1: \begin{abstract}\noindent

2:   Based on continuously recorded beam positions and corrector excitations from,

3:   for example, a closed-orbit feedback system we describe an algorithm that

4:   continuously updates an estimate of the orbit response matrix. The speed of

5:   convergence can be increased by adding very small perturbations, so-called

6:   dither, to the corrector excitations. Estimates for the rate of convergence

7:   and the asymptotically achievable accuracies are provided.

8: \end{abstract}

9: