Harmonic coordinates for diffusions in the plane (Q2565355): Difference between revisions

Let \(L:=\sum^2_{i,j=1} a_{ij}(x){\partial^2\over\partial x_i\partial x_j}+ \sum^2_{i=1} b_i(x){\partial\over\partial x_i}\) be a second-order differential operator on \(\mathbb{R}^2\), supposed elliptic (with bounded eigenvalues ratio) with locally Hölder continuous coefficients \(a_{ij}\) and \(b_i\). The aim of the article is to establish, under the (almost necessary) hypothesis: there is some \(\varepsilon>0\) such that \(|x|^{1+\varepsilon}|b(x)|= O(|a(x)|)\) as \(|x|\to\infty\), the following theorem: It exists an \(L\)-harmonic global \(C^2\) diffeomorphism of \(\mathbb{R}^2\), say \(u\), such that as \(|x|\to\infty\): \(|u(x)-x|= o(|x|)\) and \(Du(x)\to\text{Id}\). Moreover, up to an additive constant, \(u\) is unique. The proof is delicate, and essentially probabilistic. Formally, \(\text{Id}\)-\(u\) must be the \(L\)-potential of \(b\), which does not a priori converge. So the starting idea is to take a renormalized approximation of this \(L\)-potential, based on its probabilistic representation. The approximated \(L\)-potential is controlled via sharp estimates concerning the \(L\)-diffusion and its hitting times of large spheres, and its gradient is controlled via fine estimates on coupled \(L\)-diffusions. So at the limit an \(L\)-harmonic \(u\) is obtained, that satisfies \(|u(x)-x|= o(|x|)\). The other properties of \(u\) are finally more easily deduced. The same fine result remains in the higher dimensions a conjecture.