We study the problem of optimizing a dynamic stochastic state-linear and control-nonlinear system operating on an infinite time horizon with respect to a quadratic cost functional. Several algorithms for the successive improvement of a given time-varying program control are proposed. They can be used, in particular, for designing high-performance dynamic incomplete-feedback controllers in deterministic linear time-invariant systems. A number of rigorous statements regarding the original optimization problem are stated and proved to justify the improvement algorithms.