(Sorry about the missing 3F2[1/64] previous subject line.) I just (re?)found another misprint in my ancient (1974) AI Memo 304 series acceleration paper, dspace.mit.edu/handle/1721.1/6088 : p 73, 2^j-1 should be 2^j+1 in the quasidouble sum for Euler's constant: inf ==== j + 1 \ 6 k + 2 + 3 > ---------------------------------- inf / k + 1 j ==== ==== 4 binomial(2 k + 2 + 1, 2 k) \ k = 0 %gamma = > ---------------------------------------- / j j ==== 2 (2 + 1) j = 1 (Quasi because you only need ~ n log n terms for n digits.) I'll post a corrected version on gosper.org when I figure out how to mark it up with Acrobat. Oddity: inf ==== \ n 1 n > (psi (2 + -) - psi (2 )) / 0 2 0 ==== n = 0 %gamma = ------------------------------- 2 inf ==== \ n + 1 n = > (psi (x ) - psi (x ) - log(x)), / 0 0 ==== n = 0 x>1, where psi[0](k) can be replaced by H[k-1]. It must be independent of x, since grouping terms pairwise is the same as squaring x. --rwg PS: Substituting x=1, we have the surprising result that %gamma=0.-)