|a > - 2a|
We proceed to the proof of Theorem 2.7.
Proof of Theorem 2.7. Following convention, we let denote the number of errors that the codes of a G-family can correct. It is well-known that each vector in 2nm-1 of weight t must be a unique coset leader for G m-1. Moreover, using our assumption about n0 (which also holds for nm, since nm n0), Lemma 2.8 implies that
|minwt(a | l), wt | (l )|
The result in Theorem 2.7 also applies to all G-families trivially seeded with the code d.
If a G-family is trivially seeded, then, necessarily, n0 = d and .Thus,
so that Theorem 2.7 applies to the remaining codes in the family.
We now turn our attention to generating mappings which produce codes whose information rate is locally maximized. More specifically, for any code with covering radius , we will consider only generating mappings f with the property that the Hamming distance from to is exactly . We will call such mappings minimal generating mappings, and the corresponding family of codes minimal G-codes, because they locally minimize length (and hence locally maximize information rate) for a given dimension and minimum-distance.
As an example, the generating mappings for the traditional lexicodes and the trellis-oriented lexicodes are both minimal. We may now easily strengthen Lemma 2.6 by observing in its proof that = d - for minimal G-codes.
The covering radius bounds we have developed for G-codes translates naturally to length bounds as well.