On Thu, Nov 6, 2008 at 2:08 PM, Angus Hendrick <[EMAIL PROTECTED]> wrote: > Gentlemen, > > Thanks for all your help so far. I have been using fipy Trunk successfully > with the development version of numpy (1.3.0dev5945). I get the occasional > warning about deprecated names, but everything seems to work fine. I've > also gotten PyTrilinos working and the Aztec00 solver with > MultiLevelPreconditioner is very fast.
Excellent! > I did a quick comparison using a > simple Poisson problem between fipy with Aztec00 and a stock Poisson matrix > builder solved with pyamg. The fipy code was only about 5% slower, and > given all the flexibility in fipy versus roll-your-own solutions, that's > really impressive. The Poisson problem will not necessarily capture the inefficiencies in fipy. The numerical operations required to evaluate the arrays that are used to populate the matrix are generally thought to be "slow" in fipy and the Poisson problem has very few of these evaluations so I wouldn't hold it up as an example of fipy's efficiency. If you add some extra terms with complicated coefficients to the equations that that will provide a better test. > The problem I'm having now is that I want to save my simulation results > using the h5 file format with pytables. I haven't tried using pytables. > I can extract all the problem > related arrays without much difficulty, but the boundary conditions are more > vexing since in the forms I can readily access, the locations where each > boundary condition is applied live as unresolved union and intersection of > boolean arrays buried inside the FixedValue or FixedFlux instances. Is > there an approved method for extracting and resolving this location > information? If not, is there an easy hack I can use? Can you clarify, Do you want to know the faces that a boundary condition is associated with and the locations of those faces given only the boundary condition object itself? -- Daniel Wheeler
