Estimates on the transition densities of Girsanov transforms of symmetric stable processes (Q867079): Difference between revisions

For a Brownian motion \(X_t\) on \(\mathbb R^d\), the transition density of the process \(\widetilde{X}\) obtained from \(X\) by a Girsanov transformation using a natural function has Gaussian estimates. For the symmetric stable processes, the similar problem is considered. Let \((X_t,P_x)\) be an \(m\)-symmetric \(\alpha\)-stable process on \(\mathbb R^d\). For a bounded symmetric function \(F\) on \(\mathbb R^d\times \mathbb R^d\) vanishing on diagonal such that \(E_x(\int_0^t \int_X F^2(X_s,y)/| X_s-y| ^{d+\alpha} \,dy\,ds) <\infty\) and \(\inf_{x,y}F(x,y)>-1\), let \[ M_F(t)=\sum_{s\leq t} F(X_{s-},X_s) -\int_0^t F(X_s,y)/| X_s-y| ^{d+\alpha}\, dy\,ds. \] Using the multiplicative functional \(L_t\) determined by \(L_F(t)=1+\int_0^t L_F(s-)\,dM_F(s)\), define the purely discontinuous Girsanov transform \((\widetilde{X}_t,\widetilde{P}_x)\) by \(d\widetilde{P}_x= L_F(t)1_{\{t<\zeta\}}\,dP_x\) on \({\mathcal M}_t\). After giving the general properties and Dirichlet form of \(\widetilde{X}_t\), it is shown that the transition density of \(\widetilde{X}_t\) is comparable to that of \(X_t\). A similar result is also proved for the Girsanov transform given by \textit{Z.-Q. Chen} and \textit{T.-S. Zhang} [Ann. Inst. Henri Poincaré, Probab. Stat. 38, No. 4, 475--505 (2002; Zbl 1004.60077)]. As an application of these results, for the Feynman-Kac semigroup \(T_tf(x)=E_x\left(\exp(N^u_t)f(X_t)\right)\) determined by the energy zero part \(N^u_t\) of the Fukushima decomposition of \(\widetilde{u}(X_t)-\widetilde{u}(X_0)\) for a bounded function \(u \in {\mathcal F}_e\), a condition for the existence of potential density which is comparable to that of the symmetric stable process is given.