[Top][All Lists]
[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Re: [Help-glpk] Question about the behavior of lpx_exact
From: |
Andrew Makhorin |
Subject: |
Re: [Help-glpk] Question about the behavior of lpx_exact |
Date: |
Thu, 4 Oct 2007 11:35:39 +0400 |
> I'm using GLPK to solve a relatively large linear program. Using doubles
> (lpx_solve) I get what seems to be a reasonable answer in an hour or 2.
>
> I would like to verify that the solution is exactly right, so I instead
> use lpx_exact to solve the same linear program. I realize it will take
> longer, but how much longer should I expect it to take?
>
> The program has been running for 38 hours, and repeatedly output lines
> that look like the following:
>
> *203649: objval = 0.111111111111111 (8895)
>
> Is this an indication that the solution found using lpx_solve is wrong
> and only terminated due to numerical error, or have I just not waited
> long enough?
Using the exact solver to solve lp from the very beginning is not
a good idea. It is better to call it *after* the floating-point solver
has found optimal solution.
However, to check if the solution obtained by glp_simplex is accurate
enough you can use the api routine lpx_check_kkt.