Strong representation of an adaptive stochastic approximation procedure (Q1819871)

The author considers a rather general one-dimensional stochastic approximation algorithm, including the possibility of random step length. Without assuming martingale properties for the random noise he obtains a strong representation by weighted averages of the error terms. Applying known laws of large numbers, central limit theorems etc. to these weighted averages, he may derive the corresponding results for the stochastic algorithm itself. The author is able to apply his representation to Venter's adaptive procedure in the case where the noise constitutes a martingale difference sequence as well as in the case of weakly dependent noise. This work continues search of the reviewer [Ann. Probab. 5, 954--965 (1977; Zbl 0374.62082)] and \textit{D. Ruppert} [ibid. 10, 178--187 (1982; Zbl 0485.62083)].