next up previous contents
Next: 4.3.1 Non-overlapping elements Up: 4. Basic routines Previous: 4.2.2 Operations with sparse   Contents


4.3 Domain decomposition and basic numerical routines

The data distribution with respect to domain decomposition uses geometrical connections for the parallelization. The domain $ \Omega$ will be split into $ P$ subdomains $ \Omega _i$ ( $ \scriptstyle i=\overline{1,P} \makebox[0pt]{}$), whose will be mapped on a process. Fig. 4.4 presents an part of a non-overlapping domain decomposition. We denote edges between subdomain as Interfaces. The overlapping domain decomposition is not the main scope of that chapter.

Figure 4.4: Non-overlapping domain decomposition.
\begin{figure}\unitlength0.055\textwidth
\begin{center}
\begin{picture}(8,6)
...
....5){\makebox(0,0){$\Omega_3$}}
\end{picture}
\end{center} \protect\end{figure}


Concerning the data type we distinguish between element based and node based data, which we may store overlapping or non-overlapping (Domains are still non-overlapped!). This leads to 4 types of data distributions - we will investigate the non-overlapping distribution of (finite) elements in detail. We regard matrices which have been generated by Finite Element Methods (FEM), Finite Difference Methods (FDM) or Finite Volume Methods (FVM).

Subsections

Gundolf Haase 2000-03-20