Parallel Computation of Multifield Topology: Experience of Haskell in a Computational Science Application

Codes for computational science and downstream analysis (visualization and/or statistical modelling) have historically been dominated by imperative thinking, but this situation is evolving, both through adoption of higher-level tools such as Matlab, and through some adoption of functional ideas in the next generation of toolkits being driven by the vision of extreme-scale computing. However, this is still a long way from seeing a functional language like Haskell used in a live application. This paper makes three contributions to functional programming in computational science. First, we describe how use of Haskell was interleaved in the development of the first practical approach to multifield topology, and its application to the analysis of data from nuclear simulations that has led to new insight into fission. Second, we report subsequent developments of the functional code (i) improving sequential performance to approach that of an imperative implementation, and (ii) the introduction of parallelism through four skeletons exhibiting good scaling and different time/space trade-offs. Finally we consider the broader question of how, where, and why functional programming may - or may not - find further use in computational science.