OFFSET
0,6
COMMENTS
a(14) is likely correct (see A093179); a(20) is unknown; a(21) to a(23) are 2097110, 4194189, 8388581; and a(24) is unknown.
2^(2^n - a(n)) < A093179(n).
Conjecture I: the dyadic valuation of A093179(n) - 1 does not exceed 2^n - a(n).
Conjecture II: a(n) ~ 2^n as n -> oo.
From Lorenzo Sauras Altuzarra, Jan 11 2026: (Start)
Moritz Firsching (personal communication) reports that AlphaProof solved Conjecture I by applying the formula below and the fact that A007814(k-1) <= floor(log_2(k)) for every integer k >= 2. See links for formal proof by AlphaProof.
Conjecture II is thus the question whether floor(log_2(A093179(n)))/2^n tends to zero or not. If there are infinitely many Fermat primes (which is currently unknown), then it cannot tend to zero. (End)
LINKS
Lorenzo Sauras-Altuzarra, Some properties of the factors of Fermat numbers, Art Discrete Appl. Math. (2022).
Google DeepMind, Lean proof of Firsching's formula below.
Google DeepMind, Lean proof of Conjecture I.
FORMULA
a(n) = 2^n-floor(log_2(A093179(n))). - Lorenzo Sauras Altuzarra, Jan 11 2026
EXAMPLE
For n=5, the smallest prime factor of F(5) = 2^(2^5) + 1 is 641 and it falls between 2^(2^5 - 23) = 512 < 641 < 1024 = 2^(2^5 - 22) so that a(5) = 23.
CROSSREFS
KEYWORD
nonn,hard,more
AUTHOR
Lorenzo Sauras Altuzarra, Nov 26 2022
STATUS
approved
