Existence of densities for jumping stochastic differential equations (Q2490049)

Consider a \(d\)-dimensional Markov process with jumps \(\{X_t^x\}_{t\geq0}\), starting from \(x\in \mathbb R^d\), with generator \(\mathcal L\), defined, for \(\varphi : \mathbb R^d \mapsto\mathbb R\) sufficiently smooth and \(y\in \mathbb R^d\), by \[ \mathcal L\varphi(y) = b(y)\cdot \nabla\varphi(y) +\int_{\mathbb R^n}\gamma(y)\varphi(z) [\varphi(y + h(y, z))-\varphi(y)]\,dz,\tag{1} \] with possibly an additional diffusion term, and the integral part written in a (more general) compensated form. Here \(n\in\mathbb N\) is fixed, and the functions \(\gamma:\mathbb R^d \mapsto \mathbb R\) and \(\varphi : \mathbb R^n \mapsto\mathbb R\) are nonnegative. The aim of the paper is to investigate the absolute continuity of the law of \(X_t^x\) with respect to the Lebesgue measure on \(\mathbb R^d\), for \(t > 0\). The authors consider the case where the rate of jumping \(\gamma\) is constant, in the spirit of \textit{K.~Bichteler} and \textit{J.~Jacod} [in: Séminaire de probabilités XVII. Lect. Notes Math. 986, 132--157 (1983; Zbl 0525.60067)], under a strong non-degeneracy assumption on \(h\) with rather weaker assumptions and simpler proofs, not relying on the use of stochastic calculus of variations. Then the method is extended to the case where \(\gamma\) depends on the spatial variable. An example of application to some fragmentation equations is given.