset_succ L_r_type well_ord_DPow_r
                     well_ord_rlimit ltD)

lemma well_ord_L_r:
    "Ord(i) ==> ∃r. well_ord(Lset(i), r)"
apply (insert nat_times_nat_lepoll_nat)
apply (unfold lepoll_def)
apply (blast intro: Nat_Times_Nat.well_ord_L_r Nat_Times_Nat.intro)

text{*Every constructible set is well-ordered! Therefore the Wellordering Theorem and
      the Axiom of Choice hold in @{term L}!!*}
theorem L_implies_AC: assumes x: "L(x)" shows "∃r. well_ord(x,r)"
  using Transset_Lset x
apply (simp add: Transset_def L_def)
apply (blast dest!: well_ord_L_r intro: well_ord_subset)

interpretation L?: M_basic L by (rule M_basic_L)

theorem "∀x[L]. ∃r. wellordered(L,x,r)"
  fix x
  assume "L(x)"
  then obtain r where "well_ord(x,r)" 
    by (blast dest: L_implies_AC) 
  thus "∃r. wellordered(L,x,r)" 
    by (blast intro: well_ord_imp_relativized)

text{*In order to prove @{term" ∃r[L]. wellordered(L,x,r)"}, it's necessary to know 
that @{term r} is actually constructible. It follows from the assumption ``@{term V} equals @{term L''}, 
but this reasoning doesn't appear to work in Isabelle.*}