On the $\varepsilon$-Euler-Maruyama scheme for time inhomogeneous jump-driven SDEs - Inria - Institut national de recherche en sciences et technologies du numérique Accéder directement au contenu
Pré-Publication, Document De Travail Année : 2024

On the $\varepsilon$-Euler-Maruyama scheme for time inhomogeneous jump-driven SDEs

Résumé

We consider a class of general SDEs with a jump integral term driven by a time-inhomogeneous random Poisson measure. We propose a two-parameters Euler-type scheme for this SDE class and prove an optimal rate for the strong convergence with respect to the $L^p(\Omega)$-norm and for the weak convergence. One of the primary issues to address in this context is the approximation of the noise structure when it can no longer be expressed as the increment of random variables. We extend the Asmussen-Rosinski approach to the case of a fully dependent jump coefficient and time-dependent Poisson compensation, handling contribution of jumps smaller than $\varepsilon$ with an appropriate Gaussian substitute and exact simulation for the large jumps contribution. For any $p \geq 2$, under hypotheses required to control the $L^p$-moments of the process, we obtain a strong convergence rate of order $1/p$. Under standard regularity hypotheses on the coefficients, we obtain a weak convergence rate of $1/n+\epsilon^{3-\beta}$, where $\beta$ is the Blumenthal-Getoor index of the underlying L\'evy measure. We compare this scheme with the Rubenthaler's approach where the jumps smaller than $\varepsilon$ are neglected, providing strong and weak rates of convergence in that case too. The theoretical rates are confirmed by numerical experiments afterwards. We apply this model class for some anomalous diffusion model related to the dynamics of rigid fibres in turbulence.

Dates et versions

hal-04404438 , version 1 (18-01-2024)

Licence

Paternité

Identifiants

Citer

Mireille Bossy, Paul Maurer. On the $\varepsilon$-Euler-Maruyama scheme for time inhomogeneous jump-driven SDEs. 2024. ⟨hal-04404438⟩
16 Consultations
0 Téléchargements

Altmetric

Partager

Gmail Facebook X LinkedIn More