1: \begin{abstract} 

2: In a Hilbert space setting $\cH$, for convex optimization, we analyze the fast convergence properties as $t \rightarrow +\infty$ of the trajectories $t \mapsto u(t) \in \cH$ generated by  a third-order in time evolution system.

3: The  function $f: \cH \to \R$ to minimize is supposed to be convex, continuously differentiable, with  $\argmin_{\cH} f \neq \emptyset$.

4: It enters into the dynamic through its gradient.  

5: Based on this new dynamical system, we  improve the results obtained by 

6: [Attouch, Chbani,  Riahi: Fast convex optimization

7: via a third-order in time evolution equation, Optimization 2020]. As a main result,  when the  damping parameter  $\alpha$ satisfies

8:  $\alpha > 3$, we show that   $f(u(t)) - \inf_{\cH} f =   o\left(  1/t^3 \right)$  as $t \rightarrow +\infty$,

9: as well as  the convergence of the trajectories.

10: We complement these results by introducing into the dynamic an Hessian driven damping term, which reduces the oscillations. 

11: In the case of a strongly convex function $ f $, we show an autonomous evolution system of the third order in time with an exponential rate of convergence.

12: All these results have natural extensions to the case of a convex lower semicontinuous function $f: \cH \to \Rb$. Just replace $f$ with its Moreau envelope.

13: \end{abstract}

14: