On stochastic ordering for diffusion with jumps and applications (Q2643742): Difference between revisions

Let \(X_k(t)\), \(k=0,1,2\) be diffusion processes with jumps with infinitesimal generators defined by \[ \begin{multlined} L_k f(x) = \frac{1}{2} \sum_{i,j} a_{i,j}^{(k)}(x) \frac{\partial^2 f}{\partial x_i x_j} (x) - \sum_i\sum_jQ_{ij}x_j \frac{\partial f}{\partial x_i}(x)\\ + \int_{\mathbb R^d} \Big[ f(x+y)-f(x)- \sum_j y_j\frac{\partial f}{\partial x_j} \text{ind}_{\{| y| <1\}} \Big] \nu(dy) \end{multlined} \] where \(a^{(1)}\) and \(a^{(2)}\) are constant positive definite matrices. Assume the Hessian matrix of \(f\colon\mathbb R^d \rightarrow\mathbb R\) is nonnegative definite a.e. with respect to Lebesgue measure. If for all \(x\in\mathbb R^d\), \(a^{(0)}(x)-a^{(1)}\) and \(a^{(2)}-a^{(0)}(x)\) are nonnegative definite, then \[ E_{x} f(X_1(t)) \leq E_{x} f(X_0(t)) \leq E_{x} f(X_2(t)) \;, \] \(\forall t>0\), \(\forall x\in\mathbb R^d\), provided the expectations exist. This comparison result is applied to a result on the invariant measure: Assume \(a^{(0)}\) satisfies the uniform ellipticity condition \(\lambda^2\| y\| ^2\leq y^{\top} a^{(0)}(x)y\leq \Lambda^2\| y\| ^2\), for \(x,y\in\mathbb R^d\), and that all eigenvalues of \(Q\) have positive real parts. If \(\int_{\{| x| >1\}} \log | x| \,\nu(dx)<\infty\), then \(X_0(t)\) has a unique invariant probability. If the integral equals infinity, then \(\lim_{t\to\infty} \sup_{x} p(t,x,B)=0\) for every bounded \(B\), where \(p\) is the transition probability of \(X_0(t)\).