unit normal_cdf_inverse;
interface
uses math;
function normal_cdf_inverse(p: float): float;
implementation
function rational_approximation(t: float): float;
const
c: array [0..2] of float = (2.515517, 0.802853, 0.010328);
d: array [0..2] of float = (1.432788, 0.189269, 0.001308);
var numerator: float;
denominator: float;
begin
{ Abramowitz and Stegun formula 26.2.23. }
{ The absolute value of the error should be less than 4.5 e-4. }
numerator := (c[2]*t + c[1])*t + c[0];
denominator := ((d[2]*t + d[1])*t + d[0])*t + 1.0;
rational_approximation := t - numerator / denominator;
end;
function normal_cdf_inverse(p: float): float;
begin
assert((p > 0.0) and (p < 1));
{ See article above for explanation of this section. }
if p < 0.5 then
{ F^-1(p) = - G^-1(p) }
normal_cdf_inverse := -rational_approximation( sqrt(-2.0*ln(p)) )
else
{ F^-1(p) = G^-1(1-p) }
normal_cdf_inverse := rational_approximation( sqrt(-2.0*ln(1.0-p)) );
end;
end.