On Fri, Nov 21, 2014 at 4:01 PM, Raymond Smith <[email protected]> wrote:
> Well, I think I spoke a bit too soon. By doing the transient version of > the problem and by switching to GMRESSolver, I was able to approach steady > state for considerably higher applied voltages (about an order of magnitude > higher), which is considerably better and getting quite close to full salt > depletion on one side. > It's always worth checking if the problems are with the linear or non-linear solver. There is a clear separation of the linear and non-linear in FiPy. I still have trouble sweeping to convergence on the time steps for large > applied voltages, but perhaps by taking smaller and smaller time steps I'll > be able to continue to make progress. Tips for solving the steady state > directly (apart from simply beginning with great initial guesses, which > isn't feasible for me once I move to 2D and have more complicated boundary > conditions) are certainly still welcome, but I think this at least gives me > a way to move forward. > Another approach would be Newton steps instead of Picard or both together, Picard first and then Newton. Obviously, FiPy doesn't do anything to help you do this, but you can reformulate the equations to use Newton. -- Daniel Wheeler
_______________________________________________ fipy mailing list [email protected] http://www.ctcms.nist.gov/fipy [ NIST internal ONLY: https://email.nist.gov/mailman/listinfo/fipy ]
