Improved Hessian approximations for the limited memory BFGS method (Q1964059)

The paper presents simple modifications over the limited memory BFGS (L-BFGS) method for large scale optimization. It proposes algorithms based on employing certain extra updates for the L-BFGS method without increasing the number of stored vector pairs. Numerical experiments show that the new algorithms yield applicable improvement over the L-BFGS method. The paper gives a good perspective for the limited memory BFGS method. Anyone interested in minimizing the objective function will find this paper stimulating and worthwhile useful. The paper is written in a thoughtful and informative way.