The expected codeword length L(UD) of the best uniquely decodable (UD)
code satisfies H(P;Q) less-than-or-equal-to L(UD) < H(P;Q) + 1, where
H(P;Q) is the Kerridge inaccuracy. By applying the idea of the best 1
:1 code given by Leung Yan Cheong and T. Cover [IEEE Trans. Inform. Th
eory IT-24, 331-338 (1978)] a relation between inaccuracy and the best
1: I codeword length L1:1 has been obtained. Further, it is shown tha
t the average codeword length L1:1 is shorter than the average codewor
d length L(UD) by no more than log.log n + 3. Also, the lower bounds t
o the exponentiated mean codeword length in terms of the inaccuracy of
type alpha have been obtained.