Simulations of Computer, Telecommunications, Control and Social Systems

Gradient methods for large-scale minimization problems

I. G. Chernorutskiy

St. Petersburg State Polytechnical University

Abstract: Gradient methods with Chebyshev relaxation functions are developed. In contrast to the classical gradient procedures, the methods retain the convergence and efficiency for non-convex nonlinear programming problems under the conditions of high stiffness of target functionals and high dimension of the optimizedparameters vector.

Keywords: gradient methods, relaxation functions, non-convex problems, stiff functionals.

UDC: 681.3.06