> I should have been more clear: Assume the underlying representation > *is* IEEE 754. According to that, (sqrt -0.0) => -0.0, I believe, so > that doesn't help. If the underlying representation is IEEE 754, I think we should require (/ 1.0 0.0) to evaluate to +inf.0. Either way, (sqrt -0.0) might produce an inexact zero imaginary component. Kent