Hierarchical matrix approximation with blockwise constraints (Q2376862)

A technique to preserve constraints on a hierarchical matrix approximation is presented. The preservation is done block-wise and it carries over to \(LU\) factorization. It is shown that a particular set of block-wise constraints leads to the preservation of a significantly larger set of global constraints due to the structure of the hierarchical matrix, while maintaining its logarithmic-linear complexity. The block-wise preservation of a small set of piecewise constant vectors leads to the global preservation of the linear space of piecewise constants, which lie in the vicinity of eigenvectors corresponding to small eigenvalues, and this guarantees spectral equivalence. It is pointed out that complexity is reduced compared to the usual matrix approximation with a similar preconditioning effect.