| Author | Message | ||
     Larafrith (larafrith) New member Username: larafrith Post Number: 1 Registered: 06-2005 |
Please, someone can tell me exactly what FlexPDE uses to solve the non-linear problem?, Newton-Raphson or Modified N-R iteration process?. And if it’s a modified N-R process, how to use a load increment process to help a nonlinear problem with a hardening curve find a solution. Thanks | ||
| Robert G. Nelson (rgnelson) Moderator Username: rgnelson Post Number: 376 Registered: 06-2003 |
In steady-state problems, FlexPDE uses a modified NR iteration with backtracking, similar to the one shown in Numerical Recipes. The computed step is applied, and the new residual computed. If the new position improves the residual by at least "NRSLOPE", it is accepted. If not, the step is cut back until a partial step satisfies the "NRSLOPE" criterion. The applied step must be at least a fraction "NRMINSTEP" of the full step. The process is iterated a maximum of "NEWTON" times. Default values are NRSLOPE=0.0001, NRMINSTEP=0.09, NEWTON=(2/CHANGELIM)+20. If you are having trouble converging, try NRSLOPE=0.1, NRMINSTEP=0.009, NEWTON=100. In time-dependent problems, none of this is done, and the full Newton step is taken without analysis. The iteration limit is NRUPDATE, with a default of 3. The assumption is that in a time-dependent problem, the changes over a step should be small and the overhead of backtracking would be wasted. |