Formal analysis for practical gain sequence selection in recursive stochastic approximation algorithms

For many popular stochastic approximation algorithms, such as simultaneous perturbation stochastic approximation method and stochastic gradient method, the practical gain sequence selections are different from the optimal selection, which is theoretically derived from asymptotically performance. We provide formal justification for the reasons why we choose such gain sequence in practice.