The Challenges of Data Intensive and Extreme Scale Numerical Simulation in Physics, Materials Science and Chemistry
Main Organizers: Gabor Csanyi, Detlef Hohl, Stephen Jarvis, Ben Leimkuhler, Christoph Ortner, Mark Parsons
Large-scale numerical simulations in materials science, physics and chemistry are globally among the largest consumers of computer resources and are also generating, processing and storing a vast amount of data. Molecular dynamics, Monte Carlo simulations, N-body solvers, electron density functional calculations and other materials simulation methods consume millions of compute-core hours and generate petabytes of data. Yet these methods are based on computational procedures developed 40+ years ago for sequential computations in a monolithic memory hierarchy. New methods and algorithms are needed that are compatible with current and next generation parallel and data-centric computational paradigms.
Present algorithms also rely on intuition-based translation of mathematical models to computer codes, often using fixed parametric functional forms where general, non-parametric methods would be superior, or simple quadratures instead of more sophisticated adaptive methods. The existing methodologies have yet to catch up with modern developments in machine learning and the fact that the availability of vast computational capacity opens up novel approximation schemes. New methods can combine statistical principles, machine learning, and advanced data analysis to provide significant improvements in accuracy, lead to higher efficiency in statistical sampling, and help to eliminate or automate many of the parameter-optimisation processes. This workshop will bring together key researchers in data science and parallel/high performance computing with problem holders from across academia, industry and government to discuss the challenges posed by extreme-scale computational modelling in physics, materials science and chemistry. The fundamental problems are cross-disciplinary and require collaboration between often isolated communities of computer scientists, mathematicians, data scientists and engineers. We will identify gaps in the UK research portfolio in this area, avenues for technology transfer to industry, ways in which the ATI can have a unique impact. Our aim is to develop a set of recommendations to make important progress beyond the current state of the art.