(04-03-2018 01:28 PM)Werner Wrote: The 12-digit 42S gets the exact result - it must have special code for a=0, and return SQRT(ABS(b)/2)*(1+i*SIGN(b)) in that case - even when b/2 would underflow.

It must have some kind of special code. I had the same artifacts and had to add special code in newRPL to detect tiny values in the imaginary part. If found, then the real part is tried by itself as a root, and if it satisfies the equation the real root is returned instead (sort of "snap" to the real axis).