(*  Title:      HOL/Proofs/Extraction/Higman.thy

    Author:     Stefan Berghofer, TU Muenchen

    Author:     Monika Seisenberger, LMU Muenchen

*)

header {* Higman's lemma *}

theory Higman2

imports Closures

begin

text {*

  Formalization by Stefan Berghofer and Monika Seisenberger,

  based on Coquand and Fridlender \cite{Coquand93}.

*}

datatype letter = A | B

inductive emb :: "letter list \<Rightarrow> letter list \<Rightarrow> bool"

where

   emb0 [Pure.intro]: "emb [] bs"

 | emb1 [Pure.intro]: "emb as bs \<Longrightarrow> emb as (b # bs)"

 | emb2 [Pure.intro]: "emb as bs \<Longrightarrow> emb (a # as) (a # bs)"

inductive L :: "letter list \<Rightarrow> letter list list \<Rightarrow> bool"

  for v :: "letter list"

where

   L0 [Pure.intro]: "emb w v \<Longrightarrow> L v (w # ws)"

 | L1 [Pure.intro]: "L v ws \<Longrightarrow> L v (w # ws)"

inductive good :: "letter list list \<Rightarrow> bool"

where

    good0 [Pure.intro]: "L w ws \<Longrightarrow> good (w # ws)"

  | good1 [Pure.intro]: "good ws \<Longrightarrow> good (w # ws)"

inductive R :: "letter \<Rightarrow> letter list list \<Rightarrow> letter list list \<Rightarrow> bool"

  for a :: letter

where

    R0 [Pure.intro]: "R a [] []"

  | R1 [Pure.intro]: "R a vs ws \<Longrightarrow> R a (w # vs) ((a # w) # ws)"

inductive T :: "letter \<Rightarrow> letter list list \<Rightarrow> letter list list \<Rightarrow> bool"

  for a :: letter

where

    T0 [Pure.intro]: "a \<noteq> b \<Longrightarrow> R b ws zs \<Longrightarrow> T a (w # zs) ((a # w) # zs)"

  | T1 [Pure.intro]: "T a ws zs \<Longrightarrow> T a (w # ws) ((a # w) # zs)"

  | T2 [Pure.intro]: "a \<noteq> b \<Longrightarrow> T a ws zs \<Longrightarrow> T a ws ((b # w) # zs)"

inductive bar :: "letter list list \<Rightarrow> bool"

where

    bar1 [Pure.intro]: "good ws \<Longrightarrow> bar ws"

  | bar2 [Pure.intro]: "(\<And>w. bar (w # ws)) \<Longrightarrow> bar ws"

theorem prop1: "bar ([] # ws)" by iprover

theorem lemma1: "L as ws \<Longrightarrow> L (a # as) ws"

  by (erule L.induct, iprover+)

lemma lemma2': "R a vs ws \<Longrightarrow> L as vs \<Longrightarrow> L (a # as) ws"

  apply (induct set: R)

  apply (erule L.cases)

  apply simp+

  apply (erule L.cases)

  apply simp_all

  apply (rule L0)

  apply (erule emb2)

  apply (erule L1)

  done

lemma lemma2: "R a vs ws \<Longrightarrow> good vs \<Longrightarrow> good ws"

  apply (induct set: R)

  apply iprover

  apply (erule good.cases)

  apply simp_all

  apply (rule good0)

  apply (erule lemma2')

  apply assumption

  apply (erule good1)

  done

lemma lemma3': "T a vs ws \<Longrightarrow> L as vs \<Longrightarrow> L (a # as) ws"

  apply (induct set: T)

  apply (erule L.cases)

  apply simp_all

  apply (rule L0)

  apply (erule emb2)

  apply (rule L1)

  apply (erule lemma1)

  apply (erule L.cases)

  apply simp_all

  apply iprover+

  done

lemma lemma3: "T a ws zs \<Longrightarrow> good ws \<Longrightarrow> good zs"

  apply (induct set: T)

  apply (erule good.cases)

  apply simp_all

  apply (rule good0)

  apply (erule lemma1)

  apply (erule good1)

  apply (erule good.cases)

  apply simp_all

  apply (rule good0)

  apply (erule lemma3')

  apply iprover+

  done

lemma lemma4: "R a ws zs \<Longrightarrow> ws \<noteq> [] \<Longrightarrow> T a ws zs"

  apply (induct set: R)

  apply iprover

  apply (case_tac vs)

  apply (erule R.cases)

  apply simp

  apply (case_tac a)

