Statistical Distribution Library

The noncentral F CDF method starts giving nonsense answers when the non
centrality parameter exceeds a certain limit. This error can be traced to
the use of the noncentral beta CDF in the noncentral F CDF. This error can
be temporarily fixed by going to the noncentral_beta. cumulative method and
commenting out the line that reads:

//a0 = a + x0;

and substituting

a0 = a;

then increase the iteration maximum from 100 to 1000

final double itrmax = 1000;

Without going into detail, it appears that the only purpose of x0 is to
speed up the rate of convergence in the iterative part of the method when
the non centrality parameter exceeds a certain size(but I'm really not
sure). Instead it appears x0 causes the method to converge to a wrong
answer or an error. Anyway, my fix seems to work for the limited values I
tested it with, but there is no guarantee.