Re: [isabelle] Termination for ordinal-like datatypes

Hi Eddy,

as René said, I also would use primrec wherever possible. However, if you are bound to function you can prove termination by defining a subterm relation and prove that it is wellfounded:

definition "sub ≡ {(x, Succ x) | x. True} ∪ {(f n, Limit f) | n f. True}"

lemma subI[intro]:
  "(x, Succ x) ∈ sub"
  "(f n, Limit f) ∈ sub"
  unfolding sub_def by blast+

lemma wf_sub[simp]: "wf sub"
proof (rule wfUNIVI)
  fix P x
  assume "∀x. (∀y. (y, x) ∈ sub ⟶ P y) ⟶ P x"
  then show "P x" unfolding sub_def by (induct x) blast+

Both of your functions can be then proved terminating via

termination by (relation "sub <*lex*> sub") auto


On 15.07.2014 03:08, Eddy Westbrook wrote:

I was wondering if someone could help me figure out an easy way to prove termination for functions over ordinal-like datatypes. What I mean by this is datatypes T where one of the constructors for T takes a function that returns type T. The classic one is the ordinals:

datatype ord
   = Zero
   | Succ ord
   | Limit "(nat => ord)”

My question is, what is the easiest and best way to write termination proofs for recursive functions over the type ord? Are there some mechanisms already available, or do I have to define my own inductive proposition or something similar? It looks like all the default mechanisms for proving termination rely on types that satisfy “size”, which ord does not because of the Limit constructor.

For example, I want to define functions like these, which should be terminating:

function ord_plus :: "ord => ord => ord"
   "ord_plus x Zero = x"
   | "ord_plus x (Succ y) = Succ (ord_plus x y)"
   | "ord_plus x (Limit f) = Limit (\<lambda> n . ord_plus x (f n))"
   by pat_completeness auto

function ord_leq :: "ord ⇒ ord ⇒ bool"
   "ord_leq Zero x = True"
   | "ord_leq (Succ x) Zero = False"
   | "ord_leq (Succ x) (Succ y) = ord_leq x y"
   | "ord_leq (Succ x) (Limit f) = (\<exists> n. ord_leq (Succ x) (f n))"
   | "ord_leq (Limit f) y = (\<forall>n . ord_leq (f n) y)"
   by pat_completeness auto

Thanks very much for any help or suggestions.

This archive was generated by a fusion of Pipermail (Mailman edition) and MHonArc.