In the current Davidson implementation, all the pair of eigenvalues/eigenvectors are constantly optimized until every pair converge. It would be desirable to deflate the pairs that are already converge and dynamically changed the size of the block for the next iterations.
See this paper
In the current Davidson implementation, all the pair of eigenvalues/eigenvectors are constantly optimized until every pair converge. It would be desirable to deflate the pairs that are already converge and dynamically changed the size of the block for the next iterations. See this paper