An algorithm for the generalized singular value decomposition on massively parallel computers (Q2365571)

A new algorithm for the generalized singular value decomposition of type \(U^ TAX=D_ A\) and \(V^ TBX=D_ B\), arising e.g. in the solution of linear least squares with equality constraints, is presented. It combines the advantages of algorithms of \textit{C. C. Paige} [SIAM J. Sci. Stat. Comput. 7, 1126-1146 (1986; Zbl 0621.65030)] and of \textit{V. Hari} and \textit{K. Veselić} [SIAM J. Sci. Stat. Comput. 8, 741-754 (1987; Zbl 0627.65039)] by the proper selection of the elementary \(2\times 2\) Jacobi rotations. The parallelization on a \(CM-2\) is discussed and comparisons to the algorithm of Paige [loc. cit.] are presented.