An entropy-based aggregate method for minimax optimization

This paper presents a very efficient method, referred to as the aggregate method, for solving nonlinear minimax optimization problems, which converts a minimax problem to an unconstrained minimization on a differentiable aggregate function. This function is derived by means of Lagrangean duality and the Jaynes’ maximum entropy principle. It can be shown that as a parameter in the function tends to infinity, it approaches the maximum function that denotes a uniform approximation to the problem functions and is non-differentiable. Main features of the present method consist of its algorithmic simplicity and computational high efficiency.