Sufficient and necessary condition for the convergence of stochastic approximation algorithms

We present a sufficient and necessary condition for the convergence of stochastic approximation algorithms, which were proposed 50 years ago, have been widely applied to various areas and intensively investigated in theory. In the literature, only various sufficient conditions are known. The obtained condition is simple and has a clear physical meaning.