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

Page 42
Case 2: 0042-001.gif. Then by 3.23 and the fact that 0042-002.gif, choose 0042-003.gif such that 0042-004.gif and 0042-005.gif; let 0042-006.gif.
Observe that 0042-007.gif for all 0042-008.gif. Let 0042-009.gif. T is a text for L, since u(n) is added to T at stage n (so 0042-010.gif) and no nonmembers of L are ever added to T (so 0042-011.gif). Finally, F does not converge on T to an index for L, since for each 0042-012.gif either 0042-013.gif or, for some 0042-014.gif with 0042-015.gif, 0042-016.gif. Thus, there are infinitely many numbers m0, m1, m2, . . . such that either F(T[mi]) is an incorrect index or 0042-017.gif.
Intuitively, if F identifies T, then Theorem 3.22 guarantees the existence of 0042-018.gif that "locks" F onto a conjecture for L in the following sense: no presentation from L can dislodge F from F( s ). This suggests the following definition.
3.24 Definition (Blum and Blum [18]) Let 0042-019.gif, scientist F and 0042-020.gif be given.  s  is a locking sequence for F on L just in case:
(a) 0042-021.gif
(b)WF( s ) = L;
(c) for all 0042-022.gif, if 0042-023.gif, then 0042-024.gif.
So Theorem 3.22 can be put this way:
3.25 Corollary (Blum and Blum [18]) If scientist F identifies 0042-025.gif, then there is a locking sequence for F on L.
Note that Theorem 3.22 does not characterize F's behavior on elements drawn from Image-0472.gif. In particular, if 0042-026.gif is such that 0042-027.gif, then even if  s  is a locking sequence for F on L, 0042-028.gif may well differ from F( s ).
§3.6.2 Characterization
We now state and prove a characterization of the collections of identifiable languages.
3.26 Theorem (Based on Angluin [6]) 0042-029.gif is identifiable if and only if for all 0042-030.gif there is finite 0042-031.gif such that for all 0042-032.gif, if 0042-033.gif then 0042-034.gif.

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