ON SOLVING GENERALIZED EIGENVALUE PROBLEMS USING MATLAB

Matlab is a high-level computing environment that is rapidly gaining p opularity for the execution of matrix computations. This paper discuss es a major inconvenience that may arise if Matlab is used for the nume rical solution of generalized eigenvalue problems. Matlab normalizes t he eigenvectors to unit length (Euclidean norm). This leaves the eigen vectors fixed, up to a complex multiplier of modulus one. As a result additional programming may be necessary to eliminate spurious complex numbers. Specifically, problems recently described with respect to the eigensolution obtained by the generalized rank annihilation method ar e shown to be caused by Matlab's implementation of the generalized eig envalue problem. (C) 1997 by John Wiley & Sons, Ltd.