erf(1d-10) -> 1.128e-40
But the taylor series says erf(x) = 2*x/sqrt(%pi) so erf(1d-10) should be 1.128e-20.
The error is in slatec:derf which is using the wrong formula when x < sqeps.
Change summary to mention this is just for float values.
Fixed by modifying derf.f and derf.lisp