Hi, currently we are trying to use the eigen to solve a linear equation Ax = b. Where A is a sparse symmetric matrix with dimension up to 1M x 1M. Thanks to eigen, we were able to perform this analysis using

Code: Select all

    SparseLU<SparseMatrix<double>, COLAMDOrdering<int> > solverA;
    VectorXd X = solverA.compute(A).solveB;

However, what we found is that for any programme instance, the matrix A usually remains the same. Therefore it will help to improve the performance of the programme by skipping the step of re-computing the solverA. The problem now is that although there were a lot of information as to how to serialize the matrix A, there are little to no information as to how to store this solverA.

Is there anyway to do that? Or is that impossible?

Thank you

This is currently not really possible because the internal representation of the LU factors is not a standard SparseMatrix and it is subject to evolve in the future. To this end, you would have to hack the SparseLU class itself to get access to the internal representation and add the respective save/load functions.