Recently, particular attention has been paid attention to memoryless quasi-Newton methods for solving unconstrained optimization problems. Because memoryless quasi-Newton methods do not need the storage of memories for matrix and their computing cost par a iteration is low, the methods are efficient to large-scale unconstrained optimization problems. Moreover, since the methods are closely related to not only quasi-Newton methods but also nonlinear conjugate gradient methods and nonlinear three-term conjugate gradient method, it is expected that the methods are promising. This paper introduces recent studies on memoryless quasi-Newton methods.
This paper deals with kernel-based collocation methods mainly for nonlinear parabolic partial differential equations, with a special emphasis on rigorous convergence issues. In doing so, the interpolation theory with condition-ally positive definite kernels is briefly reviewed, and a variant of Barles-Souganidis viscosity solution methods is discussed.