1: \begin{abstract}

2: \noindent

3: The convergence of multigrid methods degrades significantly if a small

4: number of low quality cells are present in a finite element mesh, and

5: this can be a barrier to the efficient and robust application of

6: multigrid on complicated geometric domains. The degraded performance is

7: observed also if intermediate levels in a non-nested geometric multigrid

8: problem have low quality cells, even when the fine grid is high quality.

9: It is demonstrated for geometric multigrid methods that the poor

10: convergence is due to the local failure of smoothers to eliminate parts

11: of error around cells of low quality. To overcome this, a global--local

12: combined smoother is developed to maintain effective relaxation in the

13: presence of a small number of poor quality cells. The smoother involves

14: the application of a standard smoother on the whole domain, followed by

15: local corrections for small subdomains with low quality cells. Two- and

16: three-dimensional numerical experiments demonstrate that the degraded

17: convergence of multigrid for low quality meshes can be restored to the

18: high quality mesh reference case using the proposed smoother. The effect

19: is particularly pronounced for higher-order finite elements. The results

20: provide a basis for developing efficient, non-nested geometric multigrid

21: methods for complicated engineering geometries.

22: %

23: \\*[2ex]

24: %

25: {\bf Keywords:} multigrid, multigrid smoother, finite element methods,

26: cell quality, domain decomposition.

27: \end{abstract}

28: