In the concurrent-computing literature, there is ambiguity over the use of the terms “domain decomposition” and “data distribution.” Many authors use the term domain decomposition for data distributions in which the data local to each process topologically correspond to a subdomain of the computational domain. We shall not use the term domain decomposition in this sense. We reserve the term domain decomposition for certain numerical methods that split the computational domain into two or more sub-domains. Some elementary domain-decomposition methods were already encountered in Section 8.5, where domain-decomposed relaxation methods were discussed.
Unable to display preview. Download preview PDF.