1: \begin{abstract}

2: In this work we propose a differential geometric motivation for Nesterov's

3: accelerated gradient method (AGM) for strongly-convex problems. By

4: considering the optimization procedure as occurring on a Riemannian

5: manifold with a natural structure, The AGM method can be seen as the

6: proximal point method applied in this curved space. This viewpoint

7: can also be extended to the continuous time case, where the accelerated

8: gradient method arises from the natural block-implicit Euler discretization

9: of an ODE on the manifold. We provide an analysis of the convergence

10: rate of this ODE for quadratic objectives.

11: \end{abstract}

12: