Iterative Methods for Large Linear and Nonlinear Least Squares Problems

摘要

Under a grant to 8 graduate students the most exciting research accomplishment is a new trust region approach to global convergence for nonlinear programming problems. Testing has also begun on a variable metric variant of the Kamarkar linear programming algorithm that could be of great practical significance if very preliminary tests are any indication. Other interesting work has been a unified convergence analysis for the many variants of the conjugate gradient method, a convergence analysis of the popular Nelder-Mead algorithm, a novel use of interactive computer graphics to obtain user preferences in multi-objective optimization, a convergence analysis of the EM algorithm for mixture density estimation, and a survey of all the work done by researchers in various fields on nonlinear programming problems in which some subset of the variables always appear linearly.