Minor Description
Add a scaled complementary error function:
erfcx(x) = exp(x^2) * erfc(x)
For small x this can use the current rational function approximations for erf or erfc and remove the scaling down by exp(-x^2). For large x this can use the continued fraction expansion of erfc and remove the scaling by e^-z:
z 1
erfc z = -------- e^-z^2 -----------------------------
sqrt(pi) z^2 + 1/2
-----------------------
1 + 1
-------------------
z^2 + 3/2
-------------
1 + 2
---------
z^2 + ...
At very large x the rational function cannot be evaluated as z^2 + 0.5 ~ z^2 and the following approximation can be used:
1 1
erfcx z = -------- * -
sqrt(pi) z
This occurs at 2^26 or approximately 6.71e7.
