We consider a repeated QR updating algorithm for the solutionof equality constrained linear least squares problems. The constrainedproblem is first converted into the linear least squares problem usingweighted factor and then it is partitioned into a small well-manageableproblem by removing a pair of blocks of rows and columns. We performthe QR factorization of the small subproblem and then it is updated byappending the removed data. The proposed strategy is effective for largescale dense problems and also particulary suitable for parallel implementationdue to its partitioning by using the number of passes. Some numericalexperiments are given to illustrate the accuracy of the proposedalgorithm and the results are compared with the solution obtained throughthe nullspace method.
展开▼
