Recently, I tried to use Eigen 3.3.3 to calculate acoustic finite element problem, but I have a problem to use SimplicialLDLT solver to solve sparse matrix.
For the equation Ax=v, A is the sparse matrix, v is the sparse vector.
1. The code of LU solver is:

Code: Select all

 Eigen::SparseLU<Eigen::SparseMatrix<std::complex<double>>> solverLU;
                  solverLU.compute(matA);
                  vecX = solverLU.solve(vecY);


The value of LU solver info is zero, the results of LU agree well with results of SkyLine solver coded by Fortran.

2. The code of SimplicialLDLT solver is:

Code: Select all

 Eigen::SimplicialLDLT<Eigen::SparseMatrix<std::complex<double>>> solverLDLT;
                  solverLDLT.compute(matA);
                  vecX = solverLDLT.solve(vecY);


LDLT solver info is zero, but the results of the LDLT solver is inconsistent with expectations.

3. Then I try to use LDLT with upper triangular matrix, and the code is:

Code: Select all

 Eigen::SimplicialLDLT<Eigen::SparseMatrix<std::complex<double>>, Eigen::Upper > solverLDLT;
                  solverLDLT.compute(matA);
                  vecX = solverLDLT.solve(vecY);


LDLT solver info is zero and the results of method 2 and 3 are the same.

Ps: Sparse Matrix A is symmetrical can be identified.
By the way, how to inquire the number of nonezero elements in a sparse matrix? how to install suitesparse with Qt in windows?
Thanks

Your code looks fine, but LDLT or LLT assumes that your matrix is self-adjoint (aka hermitian), that is not the same as symmetric for complexes:

symmetric: A == A.transpose()
selfadjoint: A == A.transpose().conjugate() == A.adjoint()


For the number of non zeros: A.nonZeros()

ggael wrote:Your code looks fine, but LDLT or LLT assumes that your matrix is self-adjoint (aka hermitian), that is not the same as symmetric for complexes:

symmetric: A == A.transpose()
selfadjoint: A == A.transpose().conjugate() == A.adjoint()


For the number of non zeros: A.nonZeros()

Get it, thanks

ggael wrote:Your code looks fine, but LDLT or LLT assumes that your matrix is self-adjoint (aka hermitian), that is not the same as symmetric for complexes:

symmetric: A == A.transpose()
selfadjoint: A == A.transpose().conjugate() == A.adjoint()


For the number of non zeros: A.nonZeros()

Is there a way to copy the upper triangular part of a sparse martix to the lower part by calling function?