1: \begin{abstract}

2:     We study the convergence to local Nash equilibria of gradient methods for two-player zero-sum differentiable games.

3:     It is well-known that

4:     such dynamics converge locally when $S \succ 0$ and may diverge when $S=0$, where $S\succeq 0$ is the symmetric part of the Jacobian at equilibrium that accounts for the ``potential'' component of the game.

5:     We show that these dynamics also converge as soon as $S$ is nonzero (\emph{partial curvature}) and the eigenvectors of the antisymmetric part $A$ are in general position

6:     with respect to the kernel of $S$.

7:     We then study the convergence rates when $S \ll A$ and prove that they typically depend on the \emph{average} of the eigenvalues of $S$, instead of the minimum as an analogy with minimization problems would suggest.

8:     %

9:     To illustrate our results, we consider the problem of computing mixed Nash equilibria of continuous games.

10:     We show that, thanks to partial curvature, conic particle methods -- which optimize over both weights and supports of the mixed strategies -- generically converge faster than fixed-support methods.

11:     For min-max games, it is thus beneficial to add degrees of freedom ``with curvature'': this can be interpreted as yet another benefit of over-parameterization.

12: \end{abstract}

13: