unstable recursions & nfloat



I have been looking at the code in nfloat.lisp and I don't see any commented compromises with power series that only 
involve +-/*^ so this is really fantastic.  Thanks for pointing out this function.

Rich


--------------------------------------------------
From: "Barton Willis" <willisb at unk.edu>
Sent: Saturday, March 27, 2010 5:36 PM
To: "Richard Hennessy" <rich.hennessy at verizon.net>
Cc: "Maxima List" <maxima at math.utexas.edu>
Subject: Re: [Maxima] unstable recursions & nfloat

> -----"Richard Hennessy" <rich.hennessy at verizon.net> wrote: -----
>
>
>>So is nfloat(I(42),[],20) = 8.7541921419809776312b-3 the right answer?  It
>>is certainly faster than setting fpprec:1000
>>and calling bfloat(I(42)).  That is the brute force way which I really
>>don't like.
>
> Yes, I think that nfloat(I(42),[],20) = 8.7541921419809776312b-3 is accurate
> to all displayed decimal places. Certainly, the goal of nfloat is to accurately
> evaluate some subset of Maxima expressions even when the expressions are numerically
> ill-conditioned. The algorithm nfloat uses has some compromises, so you can't, for
> example, prove a theorem using the output of nfloat.
>
> --bw
>