Rats. The error comes about because squaring the rational and algebraic square roots does not result in the same residue mod N. The algebraic square root doesn't care if relations have large factors, and the code for the rational square root is really simple, so my guess is that it's something to do with the formation of ideals when the underlying prime exceeds 2^32. The internal checks in the square root (powers of primes, powers of ideals, product of relations, Newton iteration) all pass, so all of the postprocessing produces internally consistent answers. I'll check a few more things.
PS: does SVN 56 change anything?
Last fiddled with by jasonp on 20090922 at 04:09
