The method of fictitious extrema localization in the problem of global optimization

The problem of finding the global extremum of a non-negative function on a positive parallelepiped in $n$-dimensional Euclidean space is considered. A method of fictitious extrema localization in a bounded area near the origin is proposed, which allows to separate the global extremum point from fictitious extrema by discarding it at a significant distance from the localization set of fictitious minima. At the same time, due to the choice of the starting point in the gradient descent method, it is possible to justify the convergence of the iterative sequence to the global extremum of the minimized function.