Abstract:

In this paper, we propose a new family of continuous-time optimization algorithms for time-varying, locally strongly convex cost functions, based on discontinuous second-order gradient optimization flows with provable finite-time convergence to local optima. To analyze our flows, we first extend a well-know Lyapunov inequality condition for finite-time stability, to the case of arbitrary time-varying differential inclusions, particularly of the Filippov type. We then prove the convergence of our proposed flows in finite time. We illustrate the performance of our proposed flows on a quadratic cost function to track a decaying sinusoid.