Quantum Monte Carlo (QMC) methods and more specifically Projector Monte Carlo such as Diffusion Monte Carlo (DMC) offer a unique path towards high accuracy calculations for a broad range of electronic systems from molecules to solids. Unlike other electronic structure methods, QMC explicitly uses a many-body wavefunction, including explicitly and accurately effects such as localization, van der Waals interactions, and strong electronic correlations, which are often approximated by other methods. An exact mapping between the Schrodinger equation and an equivalent stochastic process is used both to represent and obtain the actual solution by stochastic methods. Electrons and ions are represented in the continuum, rather than on a lattice, allowing, in principle, QMC to address a much broader range of materials phenomena. The stochastic representation allows the compute intensive parts of the algorithms to be parallelized along several domains (random walkers domain, basis and orbital domain, k-point domain, etc.) independently and efficiently making it an ideal method for High Performance Computing allowing to reach, for most materials, the golden chemical accuracy (~1KCal/mol) allowing to link experiment to theory.
Recent theoretical and algorithmic development and ever-growing computing powers have increased the productivity of DMC by many orders of magnitude and opened up realistic opportunities to compute and predict materials properties. This presentation will describe briefly the QMC theory implemented in the QMCPACK code, the computational schemes used on the next generation of supercomputers, and some highlights of the applications on material science and quantum chemistry.
Argonne Physics Division Colloquium Schedule