It's a lot easier to use overkill and set fpprec to an unreasonably high value and then do it all at that precision. Also it seems like my power series suffers from severe canceling behavior. The sum of the positive term is 10^75 for example and the sum of negatives is almost exactly the same number except for a slight 10^-20 difference.
Anyway using my overkill method I have an algorithm that can figure out the eigenvectors to what appears to be around 75 digits in about 3 minutes given a reasonable guess. It's like Newton's method, but slower.
Rich
------------Original Message------------
From: "Richard Hennessy"<rvh2007 at comcast.net>
To: "Barton Willis" <willisb at unk.edu>, "Barton Willis" <willisb at unk.edu>
Cc: "Maxima List" <maxima at math.utexas.edu>, maxima-bounces at math.utexas.edu
Date: Tue, May-6-2008 4:20 PM
Subject: Re: [Maxima] Big float Round off errors
I get a situation where you need to keep track of two numbers for every operation. Maybe I should be doing my power series as a vector (or some kind of object) sum to do both at the same time with overridden +-*/^ operators. This is getting complicated and messy since I don't know how to create a new type of Maxima object for this.
------------Original Message------------
From: Barton Willis <willisb at unk.edu>
To: "Barton Willis" <willisb at unk.edu>
Cc: "Maxima List" <maxima at math.utexas.edu>, maxima-bounces at math.utexas.edu
Date: Tue, May-6-2008 3:45 PM
Subject: Re: [Maxima] Big float Round off errors
A Maxima version:
(%i1) bigfloatepsilon () := bfloat(2^-?fpprec);
Barton
-----maxima-bounces at math.utexas.edu wrote: -----
>(%i2) :lisp(defun $bigfloatepsilon () ($bfloat (div 1 (expt 2 fpprec)))));
>$BIGFLOATEPSILON
>
_______________________________________________
Maxima mailing list
Maxima at math.utexas.edu
http://www.math.utexas.edu/mailman/listinfo/maxima
_______________________________________________
Maxima mailing list
Maxima at math.utexas.edu
http://www.math.utexas.edu/mailman/listinfo/maxima