Standard quasi-Newton methods used in solving optimization
problems and nonlinear systems of equations only use data
from the step just completed in updating the approximation
to the Hessian / Jacobian matrix. In this sense, they may be termed
"single-step" methods. This project involves
the development, implementation and testing of whole classes
of new methods which utilize data from several recent steps
in carrying out the updating (and, thus, may be considered to be
"multi-step" methods). Considerable improvements in
computational efficiency have been obtained in the case of
optimization (some of the technical reports written about
these developments are available on the
Papers page).
Copies of other publications (both external and internal) are
available from the author (e-mail address below).