From averaging to homogenization in cellular flows -- an exact description of the transition (Q503079)

The authors describe the asymptotic behavior of the solution \( u^{\varepsilon ,R}\) to the problem for the elliptic equation \(\frac{1}{2} \Delta u^{\varepsilon ,R}+\frac{1}{\varepsilon }\upsilon u^{\varepsilon ,R}=-f(\frac{x}{R})\) posed in the 2D domain \(D_{R}\subseteq \mathbb{R}^{2}\) , which is obtained from a bounded and smooth domain \(D\) through a stretching procedure by a factor \(R\). Homogeneous Dirichlet boundary conditions are imposed on the boundary \(u^{\varepsilon ,R}\mid _{\partial D_{R}}=0\). Here, \(f\) is a continuous function on \(D\), and the Hamiltonian \( \upsilon \) is taken as \(\upsilon =(-\partial _{2}H,\partial _{1}H)\), where the stream function \(H\) has nondegeneracy properties like the function \( H(x_{1},x_{2})=\sin (x_{1})\sin (x_{2})\). The problem thus depends on two parameters \(\varepsilon \) and \(R\). When \(R\) is fixed and \(\varepsilon \rightarrow 0\) the limit problem is formally \(\upsilon \nabla u=0\) and an averaging procedure allows describing the precise behavior. When \( \varepsilon \) is fixed and \(R\rightarrow \infty \) the limit problem can be described through homogenization. In special cases, the asymptotic behavior of this problem has been described by \textit{G. Iyer} et al. [Ann. Inst. Henri Poincaré, Anal. Non Linéaire 31, No. 5, 957--983 (2014; Zbl 1302.35039)] through analytical methods. In the present paper, the authors use a probabilistic approach, considering the 2D diffusion process \(dX_{t}^{x,\varepsilon }=\frac{1}{\varepsilon } \upsilon (X_{t}^{x,\varepsilon })dt+dW_{t}\), starting from \( X_{0}^{x,\varepsilon }=x\), on some probability space \((\Omega ,\mathcal{F}, \mathbf{P})\), where \(W_{t}\) is a 2D Brownian motion. The main result of the paper proves the existence of a strictly positive definite matrix \(Q\) such that the law of \(\varepsilon ^{1/4}X_{t}^{x,\varepsilon }\) converges to that of a 2D Brownian motion \(\widetilde{W}_{L_{t}^{\Gamma (x)}}^{Q}\) with covariance matrix \(Q\), where \(L_{t}^{y}\) is the local time of the Fellerian Markov family \(Y_{t}^{y}\) generated by an operator associated to \(H \), and \(\Gamma \) is also associated to \(H\). For the proof, the authors first prove a limit theorem which describes the behavior of the process close to the parametrix. They also use a Lévy-type downcrossing representation of the local time at the interior vortex. Using the representation \[ u^{\varepsilon ,R}(x)=\mathbf{E}\int_{0}^{\tau _{\partial D_{R}}(X_{\cdot }^{x,\varepsilon })}f(X_{s}^{x,\varepsilon }/R)ds, \] where \(\tau _{\partial D_{R}}(\omega )\) is the first hitting of the boundary \(\partial D_{R}\) of \( D_{R}\) by the trajectory \(\omega \in \mathcal{C}([0,T];\mathbb{R}^{2})\), the authors deduce the asymptotic behaviors of \(u^{\varepsilon ,R}\) according to the cases \(R\varepsilon ^{1/4}\rightarrow 0\), or \(R\varepsilon ^{1/4}=C\in (0,\infty )\), or finally \(R\varepsilon ^{1/4}\rightarrow \infty \).