On Sun, Mar 31, 2013 at 4:39 PM, Ehsan Eftekhari <[email protected]>wrote:
> Dear list, > > I'm trying to solve the density driven natural convection due to the > dissolution of CO2 in water in an aquifer. Please see the attached script. > The script works fine for a Cartesian grid. However, it does not work on a > cylindrical grid. > See http://pastebin.com/e12MRbbd, I shifted the origin away from zero and it runs with an iterative solve (not LU). If you have Trilinos installed, run with "--trilinos" and that will use GMRES. If you don't have Triilinos, it works with the scipy solvers as well. So use either "--trilinos" or "--scipy", not "--pysparse". I installed fipy from the 'develop' branch, or at least I think I did, and > the problem still exist. > My other issue is that all my boundary conditions for the continuity > (pressure) equation are Neumann, which makes the Matrix of coefficient ill > conditioned. Is there a way to fix the value of pressure (variable p) at a > point? > See the script. I've fixed the corner point using a TransientTerm with a zero coefficient everywhere but the corner point. It doesn't seem to make much difference to the solution and there is weird yellow mark in the bottom left corner of the pressure plot although the value appears to be correct and stationary. > How can I check the condition number of the matrix of coefficients? > You can get more info about iterations and convergence if you switch on "FIPY_VERBOSE_SOLVER" in your environment. For example, with Trilinos it does http://matforge.org/fipy/browser/fipy/fipy/solvers/trilinos/trilinosAztecOOSolver.py?rev=fe00adaf9e4bc545ec5894e384213b3230e9dc2b#L89 There is an "AztecOO.AZ_condnum" option that isn't included, but which you could include. You probably need to pass in something telling the solver to record the condition number using "SetAztecOption" to get it to work. Hope this helps. -- Daniel Wheeler
_______________________________________________ fipy mailing list [email protected] http://www.ctcms.nist.gov/fipy [ NIST internal ONLY: https://email.nist.gov/mailman/listinfo/fipy ]