  apply (rule_tac b=B in T0)

  apply simp

  apply (rule R0)

  apply (rule_tac b=A in T0)

  apply simp

  apply (rule R0)

  apply simp

  apply (rule T1)

  apply simp

  done

lemma letter_neq: "(a::letter) \<noteq> b \<Longrightarrow> c \<noteq> a \<Longrightarrow> c = b"

  apply (case_tac a)

  apply (case_tac b)

  apply (case_tac c, simp, simp)

  apply (case_tac c, simp, simp)

  apply (case_tac b)

  apply (case_tac c, simp, simp)

  apply (case_tac c, simp, simp)

  done

lemma letter_eq_dec: "(a::letter) = b \<or> a \<noteq> b"

  apply (case_tac a)

  apply (case_tac b)

  apply simp

  apply simp

  apply (case_tac b)

  apply simp

  apply simp

  done

theorem prop2:

  assumes ab: "a \<noteq> b" and bar: "bar xs"

  shows "\<And>ys zs. bar ys \<Longrightarrow> T a xs zs \<Longrightarrow> T b ys zs \<Longrightarrow> bar zs" using bar

proof induct

  fix xs zs assume "T a xs zs" and "good xs"

  hence "good zs" by (rule lemma3)

  then show "bar zs" by (rule bar1)

next

  fix xs ys

  assume I: "\<And>w ys zs. bar ys \<Longrightarrow> T a (w # xs) zs \<Longrightarrow> T b ys zs \<Longrightarrow> bar zs"

  assume "bar ys"

  thus "\<And>zs. T a xs zs \<Longrightarrow> T b ys zs \<Longrightarrow> bar zs"

  proof induct

    fix ys zs assume "T b ys zs" and "good ys"

    then have "good zs" by (rule lemma3)

    then show "bar zs" by (rule bar1)

  next

    fix ys zs assume I': "\<And>w zs. T a xs zs \<Longrightarrow> T b (w # ys) zs \<Longrightarrow> bar zs"

    and ys: "\<And>w. bar (w # ys)" and Ta: "T a xs zs" and Tb: "T b ys zs"

    show "bar zs"

    proof (rule bar2)

      fix w

      show "bar (w # zs)"

      proof (cases w)

        case Nil

        thus ?thesis by simp (rule prop1)

      next

        case (Cons c cs)

        from letter_eq_dec show ?thesis

        proof

          assume ca: "c = a"

          from ab have "bar ((a # cs) # zs)" by (iprover intro: I ys Ta Tb)

          thus ?thesis by (simp add: Cons ca)

        next

          assume "c \<noteq> a"

          with ab have cb: "c = b" by (rule letter_neq)

          from ab have "bar ((b # cs) # zs)" by (iprover intro: I' Ta Tb)

          thus ?thesis by (simp add: Cons cb)

        qed

      qed

    qed

  qed

qed

theorem prop3:

  assumes bar: "bar xs"

  shows "\<And>zs. xs \<noteq> [] \<Longrightarrow> R a xs zs \<Longrightarrow> bar zs" using bar

proof induct

  fix xs zs

  assume "R a xs zs" and "good xs"

  then have "good zs" by (rule lemma2)

  then show "bar zs" by (rule bar1)

next

  fix xs zs

  assume I: "\<And>w zs. w # xs \<noteq> [] \<Longrightarrow> R a (w # xs) zs \<Longrightarrow> bar zs"

  and xsb: "\<And>w. bar (w # xs)" and xsn: "xs \<noteq> []" and R: "R a xs zs"

  show "bar zs"

  proof (rule bar2)

    fix w

    show "bar (w # zs)"

    proof (induct w)

      case Nil

      show ?case by (rule prop1)

    next

      case (Cons c cs)

      from letter_eq_dec show ?case

      proof

        assume "c = a"

        thus ?thesis by (iprover intro: I [simplified] R)

      next

        from R xsn have T: "T a xs zs" by (rule lemma4)

        assume "c \<noteq> a"

        thus ?thesis by (iprover intro: prop2 Cons xsb xsn R T)

      qed

    qed

  qed

qed

theorem higman: "bar []"

proof (rule bar2)

  fix w

  show "bar [w]"

  proof (induct w)

    show "bar [[]]" by (rule prop1)

  next

    fix c cs assume "bar [cs]"

    thus "bar [c # cs]" by (rule prop3) (simp, iprover)

  qed

qed

inductive substring ("_ \<preceq> _")

where

  "[] \<preceq> y"

| "x \<preceq> y \<Longrightarrow> c # x \<preceq> y"

| "x \<preceq> y \<Longrightarrow> c # x \<preceq> c # y" 

lemma substring_refl:

