arXiv++ Combinatorics

Experiments are conducted to improve Tokuda (1992) gap sequence in Shellsort into $γ$-sequences, and the best result is the gap sequence in which the $k$-th increment $h_k$ is given by \begin{align} h_k=\left\lceil \frac{γ^k-1}{γ-1} \right\rceil \end{align} , where $γ=2.243609061420001...$ and $k\in\mathbb{N}_1$. The first few increments of the gap sequence are \begin{align} 1,\, 4,\, 9,\, 20,\, 45,\, 102,\, 230,\, 516,\,1158,\,2599,\,5831,\,13082,\,29351,\,65853,\, 147748,\,331490,\,743735,\, ...\end{align}It empirically yields less numbers of comparison on average than Tokuda (1992) gap sequence. In the procedure of search, it reveals the potential existence of a new type of fractal.

Optimal gap sequences in Shellsort, defined as gap sequences having the minimised maximum number of comparisons for a fixed number of pairwise distinct elements, are found by minimax search in reduced permutational spaces, namely Bad $(s,1)$-sorted permutations. Exact optimal gap sequences in Shellsort for $n\leq16$ pairwise distinct elements are established, and the best known gap sequences for $17\leq n\leq 30$ are listed with conjectures made. It notably discovers some optimal gap sequences consist of increments larger than the half of the total number of the elements to sort.