Common lisps are supposed to have values for
least-positive-double-float
5.0d-324
least-positive-single-float
1.0e-45
as well as
least-positive-normalized-single-float
1.1754944e-38
least-positive-normalized-double-float
2.2250738585072014d-308
perhaps these should be used instead of #+pdp-10.
Or do you need for eps the least number such that (+ 1.0d0 eps) is
different from 1.0d0? They too should be there:
short-float-epsilon value: 5.960465e-8
single-float-negative-epsilon value: 2.9802326e-8
single-float-epsilon value: 5.960465e-8
long-float-epsilon value: 1.1102230246251568d-16
double-float-negative-epsilon value: 5.551115123125784d-17
long-float-negative-epsilon value: 5.551115123125784d-17
double-float-epsilon value: 1.1102230246251568d-16
short-float-negative-epsilon value: 2.9802326e-8
willisb@unk.edu wrote:
> Make that:
>
> The value of eps in trigi.lisp is too *large* -- this gives poor numerical
> accuracy for some trigonometric functions. An example ...
>
> BW
>
>
> _______________________________________________
> Maxima mailing list
> Maxima@www.math.utexas.edu
> http://www.math.utexas.edu/mailman/listinfo/maxima
>