Scaling by binormalization (Q1430251)

The authors develop an algorithm (BIN) for diagonally scale a symmetric matrix by a fast iterative binormalization technique. The convergence properties of this algorithm and the possible relation to optimal conditioning of the matrix are investigated. Analogous algorithms, for nonsymmetric and rectangular matrices are also developed. The conclusion of the authors is that BIN is a novel iterative binormalization algorithm that may be useful in many applications. Numerical experiments are made for the BIN algorithm and the NBIN algorithm (extended of BIN to non-symmetric matrices).