Dear all,

I'm using SimplicialLLT in 3.1.0.alpha2. In my sparse cholesky factor I have a large number of very small entries and hence I need to get the factor explicitly and delete close-to-zero entries. I also need to be able to reconstruct the approximation to my original sparse matrix using the sparsified Cholesky factor. However I can't seem to reconstruct my original matrix A from the Cholesky factors.

If I do something like

Code: Select all

SimplicialLLt <SparseMatrix<double> > solver;
solver.compute(A);
MatrixXd temp = (MatrixXd) (solver.matrixL() * solver.matrixL().transpose());
MatrixXd A_tilde = solver.permutationPinv() * temp * solver.permutationP();


Then A_tilde does not have the same entries as A; however it looks like I have an ordering issue since many of the entries of A and A_tilde are the same.

Am I missing something regarding the permutation matrices? I would expect the above should work assuming that the factorization is PAP' = LL'.

Best,
Jens

the permutations are inverted, it's A = P L L' P'

I was sure I'd tried the permutations both ways but obviously not ...

Thanks!