First-order algorithm with $${\mathcal{O}({\rm ln}(1{/}\epsilon))}$$ convergence for $${\epsilon}$$ -equilibrium in two-person zero-sum games