Thanks for all your comments. Yes, you are right. It seems 2 is known to be a particularly poor choice - mystery definitely solved there. However, the sequence based on 3n+1 (You start with 1 and iterate 3n+1 until you reach the number of elements, then back down using a divisor of 3) suggested by the university chap isn't too good either, at least for me and Blitz.
That's interesting about the empirical 1.3 gap. Another respondent insists that the square root of two is fastest for him.
My figures are about the same as yours, sswift, except that a definite minimum exists at 1.6 for me over a large range of initial data distributions and array sizes.
That's interesting about the empirical 1.3 gap. Another respondent insists that the square root of two is fastest for him.
My figures are about the same as yours, sswift, except that a definite minimum exists at 1.6 for me over a large range of initial data distributions and array sizes.