Rare Events via Parallel Forward Flux Sampling

Rare events can have extremely important consequences. A recent example is of course the credit crunch of 2008, a global event of a size not seen since the 1930s. A part of the systematic failure to predict this event has been attributed to the failure of the banks' computer models to account properly for these events which, while having very low probability, are of very high import.From the scientific perspective, such a failure is understandable. Simulating rare events is extremely hard. This is for the simple reason that any given computer simulation will simply not see any rare events in any reasonable time. If you are looking for them, most (if not all) of your computational time and effort will be wasted.In our work, we will attempt to improve the situation. To do this, we will combine a powerful new technique for capturing rare events with the use of large supercomputers. The goal of the work is to provide software which will make it easier for scientists to apply this technique (known as Forward Flux Sampling ) to more difficult problems. As an example, the method can be used to study conformational (shape) changes in proteins; rare mistakes in the folding of proteins is thought to play an important role in neurodegenerative diseases. Our new software will help to make the study of such problems more straightforward. It may even help to predict the next financial crisis!

The software development planned in this proposal will provide significantly enhanced capability in the simulation of rare events compared with what is available today. Parallel Forward Flux Sampling (FFS) will allow large non-equilibrium problems to be addressed for the first time. This encompasses many `real-world' problems, almost all of which can be regarded as non-equilibrium: examples are biological systems, the climate, and the economy. As a technique which can be employed for any kind of rare event in systems with stochastic dynamics, FFS is potentially of interest to a wide audience. This audience includes both pure and applied scientists, those engaged in health (to study for example protein folding, and its impact on neurodegenerative diseases), and modellers in the industrial, commercial and financial sectors where timely forecasts of rare events are conspicuously missing. The creation of a library for parallel FFS will make the method accessible to those non-expert users who may be less familiar with writing computer programs, and to those not familiar with parallel computing. More and more potential users, particularly in the biological sciences, now fall into this category. Specifically, users of the popular LAMMPS molecular dynamics code will benefit from the implementation of WP3.1, and users of the lattice Boltzmann code Ludwig from that of WP3.2. A plan for further improvement of accessibility (via plugins for other popular codes, and a graphical user interface analysis tool) has been outlined to form part of a possible second phase HPC software development call. We believe this represents a clear translational pathway to impact in the scientific and industrial sectors. The software developed by this proposal will conform to international (ANSI) standards for programming languages and be made publicly available under a suitable open source licence (e.g., the Gnu General Public Licence version 2). We will use the message passing interface (MPI) to ensure portability across the widest range of high performance parallel computers. The medium for distribution will be a web-based `sourceforge'-like account, now the de facto standard for open source codes.We will pay particular attention to the quality of the product: ease of installation and use, and comprehensive documentation will be priorities to encourage uptake and adoption of the new software.