Access old solution(s) in MatrixFreeOperators::Base::apply_add()

[Maxi Miller]

I followed step-37 for creating a MatrixFreeOperator, but now I also would like to access a previous solution (which is needed during apply_add()), and stored using a LinearAlgebra::distributed::Vector. In step-48 there is an example which uses previous solutions, but here the solve-function is replaced, and therefore making it difficult to implement it in my problem.

Thus, what would be the most efficient way of giving the operator-function apply_add access to those solution vectors? I tried adding a class member and initializing it during setup_system(), but when accessing it via read_dof_values(), I had a layout mismatch, resulting in

The violated condition was:
    vec.partitioners_are_compatible(*dof_info.vector_partitioner)
Additional information:
    The parallel layout of the given vector is not compatible with the parallel partitioning in MatrixFree. Use MatrixFree::initialize_dof_vector to get a compatible vector.

Are there better ways?

Thanks!

[Daniel Arndt]

Maxi,

[...] I tried adding a class member and initializing it during setup_system(), but when accessing it via read_dof_values(), I had a layout mismatch, resulting in

The violated condition was:
    vec.partitioners_are_compatible(*dof_info.vector_partitioner)
Additional information:
    The parallel layout of the given vector is not compatible with the parallel partitioning in MatrixFree. Use MatrixFree::initialize_dof_vector to get a compatible vector.

Did you initialize its layout via MatrixFree::initialize_dof_vector() in fact?

Best,

Daniel

[Maxi Miller]

I found my mistake, I could not reuse the same partition for the main matrix and the MG-matrices, thus I had to add

        LinearAlgebra::distributed::Vector<float> local_solution;

        mg_matrices[level].initialize_dof_vector(local_solution);

//        local_solution = solution;

        mg_matrices[level].set_old_solution(local_solution);

in the loop while creating those matrices. My next problem, though, is how to fill the local_solution-vector based on the solution-vector. When uncommenting the line above, I get exactly the same error as before. Are there alternatives?