Small time sharp bounds for kernels of convolution semigroups

Abstract: We study small time bounds for transition densities of convolution semigroups corresponding to pure jump L\'evy processes in $\mathbb{R}^{d}$, $d \geq 1$, including those with jumping kernels exponentially and subexponentially localized at infinity. For a large class of L\'evy measures, non-necessarily symmetric nor absolutely continuous with respect to the Lebesgue measure, we find the optimal, both in time and space, upper bound for the corresponding transition kernels at infinity. In case of L\'evy measures that are symmetric and absolutely continuous, with densities $g$ such that $g(x) \asymp f(|x|)$ for nonincreasing profile functions $f$, we also prove the full characterization of the sharp two-sided transition densities bounds of the form $$ p_t(x) \asymp h(t)^{-d} \cdot \mathbf{1}{\left{|x|\leq \theta h(t)\right}} + t \, g(x) \cdot \mathbf{1}{\left{|x| \geq \theta h(t)\right}}, \quad t \in (0,t_0), \ \ t_0>0, \ \ x \in \mathbb{R}^{d}. $$ This is done for small and large $x$ separately. Mainly, our argument is based on new precise upper bounds for convolutions of L\'evy measures. Our investigations lead to an interesting and surprising dichotomy of the decay properties at infinity for transition kernels of pure jump L\'evy processes. All results are obtained solely by analytic methods, without use of probabilistic arguments.