17388
|
1 |
(* ID: $Id$
|
|
2 |
Author: Sava Krsti\'{c} and John Matthews
|
|
3 |
*)
|
15636
|
4 |
|
17388
|
5 |
header {* Example use if an inductive invariant to solve termination conditions *}
|
|
6 |
|
|
7 |
theory InductiveInvariant_examples imports InductiveInvariant begin
|
15636
|
8 |
|
|
9 |
text "A simple example showing how to use an inductive invariant
|
|
10 |
to solve termination conditions generated by recdef on
|
|
11 |
nested recursive function definitions."
|
|
12 |
|
|
13 |
consts g :: "nat => nat"
|
|
14 |
|
|
15 |
recdef (permissive) g "less_than"
|
|
16 |
"g 0 = 0"
|
|
17 |
"g (Suc n) = g (g n)"
|
|
18 |
|
|
19 |
text "We can prove the unsolved termination condition for
|
|
20 |
g by showing it is an inductive invariant."
|
|
21 |
|
|
22 |
recdef_tc g_tc[simp]: g
|
|
23 |
apply (rule allI)
|
|
24 |
apply (rule_tac x=n in tfl_indinv_wfrec [OF g_def])
|
|
25 |
apply (auto simp add: indinv_def split: nat.split)
|
|
26 |
apply (frule_tac x=nat in spec)
|
|
27 |
apply (drule_tac x="f nat" in spec)
|
|
28 |
by auto
|
|
29 |
|
|
30 |
|
|
31 |
text "This declaration invokes Isabelle's simplifier to
|
|
32 |
remove any termination conditions before adding
|
|
33 |
g's rules to the simpset."
|
|
34 |
declare g.simps [simplified, simp]
|
|
35 |
|
|
36 |
|
|
37 |
text "This is an example where the termination condition generated
|
|
38 |
by recdef is not itself an inductive invariant."
|
|
39 |
|
|
40 |
consts g' :: "nat => nat"
|
|
41 |
recdef (permissive) g' "less_than"
|
|
42 |
"g' 0 = 0"
|
|
43 |
"g' (Suc n) = g' n + g' (g' n)"
|
|
44 |
|
|
45 |
thm g'.simps
|
|
46 |
|
|
47 |
|
|
48 |
text "The strengthened inductive invariant is as follows
|
|
49 |
(this invariant also works for the first example above):"
|
|
50 |
|
|
51 |
lemma g'_inv: "g' n = 0"
|
|
52 |
thm tfl_indinv_wfrec [OF g'_def]
|
|
53 |
apply (rule_tac x=n in tfl_indinv_wfrec [OF g'_def])
|
|
54 |
by (auto simp add: indinv_def split: nat.split)
|
|
55 |
|
|
56 |
recdef_tc g'_tc[simp]: g'
|
|
57 |
by (simp add: g'_inv)
|
|
58 |
|
|
59 |
text "Now we can remove the termination condition from
|
|
60 |
the rules for g' ."
|
|
61 |
thm g'.simps [simplified]
|
|
62 |
|
|
63 |
|
|
64 |
text {* Sometimes a recursive definition is partial, that is, it
|
|
65 |
is only meant to be invoked on "good" inputs. As a contrived
|
|
66 |
example, we will define a new version of g that is only
|
|
67 |
well defined for even inputs greater than zero. *}
|
|
68 |
|
|
69 |
consts g_even :: "nat => nat"
|
|
70 |
recdef (permissive) g_even "less_than"
|
|
71 |
"g_even (Suc (Suc 0)) = 3"
|
|
72 |
"g_even n = g_even (g_even (n - 2) - 1)"
|
|
73 |
|
|
74 |
|
|
75 |
text "We can prove a conditional version of the unsolved termination
|
|
76 |
condition for @{term g_even} by proving a stronger inductive invariant."
|
|
77 |
|
|
78 |
lemma g_even_indinv: "\<exists>k. n = Suc (Suc (2*k)) ==> g_even n = 3"
|
|
79 |
apply (rule_tac D="{n. \<exists>k. n = Suc (Suc (2*k))}" and x=n in tfl_indinv_on_wfrec [OF g_even_def])
|
|
80 |
apply (auto simp add: indinv_on_def split: nat.split)
|
|
81 |
by (case_tac ka, auto)
|
|
82 |
|
|
83 |
|
|
84 |
text "Now we can prove that the second recursion equation for @{term g_even}
|
|
85 |
holds, provided that n is an even number greater than two."
|
|
86 |
|
|
87 |
theorem g_even_n: "\<exists>k. n = 2*k + 4 ==> g_even n = g_even (g_even (n - 2) - 1)"
|
|
88 |
apply (subgoal_tac "(\<exists>k. n - 2 = 2*k + 2) & (\<exists>k. n = 2*k + 2)")
|
|
89 |
by (auto simp add: g_even_indinv, arith)
|
|
90 |
|
|
91 |
|
|
92 |
text "McCarthy's ninety-one function. This function requires a
|
|
93 |
non-standard measure to prove termination."
|
|
94 |
|
|
95 |
consts ninety_one :: "nat => nat"
|
|
96 |
recdef (permissive) ninety_one "measure (%n. 101 - n)"
|
|
97 |
"ninety_one x = (if 100 < x
|
|
98 |
then x - 10
|
|
99 |
else (ninety_one (ninety_one (x+11))))"
|
|
100 |
|
|
101 |
text "To discharge the termination condition, we will prove
|
|
102 |
a strengthened inductive invariant:
|
|
103 |
S x y == x < y + 11"
|
|
104 |
|
|
105 |
lemma ninety_one_inv: "n < ninety_one n + 11"
|
|
106 |
apply (rule_tac x=n in tfl_indinv_wfrec [OF ninety_one_def])
|
|
107 |
apply force
|
19623
|
108 |
apply (auto simp add: indinv_def inv_image_def)
|
15636
|
109 |
apply (frule_tac x="x+11" in spec)
|
|
110 |
apply (frule_tac x="f (x + 11)" in spec)
|
|
111 |
by arith
|
|
112 |
|
|
113 |
text "Proving the termination condition using the
|
|
114 |
strengthened inductive invariant."
|
|
115 |
|
|
116 |
recdef_tc ninety_one_tc[rule_format]: ninety_one
|
|
117 |
apply clarify
|
|
118 |
by (cut_tac n="x+11" in ninety_one_inv, arith)
|
|
119 |
|
|
120 |
text "Now we can remove the termination condition from
|
|
121 |
the simplification rule for @{term ninety_one}."
|
|
122 |
|
|
123 |
theorem def_ninety_one:
|
|
124 |
"ninety_one x = (if 100 < x
|
|
125 |
then x - 10
|
|
126 |
else ninety_one (ninety_one (x+11)))"
|
|
127 |
by (subst ninety_one.simps,
|
19623
|
128 |
simp add: ninety_one_tc inv_image_def)
|
15636
|
129 |
|
17388
|
130 |
end
|