1: \begin{abstract}

2:   Stochastic computational models in the form of pure jump processes

3:   occur frequently in the description of chemical reactive processes,

4:   of ion channel dynamics, and of the spread of infections in

5:   populations. For spatially extended models, the computational

6:   complexity can be rather high such that approximate multiscale

7:   models are attractive alternatives. Within this framework some

8:   variables are described stochastically, while others are

9:   approximated with a macroscopic point value.

10: 

11:   We devise theoretical tools for analyzing the pathwise multiscale

12:   convergence of this type of variable splitting methods, aiming

13:   specifically at spatially extended models. Notably, the conditions

14:   we develop guarantee well-posedness of the approximations without

15:   requiring explicit assumptions of \textit{a priori} bounded

16:   solutions. We are also able to quantify the effect of the different

17:   sources of errors, namely the \emph{multiscale error} and the

18:   \emph{splitting error}, respectively, by developing suitable error

19:   bounds. Computational experiments on selected problems serve to

20:   illustrate our findings.

21: \end{abstract}

22: