of the Pratt sequence. Moreover, Plaxton and Suel [35] establish
a lower bound of
(n log2 n/(log log n)2) for the
worst-case running time of Shellsort with any input sequence
(see also [11]) and Jiang et al. [23] establish a lower bound
of
(pn1+1/p) for the average-case running time of Shellsort.
Thus, the only way to achieve an O(n log n) averagetime
bound for Shellsort is to use an offset sequence of
length (log n), and, even then, the problem of proving
an O(n log n) average running-time bound for a version of
Shellsort is a long-standing open problem [42].