On the Global Convergence Rates of Decentralized Softmax Gradient Play in Markov Potential Games

Neural Information Processing Systems 

The stochastic game (SG) is a classical multi-agent model that has received extensive attention in recent MARL studies.