A mini-max variational formula giving necessary and sufficient conditions for recurrence or transience of multidimensional diffusion processes (Q1106561): Difference between revisions

Let X be a diffusion process in \(R^ d\) with the generator \(L=2^{- 1}\nabla \cdot a\nabla +b\cdot \nabla\), where a(x) is a positive-definite matrix at each \(x\in R^ d\), the entries \(a_{ij}\) of a and the components \(b_ i\) of b are in \(C^{1,\alpha}_{loc}(R^ d)\), and k is an arbitrary function in \(C^{2,\alpha}(R^ d)\). Denote \(\Sigma_ n=\{x:| x| \leq n\}\), \[ {\mathcal G}^{(k)}=\{g:\quad g\in W^{1,2}(\Sigma_ n-\Sigma_ 1);\quad g=e^ k\quad on\quad \partial \Sigma_ 1;\quad g=0\quad on\quad \partial \Sigma_ n; \] \[ (dist(x,\partial \Sigma_ n))^{-1}g(x)\in L^{\infty}(\Sigma_ n- \Sigma_ 1)\}, \] \[ \lambda_ n^{(k)}=\inf_{g\in {\mathcal G}^{(k)}}\sup_{h\in W^{1,2}(\Sigma_ n-\Sigma_ 1,g^ 2dx),h=k\quad on\quad \partial \Sigma_ 1}[2^{-1}\int_{\Sigma_ n- \Sigma_ 1}(\nabla g/g+ \] \[ -a^{-1}b)a(\nabla g/g-a^{-1}b)g^ 2dx- 2^{-1}\int_{\Sigma_ n-\Sigma_ 1}(\nabla h-a^{-1}b)a(\nabla h- a^{-1}b)g^ 2dx]. \] It is proved that \(\lim_{n\to \infty}\lambda_ n^{(k)}=0\) or \(\lim_{n\to \infty}\lambda_ n^{(k)}>0\) according to whether X is recurrent or transient.