Multiple representations to compute orthogonal eigenvectors of symmetric tridiagonal matrices (Q1434380)

The authors present and analyze a new O\((nk)\) algorithm, \({\mathtt MR}^3\), for computing \(k\) eigenvectors of a real \(n\times n\) symmetric tridiagonal matrix. Sufficient conditions are established for the computed eigenvectors to be orthogonal, to working accuracy, and for the residuals to be small. The algorithm combines the use of the authors' earlier algorithm \textit{Getvec} [SIAM J. Matrix Anal. Appl. 25, No. 3, 858--899 (2004)] with a shift strategy, using several different shifts described by a representation tree. The shifts enable it to cope with clusters of close eigenvalues.