A class of efficient preconditioners with multilevel sequentially semiseparable matrix structure

This paper presents a class of preconditioners for sparse systems arising from discretized partial differential equations (PDEs). In this class of preconditioners, we exploit the multilevel sequentially semiseparable (MSSS) structure of the system matrix. The off-diagonal blocks of MSSS matrices are of low-rank, which enables fast computations of linear complexity. In order to keep the low-rank property of the off-diagonal blocks, model reduction algorithm is necessary. We tested our preconditioners for 2D convection-diffusion equation, the computational results show the excellent performance of this approach.