1: \begin{abstract}

2: %In this paper we present a first-order method for convex-concave min-max optimization that we call \textit{Clairvoyant Extra-Gradient} (CEG) for which we establish near-optimal time-average and last-iterate convergence in the convex/concave case. Similarly to a recent approach in normal form games~\cite{PSS21}, the key idea of our work is to exploit the fact that the update rule of the \textit{Proximal Point method} (PP) can be approximated up to accuracy $\epsilon$ with only $\mathcal{O}(\log 1/\epsilon)$ additional gradient-calls through the iterations of a contraction map. Then (CEG) comes as a straight-forward approximation of (PP) that approximates its update rule at every step. Due to the above, (CEG) admits a significantly simpler analysis compared to other famous implementable first-order methods such as \textit{Extra-Gradient} (EG) and \textit{Optimistic Gradient Descent/Ascent} (OGDA). The flexibility of our approach allows us to capture several settings such as unconstrained optimization as well as bounded/unbounded diameter domains. 

3: %\end{abstract}

4: