The admissible block [[bar.F].sub.txs] generated from the LF-MLFMA has two features: 1) The admissible blocks generated from the same source cluster s share an aggregation matrix [[bar.[beta]].sub.s], and those generated from the same observer cluster t share a disaggregation matrix [[bar.[beta]].sub.t].
Because of the aforementioned features, the admissible blocks of [[bar.V].sub.H] and [[bar.P].sub.H] can be produced by LF-MLFMA efficiently.
Although the admissible blocks of [[bar.V].sub.H] and [[bar.P].sub.H] are compressed to Rk-matrices by LF-MLFMA, as shown in Eq.
For all the testing, the generalized minimum residual method (GMRES) is adopted for the iterative solvers, the number of multipoles in the LF-MLFMA is chosen as P = 5, and the recompression accuracy is set to be [[epsilon].sub.rec] = [10.sup.-4].
Here, a two-level LF-MLFMA is used to construct H-matrices.
The H-matrix representation of the A-EFIE system matrix is constructed by a 5-level LF-MLFMA. The H-LU preconditioner and the popular constraint preconditioner  are compared, and the iterative convergence curves are plotted in Fig.
The 5-level LF-MLFMA is employed to produce the H-matrix.