This leads to the (stochastic counterpart) approximating problem points for generated realizations (sample paths) wis required. Drawbacks of the SA method are well known -slow convergence, absence of a good stopping rule, difficulty in handling constraints. In this talk we discuss an alternative approach to the above optimization problem which became known as the stochastic counterpart (or sample path) method. In the stochastic counterpart method a (large) sample WI, ... , W n is generated and

more »

... he function f(·) is approximated by the corresponding average function ABSTRACT In this talk we consider a problem of optimizing an expected value function by Monte Carlo simult ion methods. We discuss, somewhat in details, the stochastic counterpart (sample path) method where a relatively large sample is generated and the expected value function is approximated by the corresponding average function. Consequently the obtained approximation problem is solved by deterministic methods of nonlinear programming. One of advantages of this approach, compared with the classical stochastic approximation method, is that a statistical inference can be incorporated into optimization algorithms. This allows to develop a validation analysis, stopping rules and variance reduction techniques which in some cases considerably enhance numerical performance of the stochastic counterpart method. minln(x), xeS