Re-Compressed H-Matrices for Fast Electric Field Integral Equation

In this paper, an improved Re-Compressed H-Matrix is presented. Improvement with respect to accuracy and memory saving is shown compared to conventional H-Matrix. Improvement is obtained by extracting the orthogonal components of the compressed matrices for higher error tolerance and re-compressing it with Singular Value Decomposition (SVD) for the desired tolerance. The method maintains the compression cost of r(2) (m + n) , storage and matrix-vector product cost of r (m + n) for a low-rank matrix block of rank r and size m x n.