INERTIAL GAME DYNAMICS AND
APPLICATIONS TO CONSTRAINED OPTIMIZATION
RIDA LARAKI AND PANAYOTIS MERTIKOPOULOS
Abstract. We derive a class of inertial dynamics for games and constrained optimization
problems over simplices by building on the well-known “heavy ball
with friction” method. In the single-agent case, the dynamics are generated by
endowing the problem’s configuration space with a Hessian–Riemannian structure
and then deriving the equations of motion for a particle moving under the
influence of the problem’s objective (viewed as a potential field); for normal form
games, the procedure is similar, but players are instead driven by the unilateral
gradient of their payoff functions. By specifying an explicit Nash–Kuiper embedding
of the simplex, we show that global solutions exist if and only if the interior
of the simplex is mapped isometrically to a closed hypersurface of some ambient
Euclidean space, and we characterize those Hessian–Riemannian structures
which have this property. For such structures, low-energy solutions are attracted
to isolated minimizers of the potential, showing in particular that pure Nash
equilibria of potential games are attracting; more generally, when the game is not
a potential one, we establish an inertial variant of the folk theorem of evolutionary
game theory,
No comments:
Post a Comment