Pathwise uniqueness of stochastic differential equations driven by Cauchy processes with drift (Q2231252)

This paper deals with the following SDE \[ dX_t= F(X_{t-})dZ_t,~X_0=x\tag{\(*\)} \] where \(Z\) is a Lévy process with drift \(\gamma\) driven by Lévy-Khintchine representation \[ \mathbb{E}[e^{iuZ_t}]=\exp[t(-\pi|u|+i\gamma u],~\gamma>0. \] The case \(\gamma=0\) could be solved with a change of function and \(Z\). The main result is the uniqueness of the solution of SDE (\(*\)) under the following assumptions concerning the function \(F\): \(\text{ for all } n\in \mathbb{N}\), \begin{align*} \text{there exists } C_1(n)& \text{ such that } |F(x)-F(y)|\leq C_1(n)|x-y|^{1-\beta},~\beta:=\frac{2}{\pi}\operatorname{arc}\tan\frac{\gamma}{\pi}, \\ \text{there exists } C_2(n) &\text{ such that } [F(x)-F(y)]\cdot \operatorname{sgn}(x-y)\leq C_2(n). \end{align*} The author introduces the following tools: For \(\beta\in (0,1)\), \(\Phi(x)=|x|^\beta\) is approximated by a `mollifier' \(\Phi_n\) defined using a function \(\psi\in C^\infty_c(\mathbb{R})\), \(\operatorname{supp}(\psi)\subset [-1,+1]\), \(\int_{-1}^1\psi(x)dx=1\). Denote \(\psi_n: x\mapsto n\psi(nx)\) and then the `mollifier' \(\Phi_n\) is the convolution product \(\Phi*\psi_n\). This approximation \(\Phi_n\) satisfies convenient hypotheses such that we can take the expectation of \(\Phi_n(X^1_t-X^2_t)\) (where \(X^i,~i=1,2\) are two solutions of (\(*\))) developed with Itô's formula. Finally, after the introduction of a stopping time \(T_m:=\inf\{ t>0:\inf(|X^1_t|,|X^2_t|)>m\}\) these tools allow the author to compute the expectation of \(E[|X^1_{\inf(t,T_m)}-X^2_{\inf(t,T_m)}|^\beta]\), then \(E[|X^1_{\inf(t,T_m)}-X^2_{\inf(t,T_m)}|^\beta]\) satisfies a Gronwall's inequality and getting \(m\) to infinity concludes to the uniqueness of the solution of (\(*\)).