This returns the complementary error function of x.  It is useful for calculating an integral of a tail of a normal distribution.

It would be easy enough to just use 1-erf(x), but this provides better numerical accuracy when x is large (so erf(x) is very close to 1).