Small-time sharp bounds for kernels of convolution semigroups (Q2408412): Difference between revisions

Consider the transition density \(p_t\) of a pure jump Lévy process with Lévy-Khintchin exponent \(\Phi\), infinite Lévy measure \(v\) on \(\mathbb R^d\setminus\{0\}\), and drift term \(b\in\mathbb R^d\). Suppose \(v(dx) = g(x)dx\), \(g(y)= g(-y)\), and \(g(x)\asymp f(x)\), \(x\in\mathbb R^d\setminus\{0\}\) for some nonincreasing function \(f:(0,\infty)\to(0,\infty)\), and define \(\Psi(r):= \sup_{|\xi|\leq r}\mathrm{Re}\,\Phi(\xi)\), \(r>0\), \(h(t):=1/\Psi^{-1}(1/t)\), and \[ b_r:= \begin{cases} b- \int_{r\leq| y|<1} yv(dy) & r<1, \\ b & r=1, \\ b+ \int_{1\leq| y|<r}yv(dy) & r>1. \end{cases} \] The authors give necessary and sufficient conditions for sharp two-side transition density bounds of the form \(p_t(x+ tb)\asymp[h(t)]^{-d} 1_{\{| x|\leq\theta h(t)\}}+ tg(x)1_{\{| x|\leq\theta h(t)\}}\), \(t\in(0,t_0]\), \(t_0>0\), \(x\in\mathbb R^d\), separately for small and large \(x\), as well as sufficient conditions for an estimate of the form \(p_t(x+ tb_{h(t)})\leq Ct[h(t)]^{\gamma-d} f(x)\), \(| x|> R\), \(t\in(0,t_0]\). The proof is purely analytic, based on new precise upper bounds for convolutions of Lévy measures. The conditions involved and several examples are discussed in detail.