doc-src/TutorialI/Recdef/termination.thy
author nipkow
Wed Dec 13 09:39:53 2000 +0100 (2000-12-13)
changeset 10654 458068404143
parent 10522 ed3964d1f1a4
child 10795 9e888d60d3e5
permissions -rw-r--r--
*** empty log message ***
     1 (*<*)
     2 theory termination = examples:
     3 (*>*)
     4 
     5 text{*
     6 When a function is defined via \isacommand{recdef}, Isabelle tries to prove
     7 its termination with the help of the user-supplied measure.  All of the above
     8 examples are simple enough that Isabelle can prove automatically that the
     9 measure of the argument goes down in each recursive call. As a result,
    10 $f$@{text".simps"} will contain the defining equations (or variants derived
    11 from them) as theorems. For example, look (via \isacommand{thm}) at
    12 @{thm[source]sep.simps} and @{thm[source]sep1.simps} to see that they define
    13 the same function. What is more, those equations are automatically declared as
    14 simplification rules.
    15 
    16 In general, Isabelle may not be able to prove all termination conditions
    17 (there is one for each recursive call) automatically. For example,
    18 termination of the following artificial function
    19 *}
    20 
    21 consts f :: "nat\<times>nat \<Rightarrow> nat";
    22 recdef f "measure(\<lambda>(x,y). x-y)"
    23   "f(x,y) = (if x \<le> y then x else f(x,y+1))";
    24 
    25 text{*\noindent
    26 is not proved automatically (although maybe it should be). Isabelle prints a
    27 kind of error message showing you what it was unable to prove. You will then
    28 have to prove it as a separate lemma before you attempt the definition
    29 of your function once more. In our case the required lemma is the obvious one:
    30 *}
    31 
    32 lemma termi_lem: "\<not> x \<le> y \<Longrightarrow> x - Suc y < x - y";
    33 
    34 txt{*\noindent
    35 It was not proved automatically because of the special nature of @{text"-"}
    36 on @{typ"nat"}. This requires more arithmetic than is tried by default:
    37 *}
    38 
    39 apply(arith);
    40 done
    41 
    42 text{*\noindent
    43 Because \isacommand{recdef}'s termination prover involves simplification,
    44 we include with our second attempt the hint to use @{thm[source]termi_lem} as
    45 a simplification rule:\indexbold{*recdef_simp}
    46 *}
    47 
    48 consts g :: "nat\<times>nat \<Rightarrow> nat";
    49 recdef g "measure(\<lambda>(x,y). x-y)"
    50   "g(x,y) = (if x \<le> y then x else g(x,y+1))"
    51 (hints recdef_simp: termi_lem)
    52 
    53 text{*\noindent
    54 This time everything works fine. Now @{thm[source]g.simps} contains precisely
    55 the stated recursion equation for @{term g} and they are simplification
    56 rules. Thus we can automatically prove
    57 *}
    58 
    59 theorem "g(1,0) = g(1,1)";
    60 apply(simp);
    61 done
    62 
    63 text{*\noindent
    64 More exciting theorems require induction, which is discussed below.
    65 
    66 If the termination proof requires a new lemma that is of general use, you can
    67 turn it permanently into a simplification rule, in which case the above
    68 \isacommand{hint} is not necessary. But our @{thm[source]termi_lem} is not
    69 sufficiently general to warrant this distinction.
    70 
    71 The attentive reader may wonder why we chose to call our function @{term g}
    72 rather than @{term f} the second time around. The reason is that, despite
    73 the failed termination proof, the definition of @{term f} did not
    74 fail, and thus we could not define it a second time. However, all theorems
    75 about @{term f}, for example @{thm[source]f.simps}, carry as a precondition
    76 the unproved termination condition. Moreover, the theorems
    77 @{thm[source]f.simps} are not simplification rules. However, this mechanism
    78 allows a delayed proof of termination: instead of proving
    79 @{thm[source]termi_lem} up front, we could prove 
    80 it later on and then use it to remove the preconditions from the theorems
    81 about @{term f}. In most cases this is more cumbersome than proving things
    82 up front.
    83 %FIXME, with one exception: nested recursion.
    84 *}
    85 
    86 (*<*)
    87 end
    88 (*>*)