On one way то design randomised algorithms

A probabilistic approach is developed to minimizing a scalar function of a vectorvalued argument. Convergence conditions of randomised algorithms and a way to incorporate prior information on the zome of global extremum of the goal function are described.