Constructing the physical parameters of a damped vibrating system from eigendata (Q2465325)

The reconstruction of the quadratic pencil \(P(\lambda) = \lambda^2 M + \lambda C + K\) from given eigenvalue/eigenvector data of \(P(\lambda) x = 0\) is considered. Hereby, \(M\), \(C\), and \(K\) are tridiagonal matrices. At first, the following two problems are discussed: Construct the pencil from two real eigenvalues and three real eigenvectors or from one real eigenvector and two complex conjugate eigenpairs. Procedures for solving both problems are presented and solvability conditions are given. Then, the author considers the case, where \(k\) eigenpairs are given. Depending on \(k\) the problem is under-determined or over-determined. Necessary and sufficient conditions on the two given real or complex eigenpairs are presented such that the under-determined system has a physical solution. An algorithm for the reconstruction of the pencil is described. Furthermore, the solution of the under- and over-determined systems in the case of noisy data is discussed. To reduce the sensitivity of the reconstruction to perturbations a least-squares approach is proposed in the case of the over-determined system. In the case of the under-determined problem a minimum norm solution is computed. A regularization method for finding a stable and physically feasible solution is presented. Finally, numerical experiments demonstrate the effectiveness of the proposed methods.