|
|
Blitz Support : |
From: Patrik Jonsson (patrik_at_[hidden])
Date: 2004-12-22 16:55:58
This is true, but using 64-bit arithmetic the results should hopefully not
differ by 10^-2...
/Patrik
At 02:17 PM 12/19/2004, you wrote:
>Hi Patrik,
>
>I don't think that this is a bug, because arithmetic with computers are
>neither commutative, associative nor the distributive law is valid.
>
>Jiri Kraus
> > Hi for the third time,
> > As part of this whole deal, I've also discovered something that looks like
> > an evaluation bug unless I'm missing something stupid. In the following
> > code, T_indep_vector means Array<TinyVector<double,4>,1> and vector means
> > Array<double,1>.
> >
> > vector operator()(const T_indep_vector& x, const vector& par) {
> > T_indep_vector xx(static_cast<T_indep_vector>(x));
> > const vector vv(par(0)*pow(xx[0]/1e11,par(1))*
> > pow(xx[1]/0.02,par(2))*pow(xx[2]/1e11,par(3))*
> > *pow(xx[3]/1,par(4)));
> > const vector v1(pow(xx[0]/1e11,par(1)));
> > const vector v2(pow(xx[1]/0.02,par(2)));
> > const vector v3(pow(xx[2]/1e11,par(3)));
> > const vector v4(pow(xx[3]/1,par(4)));
> > const vector crap(par(0)*v1*v2*v3*v4);
> >
> > I hope that I haven't screwed up and that in this example, vv and crap
> > should contain identical numbers at the end. They don't. crap is right,
> but
> > vv is incorrectly calculated (by a small amount, very deceptive...) Looks
> > pretty serious if it's indeed true and I haven't screwed up.
> >
> > That's the last bug for now. Promise. :-)
> >
> > cheers,
> >
> > /Patrik
> >
> >
> > ============================================================
> > Patrik Jonsson (831) 459-3828
> > Department of Astronomy & Astrophysics
> > University of California, Santa Cruz, CA 95064
> >
> > This message has been written using a voice recognition system.
> > Words that don't make sense or not the fault of the author...
> >
> > _______________________________________________
> > Blitz-support mailing list
> > Blitz-support_at_[hidden]
> > http://www.oonumerics.org/mailman/listinfo.cgi/blitz-support
> >
>
>_______________________________________________
>Blitz-support mailing list
>Blitz-support_at_[hidden]
>http://www.oonumerics.org/mailman/listinfo.cgi/blitz-support