ErrorFunction.java

Below is the syntax highlighted version of ErrorFunction.java from §2.1 Static Methods.

/******************************************************************************
 *  Compilation:  javac ErrorFunction.java
 *  Execution:    java ErrorFunction z
 *
 *  Implements the Gauss error function.
 *
 *              erf(z) = 2 / sqrt(pi) * integral(exp(-t*t), t = 0..z)
 *
 *
 *  % java ErrorFunction 1.0
 *  erf(1.0) = 0.8427007877600067         // actual = 0.84270079294971486934
 *  Phi(1.0) = 0.8413447386043253         // actual = 0.8413447460
 *
 *
 *  % java ErrorFunction -1.0
 *  erf(-1.0) = -0.8427007877600068
 *  Phi(-1.0) = 0.15865526139567465
 *
 *  % java ErrorFunction 3.0
 *  erf(3.0) = 0.9999779095015785         // actual = 0.99997790950300141456
 *  Phi(3.0) = 0.9986501019267444
 *
 *  % java ErrorFunction 30
 *  erf(30.0) = 1.0
 *  Phi(30.0) = 1.0
 *
 *  % java ErrorFunction -30
 *  erf(-30.0) = -1.0
 *  Phi(-30.0) = 0.0
 *
 *  % java ErrorFunction 1E-20
 *  erf(1.0E-20)  = -3.0000000483809686E-8     // true anser 1.13E-20
 *  Phi(1.0E-20)  = 0.49999998499999976
 *
 *
 ******************************************************************************/

public class ErrorFunction {


    // fractional error in math formula less than 1.2 * 10 ^ -7.
    // although subject to catastrophic cancellation when z in very close to 0
    // from Chebyshev fitting formula for erf(z) from Numerical Recipes, 6.2
    public static double erf(double z) {
        double t = 1.0 / (1.0 + 0.5 * Math.abs(z));

        // use Horner's method
        double ans = 1 - t * Math.exp( -z*z   -   1.26551223 +
                                            t * ( 1.00002368 +
                                            t * ( 0.37409196 +
                                            t * ( 0.09678418 +
                                            t * (-0.18628806 +
                                            t * ( 0.27886807 +
                                            t * (-1.13520398 +
                                            t * ( 1.48851587 +
                                            t * (-0.82215223 +
                                            t * ( 0.17087277))))))))));
        if (z >= 0) return  ans;
        else        return -ans;
    }

    // fractional error less than x.xx * 10 ^ -4.
    // Algorithm 26.2.17 in Abromowitz and Stegun, Handbook of Mathematical.
    public static double erf2(double z) {
        double t = 1.0 / (1.0 + 0.47047 * Math.abs(z));
        double poly = t * (0.3480242 + t * (-0.0958798 + t * (0.7478556)));
        double ans = 1.0 - poly * Math.exp(-z*z);
        if (z >= 0) return  ans;
        else        return -ans;
    }

    // cumulative normal distribution
    // See Gaussia.java for a better way to compute Phi(z)
    public static double Phi(double z) {
        return 0.5 * (1.0 + erf(z / (Math.sqrt(2.0))));
    }



   /***************************************************************************
    *  Test client
    ***************************************************************************/
    public static void main(String[] args) {
        double x = Double.parseDouble(args[0]);

        StdOut.println("erf(" + x + ")  = " + ErrorFunction.erf(x));
        StdOut.println("erf2(" + x + ") = " + ErrorFunction.erf2(x));
        StdOut.println("Phi(" + x + ")  = " + ErrorFunction.Phi(x));
        StdOut.println();
    }

}