Numerical Stability and Instability in Matrix Sign Function Based Algorithms

r s The matrix sign function is the basis of a family of algorithms fo olving a variety of invariant subspace related problems. Rounding t s errors corrupting the computation corrupt the calculated invarian ubspaces. We present a forward and a backward error analysis for. P the invariant subspaces calculated through the matrix sign function roper scaling is essential for numerical stability as well as for rapid x s convergence. We identify some problems upon which the matri ign function algorithms may be expected to perform well and others