  "x \<preceq> x"

apply(induct x)

apply(auto intro: substring.intros)

done

definition

 "SUBSEQ C \<equiv> {x. \<exists>y \<in> C. x \<preceq> y}"

lemma

 "SUBSEQ (SUBSEQ C) = SUBSEQ C"

unfolding SUBSEQ_def

apply(auto)

apply(erule substring.induct)

apply(rule_tac x="xb" in bexI)

apply(rule substring.intros)

apply(simp)

apply(erule bexE)

apply(rule_tac x="ya" in bexI)

apply(rule substring.intros)

apply(auto)[2]

apply(erule bexE)

apply(rule_tac x="ya" in bexI)

apply(rule substring.intros)

apply(auto)[2]

apply(rule_tac x="x" in exI)

apply(rule conjI)

apply(rule_tac x="y" in bexI)

apply(auto)[2]

apply(rule substring_refl)

done

lemma

  "x \<in> SUBSEQ C \<and> y \<preceq> x \<Longrightarrow> y \<in> SUBSEQ C"

unfolding SUBSEQ_def

apply(auto)

definition

 "CLOSED C \<equiv> C = SUBSEQ C"

primrec

  is_prefix :: "'a list \<Rightarrow> (nat \<Rightarrow> 'a) \<Rightarrow> bool"

where

    "is_prefix [] f = True"

  | "is_prefix (x # xs) f = (x = f (length xs) \<and> is_prefix xs f)"

theorem L_idx:

  assumes L: "L w ws"

  shows "is_prefix ws f \<Longrightarrow> \<exists>i. emb (f i) w \<and> i < length ws" using L

proof induct

  case (L0 v ws)

  hence "emb (f (length ws)) w" by simp

  moreover have "length ws < length (v # ws)" by simp

  ultimately show ?case by iprover

next

  case (L1 ws v)

  then obtain i where emb: "emb (f i) w" and "i < length ws"

    by simp iprover

  hence "i < length (v # ws)" by simp

  with emb show ?case by iprover

qed

theorem good_idx:

  assumes good: "good ws"

  shows "is_prefix ws f \<Longrightarrow> \<exists>i j. emb (f i) (f j) \<and> i < j" using good

proof induct

  case (good0 w ws)

  hence "w = f (length ws)" and "is_prefix ws f" by simp_all

  with good0 show ?case by (iprover dest: L_idx)

next

  case (good1 ws w)

  thus ?case by simp

qed

theorem bar_idx:

  assumes bar: "bar ws"

  shows "is_prefix ws f \<Longrightarrow> \<exists>i j. emb (f i) (f j) \<and> i < j" using bar

proof induct

  case (bar1 ws)

  thus ?case by (rule good_idx)

next

  case (bar2 ws)

  hence "is_prefix (f (length ws) # ws) f" by simp

  thus ?case by (rule bar2)

qed

text {*

Strong version: yields indices of words that can be embedded into each other.

*}

theorem higman_idx: "\<exists>(i::nat) j. emb (f i) (f j) \<and> i < j"

proof (rule bar_idx)

  show "bar []" by (rule higman)

  show "is_prefix [] f" by simp

qed

text {*

Weak version: only yield sequence containing words

that can be embedded into each other.

*}

theorem good_prefix_lemma:

  assumes bar: "bar ws"

  shows "is_prefix ws f \<Longrightarrow> \<exists>vs. is_prefix vs f \<and> good vs" using bar

proof induct

  case bar1

  thus ?case by iprover

next

  case (bar2 ws)

  from bar2.prems have "is_prefix (f (length ws) # ws) f" by simp

  thus ?case by (iprover intro: bar2)

qed

theorem good_prefix: "\<exists>vs. is_prefix vs f \<and> good vs"

  using higman

  by (rule good_prefix_lemma) simp+

subsection {* Extracting the program *}

declare R.induct [ind_realizer]

declare T.induct [ind_realizer]

declare L.induct [ind_realizer]

declare good.induct [ind_realizer]

declare bar.induct [ind_realizer]

extract higman_idx

text {*

  Program extracted from the proof of @{text higman_idx}:

  @{thm [display] higman_idx_def [no_vars]}

  Corresponding correctness theorem:

  @{thm [display] higman_idx_correctness [no_vars]}

  Program extracted from the proof of @{text higman}:

