Discrete least-squares technique for eigenvalues. I: The one-dimensional case (Q1365536)

A discrete least squares technique is proposed for computing eigenvalues of linear ordinary differential equations. The desired eigenvalues are obtained by solving a quadratic matrix eigenvalue problem resulting from the least squares technique. The latter eigenvalue problem is transformed by Frobenius linearization, and standard routines are applied to find initial approximations, which are then improved using the Gauss-Newton method. The efficiency and flexibility of the proposed method is demonstrated by treating simple model problems of second and forth order, among them two singular eigenvalue problems.