New Proof: Multi-Agent Learning Equilibria Are Unstable Saddle Points
arXiv·medium signal
Paradoxes of Game Theoretic Equilibria and Price of Anarchy (arXiv 2607.11752, DeepMind) proves the worst-case equilibria anchoring Price-of-Anarchy bounds are topologically unstable saddle points, and that discrete-time learning in non-atomic congestion games produces Li-Yorke chaos with time-averaged inefficiency degrading exponentially as 2^p. A theoretical caution for anyone relying on equilibrium convergence guarantees in multi-agent systems.