Newton's Method for a Generalized Inverse Eigenvalue Problem

From MaRDI portal





A family of matrices \(A(c)\) and \(B(c)\) dependent on a vector \(c=(c_1,\dots,c_n)\in \Omega \subset \mathbb R^n\) is introduced, \(A(c)=A_0+\sum_{k=1}^n c_kA_k\), \(B(c)=B_0+\sum_{k=1}^n c_kB_k\), \(B(c)>0\), where \(A_k,B_k\), \(k=0,\dots n\), are given real symmetric matrices. The solutions of the generalized eigenvalue problem \(A(c)x(c)=\widetilde \lambda(c)B(c)x(c)\) depend on \(c\) and form an ordered \(n\)-tuple \(\lambda(c)\). For a given vector \(\widehat \lambda =(\lambda_1,\dots,\lambda_n)\), the vector-valued function \(f(c)=\lambda(c)-\widehat \lambda\) is defined. Then the generalized inverse eigenvalue problem (GIEP) reads: Find \(c^*\in \Omega\) such that \(f(c^*)=0\). Distinct and multiple eigenvalue problems are taken into account separately. In the latter case, the authors show how to state the problem so that it is not overdetermined. Two Newton-type algorithms for solving the GIEP are presented. It is proved that, under some assumptions, they have locally quadratic convergence. Numerical examples illustrate the theory which, as the authors emphasize, extends the results achieved by \textit{S. Friedland}, \textit{J. Nocedal} and \textit{M. L. Overton} [SIAM J. Numer. Anal. 24, 634-667 (1987; Zbl 0622.65030)] and \textit{J. Nocedal} and \textit{M. L. Overton} [Lect. Notes Math. 1005, 212-225, Springer-Verlag, Berlin (1983; Zbl 0525.65020)].




Cited in
(28)








This page was built for publication: Newton's Method for a Generalized Inverse Eigenvalue Problem

Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q4383416)