1: \begin{abstract}

2: Among first order optimization methods, Polyak's heavy

3: ball method has long been known to guarantee the asymptotic rate of

4: convergence matching Nesterov's lower bound for functions

5: defined in an infinite-dimensional space. In this paper, we use results on

6: the robust gain margin of linear uncertain feedback control systems to show that the

7: heavy ball method is provably worst-case asymptotically optimal when

8: applied to quadratic functions in a finite dimensional space.  

9: 

10: \medskip

11: \noindent\textbf{Keywords:} 

12: Optimization methods; the heavy ball method; analysis of systems with

13: uncertainties. 

14: \end{abstract}

15: