Materials Genome Initiative: Quantum Monte Carlo for Many-Body Systems
Ceperley’s current research is related to the “Materials Genome Initiative:” the federally supported cross-agency program to develop computational tools to design materials. Quantum Monte Carlo calculations provide nearly exact information on quantum many-body systems and are able to use Blue Waters effectively. They are the only methods capable of treating electron correlation realistically, thus have to be the kernel of any materials design initiative. Ceperley is a world expert on these methods, has trained most of the researchers in this field. His group has been at the forefront of high performance computing for more than 30 years in the area of quantum simulation.
Blue Waters capability is needed to take the research from the model level to realistic description of materials where electron correlation is important. The quantum Monte Carlo algorithm is a mapping from the quantum world to a Markov process. Typical materials require simulation of a unit cell with at least one thousand electrons. Just holding the single-body orbitals in memory for a realistic material requires a multiprocessor computer. