skip to primary navigationskip to content


Accelerating first-principles simulations on emerging computer architectures

The use of quantum mechanical "first-principles" simulations, particularly those based on Density Functional Theory (DFT), has seen a dramatic increase over the past three decades, modelling electronic, chemical and physical properties of matter in fields across the physical sciences and, increasingly, biology and medicine.

The enormous success of DFT has only been possible because of the development of high-performance software implementations. One such software package is Castep ( [1]), developed jointly by researchers at the Universities of Cambridge, Durham, Oxford, Royal Holloway and York. Castep is written in modern Fortran, using OpenMP and MPI for its parallelism, and is well-optimised for conventional High-Performance Computing (HPC); however recent hardware developments have seen a move towards new architectures with large numbers of low-power cores. Such machines include hybrid CPU-Accelerator systems (e.g. Intel MIC (Xeon Phi), nVidia Tesla or ARM v8), or massively-parallel systems like IBM's BlueGene. The theoretical peak performance of these emerging technologies is enormous, but exploiting their unusual architecture effectively requires a re-imagining of many of Castep's key operations.

This project involves redesigning and developing the Castep DFT program to take advantage of these new technologies. The work is likely to involve new methods and algorithms, as well as new software paradigms (e.g. OpenACC), and will be carried out in collaboration with researchers at the University of York. Exploiting these cutting-edge technologies effectively will dramatically increase the performance of Castep, increasing practical simulation sizes, speeding up molecular dynamics calculations and opening up its application to new scientific areas.