1: \begin{abstract}

2: Sparse grids are tailored to the approximation of smooth high-dimensional

3: functions. On a $d$-dimensional cube, the number of grid points is

4: $N = \mathcal O(h^{-1} |\log h|^{d-1})$ with a mesh size parameter $h$.

5: The so-called combination technique, based on hierarchical decomposition,

6: facilitates the numerical solution of partial differential equations on these

7: grids.

8: Key to the convergence analysis are specific multivariate error expansions,

9: which we derive in a generic way for linear difference schemes

10: through an error correction technique employing semi-discretisations.

11: We obtain error formulae of the structure $\varepsilon = \mathcal{O} (h^p |\log

12: h|^{d-1})$

13: and illustrate the convergence by numerical examples.

14: \end{abstract}

15: