Markov chain approximations for transition densities of Lévy processes (Q2637751)

The paper is devoted to a comprehensive analysis of the convergence rates of several quantities (characteristic exponents, transition kernels, expectations and algorithms for simulations) in the approximation of Lévy processes \(X_t\) by means of (continuous time) Markov chains \(X^h_t\) with states \(kh\) (\(k\in \mathbb{Z}\)) for \(h\to0\). The proofs are mainly developed for the one-dimensional case, but extensive discussions are allotted also to the \(d\)-dimensional generalizations of the results. The outcomes are then properly summarized in a few tables according to both the presence/absence of a diffusive component, and the behavior of the Lévy measure \(d\lambda(x)\) in the target Lévy process \(X_t\). To make short a long and detailed paper, the main results about the transition kernels (which are not a foregone consequence of the previous results about the characteristic exponents: see Propositions 4.5 and 4.6 syntesized in Theorem 2.4, with a multivariate generalization in the Theorem 2.8) can be summed up as follows: if \(P^h_{t,T}(x,y)\) and \(p_{t,T}(x,y)\) denote the transition probabilities and densities respectively of \(X^h\) and \(X\) from \(x,t\) to \(y,T\), then -- under several technical assumptions -- for \(h\to0\) the error assessment \[ \Delta_{T-t}(h)=\sup_{x,y}\left|p_{t,T}(x,y)-h^{-1}P^h_{t,T}(x,y)\right| \] behaves either as \(O(h^2)\), or as \(O(h)\), respectively, when \(\lambda(\mathbb{R})=0\), or \(0<\lambda(\mathbb{R})\leq\infty\) (but \(\kappa(0)<\infty\)), and in the presence of a non-zero diffusive component; while, on the other hand, for \(\kappa(0)=\infty\), the error goes as \(O\left(h\kappa\left({h\over 2}\right)\right)\), where, with \(\delta>0\), \[ \kappa(\delta)=\int_{[-1,1]\backslash[-\delta,\delta]}|x|\,d\lambda(x)\qquad (\kappa(0)\leq\lambda(\mathbb{R})). \] The paper is lavish then on details and proofs about every other possible generalization or offspring of the previous results, and also presents several examples of numerical simulations visually displaying the actual approximation rates to particular instances of Lévy processes.