  @{thm [display] higman_def [no_vars]}

  Program extracted from the proof of @{text prop1}:

  @{thm [display] prop1_def [no_vars]}

  Program extracted from the proof of @{text prop2}:

  @{thm [display] prop2_def [no_vars]}

  Program extracted from the proof of @{text prop3}:

  @{thm [display] prop3_def [no_vars]}

*}

subsection {* Some examples *}

instantiation LT and TT :: default

begin

definition "default = L0 [] []"

definition "default = T0 A [] [] [] R0"

instance ..

end

function mk_word_aux :: "nat \<Rightarrow> Random.seed \<Rightarrow> letter list \<times> Random.seed" where

  "mk_word_aux k = exec {

     i \<leftarrow> Random.range 10;

     (if i > 7 \<and> k > 2 \<or> k > 1000 then Pair []

     else exec {

       let l = (if i mod 2 = 0 then A else B);

       ls \<leftarrow> mk_word_aux (Suc k);

       Pair (l # ls)

     })}"

by pat_completeness auto

termination by (relation "measure ((op -) 1001)") auto

definition mk_word :: "Random.seed \<Rightarrow> letter list \<times> Random.seed" where

  "mk_word = mk_word_aux 0"

primrec mk_word_s :: "nat \<Rightarrow> Random.seed \<Rightarrow> letter list \<times> Random.seed" where

  "mk_word_s 0 = mk_word"

  | "mk_word_s (Suc n) = exec {

       _ \<leftarrow> mk_word;

       mk_word_s n

     }"

definition g1 :: "nat \<Rightarrow> letter list" where

  "g1 s = fst (mk_word_s s (20000, 1))"

definition g2 :: "nat \<Rightarrow> letter list" where

  "g2 s = fst (mk_word_s s (50000, 1))"

fun f1 :: "nat \<Rightarrow> letter list" where

  "f1 0 = [A, A]"

  | "f1 (Suc 0) = [B]"

  | "f1 (Suc (Suc 0)) = [A, B]"

  | "f1 _ = []"

fun f2 :: "nat \<Rightarrow> letter list" where

  "f2 0 = [A, A]"

  | "f2 (Suc 0) = [B]"

  | "f2 (Suc (Suc 0)) = [B, A]"

  | "f2 _ = []"

ML {*

local

  val higman_idx = @{code higman_idx};

  val g1 = @{code g1};

  val g2 = @{code g2};

  val f1 = @{code f1};

  val f2 = @{code f2};

in

  val (i1, j1) = higman_idx g1;

  val (v1, w1) = (g1 i1, g1 j1);

  val (i2, j2) = higman_idx g2;

  val (v2, w2) = (g2 i2, g2 j2);

  val (i3, j3) = higman_idx f1;

  val (v3, w3) = (f1 i3, f1 j3);

  val (i4, j4) = higman_idx f2;

  val (v4, w4) = (f2 i4, f2 j4);

end;

*}

text {* The same story with the legacy SML code generator,

this can be removed once the code generator is removed. *}

code_module Higman

contains

  higman = higman_idx

ML {*

local open Higman in

val a = 16807.0;

val m = 2147483647.0;

fun nextRand seed =

  let val t = a*seed

  in  t - m * real (Real.floor(t/m)) end;

fun mk_word seed l =

  let

    val r = nextRand seed;

    val i = Real.round (r / m * 10.0);

  in if i > 7 andalso l > 2 then (r, []) else

    apsnd (cons (if i mod 2 = 0 then A else B)) (mk_word r (l+1))

  end;

fun f s zero = mk_word s 0

  | f s (Suc n) = f (fst (mk_word s 0)) n;

val g1 = snd o (f 20000.0);

val g2 = snd o (f 50000.0);

fun f1 zero = [A,A]

  | f1 (Suc zero) = [B]

  | f1 (Suc (Suc zero)) = [A,B]

  | f1 _ = [];

fun f2 zero = [A,A]

  | f2 (Suc zero) = [B]

  | f2 (Suc (Suc zero)) = [B,A]

  | f2 _ = [];

val (i1, j1) = higman g1;

val (v1, w1) = (g1 i1, g1 j1);

val (i2, j2) = higman g2;

val (v2, w2) = (g2 i2, g2 j2);

val (i3, j3) = higman f1;

val (v3, w3) = (f1 i3, f1 j3);

val (i4, j4) = higman f2;

val (v4, w4) = (f2 i4, f2 j4);

end;

*}

end