From: Ahmed EL-S. <els...@mc...> - 2004-06-04 04:08:12
|
The problem with the number of unknowns appears for the elements on the boundary.. linear chain of quad elements is simple example. As you said for linear and quadratic reconstructions it is some how ok but if the initial mesh is bad there might be a problem. > Eventually I'd like the library to look for LAPACK in ./configure and > then use it internal to the DenseMatrix<> and DenseVector<> class. In > the meantime, if you'd like to "share" an SVD implementation we'd be > happy to include it! I would recommend to adopt an SVD implementation like the LU one so if you run it in parallel it works too. I will attach the code which I have it is the TNT implementation of the SVD. I modified it to work with the DenseMatrix<> and DenseVector<>. I call it from the error estimator like //---------------------------- SVD<double> M_svd(M); M_svd.solve(bx, solx); M_svd.solve(by, soly); //---------------------------- BUT this code may loop for ever. I am checking how to put a stopping criteria if there is a convergence problem. Ahmed > > -Ben > > Ahmed EL-Sheikh wrote: > > That is great.. I have already developed a similar code :-( > > I started with the same code but I added some modifications > > > > The first is the need to use a local coordinate system for the > > patch to avoid the different weighting of x and y. > > So I did something like this: > > > > // determine the x_min, x_max, y_min, y_max , z_min, z_max for all the points on the > > patch > > // Scaled x and y by z by > > // x=(x-x_min)/(x_max-x_min); > > > > > > The other thing is using Singular value decomposition to solve. > > I adopted a code from TNT for that. > > > > Do you have any plans for different error estimators .... > > > > > > John Peterson wrote: > > > > > >>Note that your technique will only add *face* neighbors of the > >>current element to the patch, i.e. you will not in general include > >>the full support of each of the nodal basis functions in your > >>patch. > > > > > > I new that .. It is a bit problem here, > > What I want is a patch ... with the same system of equation > > as well as the current solution. This patch may use the same assemble function > > but on the residuals or attach its own assembling function. > > > > > > One choice is use these faces I build a map from the patch mesh to > > the original mesh ... then initialize a different equation system for the > > patch .. obtain what is needed from the old mesh when needed. > > Attach another assembling function for the error estimation with > > different boundary conditions. Then instead of solving it using the method > > solve use SVD. > > > > Is that make sense !! > > > > Ahmed > > > > > > > > ------------------------------------------------------- > > This SF.Net email is sponsored by the new InstallShield X. > >>From Windows to Linux, servers to mobile, InstallShield X is the one > > installation-authoring solution that does it all. Learn more and > > evaluate today! http://www.installshield.com/Dev2Dev/0504 > > _______________________________________________ > > Libmesh-devel mailing list > > Lib...@li... > > https://lists.sourceforge.net/lists/listinfo/libmesh-devel |