We consider using seed projection methods for solving unsymmetric shifted systems with multiple right-hand sides (A - σjI)x^(j) = b^(j) for 1 ≤ j ≤ p. The methods use a single Krylov subspace corresponding to a seed system as a generator of approximations to the nonseed systems. The residual evaluates of the methods are given. Finally, numerical results are reported to illustrate the effectiveness of the methods.
In this paper, a generalized block Broyden’s method is presented for solvinga collection of overdetermined equations. We have proven that for the p linear overdetermined equations with a m×n coefficient matrix, the method is terminated with the p least squared solutions after 2m/p steps at most, and two numerical examples are given.
In 1994, O’leary and Yeremin extended the quasi-Newton method for minimizing a collection of functions with a common Hessian matrix to the block version,and discussed some algebraic properties of this block quasi-Newton method. In thispaper, we derive compact representations of the block BFGS’s updating matrices.These representations allow us to efficiently implement limited memory methods,e.g., the limited memory BFGS method, for minimizing a collection of functionswith a common Hessian matrix. The method relieves the requirement for the storage counts and has the savings in the operation counts, in particular, for large scaleproblems. The numerical experiments for the multiple unconstrained optimizationproblems show that the method works efficiently. Compared with O’Leary’s multiple version of BFGS method, our multiple version of the limited memory BFGSmethod is more efficient in the total operation counts and the storage counts.
