From: lrmsx AT my-deja DOT com
Newsgroups: comp.os.msdos.djgpp
Subject: Re: multidimensional arrays and math inaccuracy
Date: Sun, 30 May 1999 20:09:51 GMT
Organization: Deja.com - Share what you know. Learn what you don't.
Lines: 51
Message-ID: <7is5uf$7u5$1@nnrp1.deja.com>
References: <374edbb4 DOT 18530235 AT news DOT ndh DOT net>
NNTP-Posting-Host: 200.248.54.9
X-Article-Creation-Date: Sun May 30 20:01:58 1999 GMT
X-Http-User-Agent: Mozilla/4.6 [en] (Win95; I)
To: djgpp AT delorie DOT com
DJ-Gateway: from newsgroup comp.os.msdos.djgpp
Reply-To: djgpp AT delorie DOT com

In article <374edbb4 DOT 18530235 AT news DOT ndh DOT net>,
  fnatter AT gmx DOT net wrote:
> is this 100% correct when I then access the
elements via elems[x][y]?
I see no problem in the memory acessing or
allocation.

(This is really not DJGPP realted it's numerical
math but anyway...)

> the reason I'm asking is because I use gaussian
elimination to
> transform the matrix to "diagonal" form (for
solving a linear system
> of equations), and when I do so, on a system
where there is a unique
> solution (in this code it's .2, .4, .2), there
will be a math
> inaccuracy: where an entry should have been
reduced to 0.0, it becomes
> something*10^(-15), resulting in the algorithm
continuing and giving a
> wrong result!
After making the matrix diagonal you should only
use the elements at the diagonal itself to
calculate the result. Supposing you are really
using (or want to) a full maxtrix multiply to get
the solution that's what you should do: (
that's what I learned in the numerical methods) in
the Gauss Elimination algorithm where some element
is supposed to be eleminated (zeroed) simply put a
zero (0.0) there, don't really calculate it's
value as every single machine in the world will
have limited precision and result will not be
accurate for all digitis. Simply zeroing the
elements will also make the calculation faster, as
the number of calculated columns will decrease
after each line is calculated. Remember also that
to get 7 digiti precision after doing some
operations with floating point numbers you will
need more than 7 digitis (in the floating point
operations). Double precision has about 15 digitis
(53 bits in the mantissa if a remember correctly).
So it's expected to have some 10^-15 (or smaller)
result where it really should be zero.

LRMS


Sent via Deja.com http://www.deja.com/
Share what you know. Learn what you don't.