// the switching point was empirically determined. the optimal point
// depends on hardware, Digits, ... so an approx value is okay.
// it solves also the problem with precision due to the u=-log(1-x) transformation
// the switching point was empirically determined. the optimal point
// depends on hardware, Digits, ... so an approx value is okay.
// it solves also the problem with precision due to the u=-log(1-x) transformation