Another proof of the averaging principle for fully coupled dynamical systems with hyperbolic fast motions (Q2574723)

The author considers a system of differential equations (*) for \(X^\varepsilon= X^\varepsilon_{x,y}\) and \(Y^\varepsilon= Y^\varepsilon_{x,y}\), \[ dX^\varepsilon /dt=\varepsilon\cdot B(X^\varepsilon(t),Y^\varepsilon(t)), \quad dY^\varepsilon/dt= b (X^\varepsilon(t),Y^\varepsilon(t)),\tag \(*\) \] with the initial conditions \(X^\varepsilon (0)= x\), \(Y^\varepsilon(0)=y\) on the product \(\overline{\mathcal X}\times M\), where \({\mathcal X}\subset\mathbb{R}^\alpha\) is a bounded connected open set, \(\overline{\mathcal X}\) is its closure and \(M\) is a compact \(C^2\) Riemannian manifold, and where \(B(x,y)\), \(b(x,y)\) are vector fields on \({\mathcal X}\) and \(M\), respectively, which are assumed to be bounded and Lipschitz continuous. To generalize the results in the author's previous paper [Ergodic Theory Dyn. Syst. 24, 847--871 (2004; Zbl 1055.37025)], the author assumes that each flow \(F^t_x(x\in{\mathcal X})\) on \(M\) given by \(dF^t_x (y)/dt=b(x,F^t_xy)\), \(F^0_xy=y\) possesses a basic hyperbolic attractor \(\Lambda_x\). Moreover, the author assumes the boundedness of the \(C^1\)-norm of \(B(x,y)\) and \(C^2\)-norm of \(b(x,y)\) for \((x,y)\in \overline{\mathcal X}\times M\). Denoting by \(\overline B(x)\) the average of \(B(x,y)\) with respect to some measure \(\mu^0_x\) on \(\Lambda_x\), we see that \(\overline B(x)\) is \(C^1\) in \(x\) and so \(\overline B\) is bounded and Lipschitz continuous. Thus, we know the existence of uniqueness of the solutions of the equations (*) and the equations \(d\overline X_x^\varepsilon(t)/dt=\varepsilon\cdot\overline B(\overline X_x^\varepsilon(t))\), \(\overline X^\varepsilon_x(0)=x\). Under these assumptions, the author proves the convergence in average in the averaging principle, i.e., the average of the suprema of \(|X^\varepsilon_{x,y}(t)-\overline X_x^\varepsilon(t)|\) for \(0\leq t<T/ \varepsilon\), with respect to a probability measure on \({\mathcal X}_T\times M\) converges to 0 for \(\varepsilon\to 0\), where \({\mathcal X}_T\) is some open set in \({\mathcal X}\). The author asserts that we can formulate the corresponding results in the discrete time setup with difference equations, instead of differential equations (*).