CHIMPS: A high-performance scalable module for multi-physics simulations

As computational methods attempt to simulate ever more complex physical systems the need to couple independently-developed numerical models and solvers arises. This often results from the requirement to use different physical or numerical models for various portions of the domain of interest. In many situations it is also common to use different physical models that influence each other within the same domain of interest. The interaction between these models normally requires an exchange of information between the participating solvers. When the solvers that exchange information are distributed over a large number of processors in a parallel computer, the problem of exchanging information in an efficient and scalable fashion becomes complicated. This paper describes our efforts to develop a Coupler for High-Performance Integrated Multi-Physics Simulations library, CHIMPS, that can enable the exchange of information between solvers and that automates the search, interpolation and communication processes in order to allow the developer to focus on other matters of interest such as the appropriate strategies to couple the solvers in an accurate and stable fashion. Our basic approach, the underlying technology, the CHIMPS API, and a number of examples are presented. In addition, a series of appendices are included with actual sample code that can be used to become familiar with the CHIMPS library.