Some models of continuous-time stochastic approximation (Q1816014)

Consider the problem of finding the zero of the function \(f(x)\) given that the function is observed in a mixture with random noise, and \(dy(t,x)= f(x)+\sigma(t,x)d\xi(t)\), where \(\sigma(t)\) is a continuous square integrable martingale. To find the root of the function \(f(x)\), consider the equation \[ dx(t)= {w(t)\over t} (f(x(t))dt+ \sigma(t,x(t))d\xi(t)),\quad x(1)=x,\tag{\(*\)} \] where \(\sigma(t,x)\) is a nonstochastic function, \(w(t)\) is some \(F_t\)-measurable stochastic process which has a limit with probability as \(t\to\infty\); \(\{F_t, t\geq 0\}\) is a flow of \(\sigma\)-algebras. In addition to the case of a general observed process, the author examines some other issues associated with the weak convergence of the processes generated by procedures of the form \((*)\).