RECOMPRESSION TECHNIQUESFOR ADAPTIVE CROSS APPROXIMATION

摘要

The adaptive cross approximation method(ACA) generates low-rank approximations to suitable in x nsub-blocks of discrete integral formulations of elliptic bound-ary value problems. A characteristic property is that the ap-proximation, which requires k(m+n), k~|logε|~*,units ofstorage, is generated in an adaptive and purely algebraic man-ner using only few of the matrix entries. In this article wepresent further recompression techniques which are based onACA and bring the required amount of storage down to sub-linear order kk', where k' depends logarithmically on the ac-curacy of the approximation but is independent of the matrixsize. The additional compression is due to a certain smooth-ness of the vectors generated by ACA.