Seminar 8th March 2013 4 p.m. 27:2001
The End of Atomistic Simulation?
Professor Mike Payne FRS
Cavendish Laboratory, University of Cambridge
- CASTEP, Complex Systems, Computer Science, CUDA, CUDA Fortran, CVS, Data Management, Density functional Theory, e-Research, FFT, Fortran, GPU, HECToR, HPC, HPCx, Iridis, Jaguar, Level set, Materials, MEMS, Metals, Molecular Mechanics, Monte Carlo, Multi-core, Multi-physics, Multi-scale, Multigrid solvers, NWCHEM, Onetep, Particle Collisions, Photonics, Quantum Chemistry, Scientific Computing, Software Engineering, Symbolic calculation, Visualisation, Voxel imaging, Xmgrace
- Chris-Kriton Skylaris
The era of predictive quantum mechanical atomistic simulations, based either on density functional theory (DFT) or quantum chemistry techniques, began in the early 1980s. Within a decade, DFT codes had been packaged into ‘black box’ techniques which could be routinely applied to systems containing many tens of atoms and, most importantly, could be used by non-experts. This has led to an explosion in the number of scientific publications based on DFT and its widespread use in industry. However, despite the extraordinary success of DFT it has many limitations such as: (i) restricted system size, even with the development of linear scaling methods that can be applied to thousands or even millions of atoms: (ii) very short simulation timescales, typically of the order of picoseconds (iii) limited accuracy. Given the computational efficiency of modern DFT codes, further increases in system sizes and timescales will be primarily driven by Moore’s law and, as such, will be relatively slow. Therefore, to develop a universal methodology for quantum mechanical based atomistic simulations we need to move away from regarding DFT as the universal panacea and develop : (i) a ‘black box’ multiscale modelling methodology that links beyond-DFT accuracy methods to DFT calculations which, in turn, are linked to empirical atomistic simulations based on simple interatomic potentials which, in turn, are linked to continuum simulations; (ii) fast phase space searching techniques; and (iii) acceleration techniques for dynamical simulations to access much longer timescales. Furthermore, to make the methodology of general use, so that has the same level of impact as DFT, all of these must be automated ‘black box’ approaches which can be used by non-experts. Remarkably, I believe that all the intellectual challenges that need to be overcome to develop such a methodology have now been addressed. As such, we may be close to the end of the era in atomistic simulation where we have been held back by challenges that could only be overcome through the development of innovative new methodologies. This does not, however, mean that further innovation will not be welcome or that there will not be almost infinite need for application of the methodology once all the components are put together.