But do you really know the roots?  You do know the roots
of the polynomial q=:1;(i.20),20.0001 , but the roots of
the polynomial p. q are a different matter.  Is it possible
to represent p. q in IEEE 64-bit floating point to get
good answers for p. p. q ?

A warning about a polynomial root finder should say 
something to the effect that PROOT finding (by any
root finder) is extremely sensitive to the coefficients.
Vide the Wilkinson monster.  



----- Original Message -----
From: John Randall <[EMAIL PROTECTED]>
Date: Sunday, July 27, 2008 15:58
Subject: Re: [Jprogramming] Roots robustness?
To: Programming forum <[email protected]>

> Roger Hui wrote:
> > How do you know that those are not the right answers?
> > http://en.wikipedia.org/wiki/Wilkinson_polynomial
> 
> Because I know the roots.  Wilkinson's polynomial comes 
> from perturbing a
> coefficient after the polynomial is expanded.
> 
> >From the practical point of view:
> 
> roots=: >@(1&{)@p.
> q=:1;(i.20),20.0001
> p=:p. q
> r=:roots p
> 
>    >./|p&p. r
> 2.8237e14
----------------------------------------------------------------------
For information about J forums see http://www.jsoftware.com/forums.htm

Reply via email to