On the numerical stability and accuracy of the conventional recursive least squares algorithm

We study the nonlinear round-off error accumulation system of the conventional recursive least squares algorithm, and we derive bounds for the relative precision of the computations in terms of the conditioning of the problem and the exponential forgetting factor, which guarantee the numerical stability of the finite-precision implementation of the algorithm; the positive definiteness of the finite-precision inverse data covariance matrix is also guaranteed. Bounds for the accumulated round-off errors in the inverse data covariance matrix are also derived. In our simulations, the measured accumulated roundoffs satisfied, in steady state, the analytically predicted bounds. We consider the phenomenon of explosive divergence using a simplified approach; we identify the situations that are likely to lead to this phenomenon; simulations confirm our findings