A simulation and decision framework for selection of numerical solvers in scientific computing

Abstract

Selecting the right numerical solver or the most appropriate numerical package for a particular simulation problem it is increasingly difficult for users without an extensive mathematical background and deeper knowledge in numerical analysis. In this paper, we propose a model-driven combined decision-simulation framework for automatically selecting a numerical method for a given set of equation system. We also propose a formal paradigm based on domain-specific languages for specification of structural and behavioral aspects of the numerical equation solving process. Starting from a declarative description of the equation system that need to be solved, our system is able to detect the nature of the equations, perform symbolic manipulations of the equations, and transform them into a domain-specific model. We describe the motivation for such a system, its main features, and a prototype environment together with a usage example.