Hi,
I'm using SimplicialLDLt to solve a sparse linear system, but I'm not getting right results. My code looks like the following excerpt:

Code: Select all

   MappedSparseMatrix<double, RowMajor> mat(dim, dim, nnz_lin * dim, rowPtr, colInd, nonZero);

   Matrix<double, Dynamic, 1> b(dim), x(dim);
   for (int i = 0; i < dim; i++)
      b(i) = 1;

   SimplicialLDLt<SparseMatrix<double>> solver;
   solver.compute(mat);
   x = solver.solve(b);

I'm converting the sparse matrix to dense and printing it, and the values are fine. What am I doing wrong?

How is mat? is it really selfadjoint? does it store all coefficients? only the lower half? only the upper half? Then you should check the success of each step:
solver.compute(mat);
if(!solver.info()==Success) {...}
x = solver.solve(b);
if(!solver.info()==Success) {...}

The test matrix I'm using is the following:

Code: Select all

mat =
     0     1     2     0     0
     0     3     4     5     0
     0     0     6     7     8
     9     0     0    10    11
    12    13     0     0    14

It's being solved fine by Matlab and by a SuperLU routine that I've written some time ago. The solver.info() is ok too.

this matrix is not symmetric, so you cannot solve it with a Cholesky factorization.

Yes, that's right.
Now I'm trying to solve it with the BiCGSTAB method, but I'm getting the following error:

Unhandled exception at 0x0041751f in eigen_test.exe: 0xC0000005: Access violation reading location 0x00846838.

In the following line, when I call solver.solve(b):

Code: Select all

else if(RhsIsVector)              dest.coeffRef(it.index()) += it.value() * rhs_j;


in SparseDenseProduct.h

My code looks like the following:

Code: Select all

   colInd = new int[nnz_lin * dim];
   rowPtr = new int[dim + 1];
   nonZero = new double[nnz_lin * dim];
   GenSquareMatrix(dim, nnz_lin, nonZero, rowPtr, colInd);
   PrintMatrix(dim, nnz_lin, nonZero, rowPtr, colInd);

   MappedSparseMatrix<double, ColMajor> mat(dim, dim, nnz_lin * dim, rowPtr, colInd, nonZero);
   MatrixXd m = mat.toDense();
   //Print it and check if the values are ok

   VectorXd b(dim), x(dim);
   for (int i = 0; i < dim; i++)
      b(i) = 1;

   BiCGSTAB<SparseMatrix<double>> solver;
   solver.compute(mat);
   if(!solver.info()==Success) {cout << "Error on compute(mat)";}
   x = solver.solve(b);
   if(!solver.info()==Success) {cout << "Error on solve(b)";}

Thanks a lot,
Guilherme.