Shor's r-algorithm is an iterative method for unconstrained optimization, designed for minimizing nonsmooth functions, for which its reported success has been considerable. Although some limited convergence results are known, nothing seems to be known about the algorithm's rate of convergence, even in the smooth case. We study how the method behaves on convex quadratics, proving linear convergence in the two-dimensional case and conjecturing that the algorithm is always linearly convergent, with an asymptotic convergence rate that is independent of the conditioning of the quadratic being minimized.
Submitted to IMA J. Numer. Anal.