Multiple time scale analysis of interacting diffusions (Q1326341)

We consider the questions: How can be the long term behavior of large systems of interacting components be described in terms of infinite systems? On what time scale does the infinite system give a qualitatively correct description and what happens at large (resp. critical) time scales? Let \(Y^ N(t)\) be a solution \((y_ i^ N(t))_{i \in [-N,N]}\) of the system of stochastic differential equations \((w_ i(t)\) are i.i.d. Brownian motions) \[ dy_ i^ N(t)=\left( {1 \over 2N+1} \sum^ N_{j=-N} y_ j^ N(t)-y^ N_ i(t) \right) dt+\sqrt {2g(y_ i^ N(t))} dw_ i(t). \] In the McKean-Vlasov limit, \(N \to \infty\), we obtain the infinite independent system \[ dy_ i^ \infty (t)=(E(y_ i^ \infty (t))-y^ \infty_ i(t)) dt+\sqrt {2g(y_ i^ \infty (t))} dw_ i(t),\quad i \in \mathbb{Z}. \] This infinite system has a one parameter set of invariant measures \(\nu_ \Theta=\bigotimes_{x \in \mathbb{Z}} \Gamma_ \Theta\) with \(\Gamma_ \Theta\) the equilibrium measure of \(dx(t)=(\Theta-x(t))dt+\sqrt {2g(x(t))} dw(t)\). Let \(Q_ s (\cdot, \cdot)\) be the transition kernel of the diffusion with generator \(u_ g(x) \bigl({\partial \over \partial x}\bigr)^ 2\) with \(u_ g(x)=\int g(y) \Gamma_ x(dy)\). Then one main result is that, as \(N \to \infty\), \[ {\mathcal L} \Bigl( \biggl( Y^ N \bigl( s(2N+1) \bigr) \biggr) \Bigr) \Rightarrow \int Q_ s (\Theta',d \Theta) \nu_ \Theta,\quad \Theta'=E(y_ 0). \] This provides a new example of a phenomenon also exhibited by the voter model and branching random walk. In particular we are also able to modify our model by adding the term \(cN^{-1} (A-y^ N_ i(t))dt\) to obtain the first example in which the analog of \(Q_ s(\cdot,\cdot)\) converges to a honest equilibrium instead of absorption in traps as in all models previously studied in the literature. Finally, we discuss a hierarchical model with two levels from the point of view discussed above but now in two fast time scales.