Opportunities in Computational Science to Advance Software Engineering

Software development in Computational Science and Engineering (CSE) is in a different universe. CSE is mostly mathematical computations. Tools are needed to help the few experts who write scientific libraries; these libraries are used by the CSE masses. CSE software is broadly believed to be different from “normal” software. First, it is mathematical – there are precise (if not formal) specifications for code to be developed. Relationships among program elements are defined by mathematics – few domains in SE have such rich underpinnings. Our guess is that typical SE domains do have some of this richness, but finding such relationships is hardly mainstream SE research and certainly not practiced by or taught to typical programmers. In short, contemporary SE education and interests are misaligned for long-term CSE needs. Second, CSE software tends to be algorithm-centric, not behavior-centric. Early SE researchers (Dijkstra, Hoare) focused on disciplined development of algorithms. Today’s systems stress the coordination of agents (classes, subsystems) to produce particular behaviors; algorithms tend to be a tiny part of a system’s overall design. Consequently, few papers relevant to CSE software technology find their way into major SE conferences. Simply put, CSE is not on the cutting-edge or horizon of today’s SE researchers. Although a discouraging but realistic observation, we see it as an opportunity in CSE to advance SE in a general way. 2. What We Think is Needed