RE: [Help-glpk] control parameters and round-off error

[Sebastian Pokutta]

Ali,

If you use a linear program you might also consider using the exact simplex
solver in glpk. In this case the simplex solution should be exact.

Regards,
Sebastian

-----Original Message-----
From: address@hidden
[mailto:address@hidden On Behalf Of
Michael Hennebry
Sent: 15 January 2008 06:24
To: Ali Baharev
Cc: glpk
Subject: Re: [Help-glpk] control parameters and round-off error

On Tue, 15 Jan 2008, Ali Baharev wrote:

> I do not know the solution with guaranteed accuracy, it would require
> a self-validating method such as interval arithmetic. I only suspect
> that the solver has numerical problems as i have inconsistent results.
> An error estimate on the result of the solver could help me.

In what way are they inconsistent?

There is an algorithm for computing sums fairly precisely.
I can't think of the name right now.
If doing a dot product, your roundoff should be
bounded by roughly 3*eps*(sum of absolute values).
A third of that is the roundoff from forming the products.
With IEEE binary arithmetic, eps is the largest
number such that the machine computes 1+eps to be 1.

-- 
Michael   address@hidden
"Those parts of the system that you can hit with a hammer (not advised)
are called Hardware;  those program instructions that you can only
curse at are called Software."



_______________________________________________
Help-glpk mailing list
address@hidden
http://lists.gnu.org/mailman/listinfo/help-glpk