1: \begin{abstract}

2: It is shown the almost sure convergence and asymptotical normality of a generalization 

3: of Kesten's stochastic approximation algorithm for

4: multidimensional case.

5: 

6: In this generalization, the step  increases or decreases 

7: if the scalar product of two subsequente increments 

8: of the estimates is positive or negative.

9: 

10: This rule is intended to accelerate the entrance in the 

11: `stochastic behaviour' when initial conditions 

12: cause the algorithm to behave in a `deterministic fashion' for the

13: starting iterations.

14: \end{abstract}

15: