On 11/2/10 12:15 PM, Robert Dodier wrote:
> Looks like the criterion for figuring out when to stop iterating
> depends on fpprec. Not surprisingly, if you load newton.mac
> when a larger value of fpprec is in effect, it takes more iterations
> or it fails to converge according to the criterion.
> I think if you set newtonepsilon (? not sure about the name)
> to a different value, it will change the behavior of newton.
Dieter fixed this yesterday with rev 1.15 of share/contrib/mnewton. The
default value of newtonepsilon is now always 10^(-8) instead of
depending on fpprec. If you want a different precision, you have to set
newtonepsilon to the appropriate value, after loading mnewton.
Ray