An object-oriented approach to development and testing of parallel solution algorithms for nonlinear PDES

An object-oriented design that provides flexibility in simulation codes is presented. This flexibility allows programmers freedom to easily change solution algorithms and discretization schemes as well as add new solver packages as they become available. Careful attention is paid to separating algorithm, data, and specific problem classes to provide for ease in changing any of these components. Furthermore, data structures are chosen so that each component works with data in a form best suited to its needs. Lastly, we present some experiences and comments on the tradeoffs involved with this design