The formula boils down to k - 1 ==== \ k + 1 k - j > B ( ) n U (n) / j + 1 j + 1 j ==== ==== k + 1 \ k j = 1 phi(n ) > t = ----------------------------------- + -----------, / k + 1 k + 1 ==== t < n (t,n)=1 where B are Bernoulli numbers, j a b c a-1 b-1 c-1 phi(p q r ...) := p (1 - p) q (1 - q) r (1 - r) ..., and a b c k k k U (p q r ...) := (1 - p ) (1 - q ) (1 - r ) ..., k where p, q, r, ... are distinct primes. This violates my principle of always writing B_j(y) instead of B_j, but what neat thing does this compute with Bernpolys instead of Berns?? --rwg