Abstract.A new method for quasi-Newton minimization outperforms BFGS by combining least-change updates of the Hessian with step sizes estimated from a Wishart model of uncertainty. The Hessian update is in the Broyden family but uses a negative parameter, outside the convex range, that is usually regarded as the safe zone for Broyden updates. Although full Newton steps based on this update tend to be too long, excellent performance is obtained with shorter steps estimated from the Wishart model. In numerical comparisons to BFGS the new statistical quasi-Newton (SQN) algorithm typically converges with about 25% fewer iterations, functions, and gradient evaluations on the top 1/3 hardest unconstrained problems in the CUTE library. Typical improvement on the 1/3 easiest problems is about 5%. The framework used to derive SQN provides a simple way to understand differences among various Broyden updates such as BFGS and DFP and shows that these methods do not preserve accuracy of the Hessian, in a certain sense, while the new method does. In fact, BFGS, DFP, and all other updates with nonnegative Broyden parameters tend to inflate Hessian estimates, and this accounts for their observed propensity to correct eigenvalues that are too small more readily than eigenvalues that are too large. Numerical results on three new test functions validate these conclusions.
Key words. BFGS, DFP, negative Broyden family, Wishart model
AMS subject classifications. 65K10, 90C53DOI. 10.1137/040614700 1. Introduction. Quasi-Newton methods for unconstrained optimization are important computational tools in many scientific fields and are a standard subject in textbooks on computation. The BFGS method, proposed individually in [6], [14], [20], and [30], is implemented in most optimization software and is widely recognized as efficient. Generalizations of BFGS are available for large problems with memory limitations, for problems with bound constraints, and for a parallel computing environment. In theoretical investigations BFGS is known as a special case of the Broyden class [5]. Some Broyden updates with negative Broyden parameters have been found to produce faster convergence than BFGS updates [31], [8] but, for various reasons, have not been widely adopted. Indeed, Byrd et. al. conclude that "practical algorithms that preserve the excellent properties of the BFGS method are difficult to design." Nocedal and Wright [29] state that "the BFGS formula. . . is presently considered to be the most effective of all quasi-Newton updating formulae." In our opinion, BFGS remains the most popular front-runner because of two important unanswered