[Cover] [Contents] [Index] Previous page Next Section

Page 115
§5.6 Strategies for Function Identification
We now turn our attention to strategies for function identification. As in the case of languages, we consider both the global and class versions where they are meaningful. Notation: Recall from Chapter 4 that f[n] denotes both the partial function 0115-001.gif and the initial segment of length n of the canonical text for f. Also recall from Definition 4.18 that 0115-002.gif.
§5.6.1 Consistency
The next definition formalizes consistency in the context of function identification.
5.61 Definition (Barzdins * [13], Blum and Blum [18])
(a) M is consistent on f just in case for all n, 0115-003.gif.
(b) M is consistent on 0115-004.gif just in case M is consistent on each 0115-005.gif.
(c) M is consistent just in case M is consistent on Image-1216.gif.
(d) 0115-006.gif.
(e) 0115-007.gif.
Just as in the language paradigm, consistency has the ring of rationality. However, similarly to before, implementing the strategy can entail a loss of inductive power.
5.62 Proposition 0115-008.gif.
Proof: Clearly, 0115-009.gif.
We first show that 0115-010.gif.
Consider the collection of self-describing functions 0115-011.gif. It is easy to show that 0115-012.gif. Suppose by way of contradiction that a scientist M that is consistent (on Image-1217.gif) identifies Image-1218.gif. Then, by Kleene's recursion theorem (Theorem 2.3), there exists an index e such that, for all x,
0115-013.gif
It is clear that 0115-014.gif. Also, since M is consistent, for each x > 0, either 0115-015.gif or 0115-016.gif. Hence, for each x > 0, 0115-017.gif It follows that 0115-018.gif, a contradiction.

 
[Cover] [Contents] [Index] Previous page Next Section