[aerosayan]

Hello everyone,

I'm trying to implement this implicit solver : https://doi.org/10.1016/j.jcp.2018.04.005

They use an approximate Jacobian to linearize the system, then relax it using 10 inner iterations of a Gauss-Seidel solver.

Essentially it is a Newton's method based solver, and in the inner iterations, the solver takes small steps following the gradient/direction prescribed in the Jacobian

Since it is not required to completely solve the linear system in one step, a simple Gauss-Seidel solver seems to be fast enough for the inner iterations.

In future I will use PETSc, but for now I'm trying to use Eigen 3.

However, Eigen 3 does not have a sparse Gauss-Seidel solver, so I'm not sure which solver I should use for the inner iterations.

Eigen has Sparse LU, Sparse QR, Conjugate Gradient, Least Square Conjugate Gradient, BiCGSTAB, and few other sparse solvers : https://eigen.tuxfamily.org/dox/grou...seSystems.html

Can we use any of these solvers instead of Gauss-Seidel for the inner iterations?

Thanks.