From: jarausch@numa1.igpm.rwth-aachen.de (Helmut Jarausch)
Newsgroups: sci.math.num-analysis
Subject: Re: Precision, Precision - where can I get you?
Date: 2 Jan 1998 09:43:12 GMT

In article <34a95c61.334783192@news-s01.ny.us.ibm.net>, CarlDS@ibm.net (Carl D. Speare) writes:
|> Ah the time old question. Precision is alwyas a problem.
|> Never enough of it I suppose. I am using a C/C++ development
|> environment, Borland C++ Development Suite 5.0.2, to be exactly
|> precise, and I'm tired of long double types only getting 19 digits of
|> precision. I've tried to get Arjen Lenstra's freeLIP package to work,
|> but I keep running into a problem. Anyone know of a multiple precision
|> package, preferably free, that works with Borland products?

Precision is alwyas a problem - How true!

Precision in a single (floating point) operation is fine but may give
you a misleading feeling of safety. What one has really to care about
is the accumatation and amplification of errors.

This all is the subject of 'Reliable Computing'.
There is a journal with this same title and a mailing list
majordomo@interval.usl.edu
subscribe reliable_computing

To start with I recommend a very nice package -- written in C++ --
which is called Range by O. Aberth.
It implements variable precision 'range arithmetic'
but what is even more important, it has ready to use examples which 
solve quite a wide range of tasks, as there is
- solving real and complex linear and nonlinear equations in several variables
- solving linear equations in rational arithmetic -> ultimate precision
- computing integrals with arbitrary and verified precision
- solving ordinary boundary value problems
- computing eigenvalues
- and of course a nice C++ library for doing computations yourself

The best news - it's free

Look at

ftp://ftp.math.tamu.edu/pub/aberth/range.tar.Z

I am sure you'll enjoy it - as I do.

-- 
Helmut Jarausch
Lehrstuhl f. Numerische Mathematik
Institute of Technology, RWTH Aachen
D 52056 Aachen, Germany