Iterative gauss-seidel solver
Posted: Mon Nov 21, 2005 4:32 pm
Hi Guys
I'm working on a simulator where I use a iterative gauss-seidel method to solve for lagrange multipliers in my lcp problem.
This works quite well in most of the cases, but in some extreme cases I get a bad solution.
The problem is that my solver starts to cycle between solutions for the multipliers.
After my first iteration I might have a solution like
\lambda_1,1 , \lambda_1,2 , \lambda_1,3
and then
\lambda_2,1 , \lambda_2,2 , \lambda_2,3
and then another iteration where I get
\lambda_3,1 , \lambda_3,2 , \lambda_3,3.
Then the solver can start over and will find
\lambda_1,1 , \lambda_1,2 , \lambda_1,3
as a new solution. I guess it is because my problem somehow are illconditioned, but how should I solve this. Have any other experienced this behaviour with iterative gauss seidel solvers
Thanks
Niels
I'm working on a simulator where I use a iterative gauss-seidel method to solve for lagrange multipliers in my lcp problem.
This works quite well in most of the cases, but in some extreme cases I get a bad solution.
The problem is that my solver starts to cycle between solutions for the multipliers.
After my first iteration I might have a solution like
\lambda_1,1 , \lambda_1,2 , \lambda_1,3
and then
\lambda_2,1 , \lambda_2,2 , \lambda_2,3
and then another iteration where I get
\lambda_3,1 , \lambda_3,2 , \lambda_3,3.
Then the solver can start over and will find
\lambda_1,1 , \lambda_1,2 , \lambda_1,3
as a new solution. I guess it is because my problem somehow are illconditioned, but how should I solve this. Have any other experienced this behaviour with iterative gauss seidel solvers
Thanks
Niels