%I #26 Jan 18 2017 23:33:05
%S 11,1093,1093,3511,3511,5557,104891,1006003
%N GCDs arising in A126196.
%C All terms are primes. Note a connection to the Wieferich primes A001220: a(2) = a(3) = A001220(1), a(3) = a(4) = A001220(2).
%C From _John Blythe Dobson_, Jan 14 2017: (Start)
%C All Wieferich primes p will belong to this sequence twice, because if H([p/k]) denotes the harmonic number with index floor(p/k), then p divides all of H([p/4]), H([p/2]), and H(p-1). The first two of these elements gives one solution, and the second and third another. This property of the Wieferich primes predates their name, and was apparently first proved by Glaisher in "On the residues of r^(p-1) to modulus p^2, p^3, etc.," pp. 21-22, 23 (see References).
%C Note also a connection to the Mirimanoff primes A014127: a(1) = A014127(1), a(8) = A014127(2). All Mirimanoff primes p will belong to this sequence, because p divides both H([p/3]) and H([2p/3]). This property of the Mirimanoff primes likewise predates their name, and was apparently first proved by Glaisher in "A general congruence theorem relating to the Bernoullian function," p. 50 (see Links).
%C The Wieferich primes and Mirimanoff primes would seem to be the only cases for which the value of n in A126196(n) is predictable from knowledge of p. It is not obvious that all members of the present sequence are prime; however, by definition all their divisors must be non-harmonic primes A092102. Furthermore, it is clear from the cited literature under that entry that H([n/2]) == H(n) == 0 (mod p) is only possible when n < p. Thus, all divisors of the present sequence must belong to the harmonic irregular primes A092194.
%C One possible reason for interest in this sequence is a 1995 result of Dilcher and Skula (see Links) which among other things shows that if a prime p were an exception to the first case of Fermat's Last Theorem, then p would divide both H([p/k]) and H([2p/k]) for every value of k from 2 to 46. To date, the only values for which such coincidences have been found have k = 2, 3, or 4. For k = 6 to hold, p would have to be simultaneously a Wieferich prime and a Mirimanoff prime, while for k = 5 to hold, p would have to be simultaneously a Wall-Sun-Sun prime and a member of A123692. The sparse numerical results for the present sequence suggest that even the more relaxed condition H([n/2]) == H(n) == 0 (mod p) is rarely satisfied. (End)
%D J. W. L. Glaisher, On the residues of r^(p-1) to modulus p^2, p^3, etc., Quarterly Journal of Pure and Applied Mathematics 32 (1900-1901), 1-27.
%H Karl Dilcher and Ladislav Skula, <a href="https://doi.org/10.1090/S0025-5718-1995-1248969-6">A new criterion for the first case of Fermat's Last Theorem</a>, Mathematics of Computation 64 (1995), 363-392.
%H J. W. L. Glaisher, <a href="https://archive.org/stream/proceedingslond00unkngoog#page/n33/mode/2up">A general congruence theorem relating to the Bernoullian function</a>, Proceedings of the London Mathematical Society 33 (1900-1901), 27-56.
%t f[n_] := GCD @@ Numerator@ HarmonicNumber@ {n, Floor[n/2]}; f@ Select[ Range[5000], f[#] > 1 &] (* _Giovanni Resta_, May 13 2016 *)
%Y Cf. A001008, A001220, A002805, A058313, A074791, A121594, A125581, A126196.
%K nonn,more
%O 1,1
%A _Max Alekseyev_ and _Tanya Khovanova_, Mar 07 2007
%E a(8) from _Giovanni Resta_, May 13 2016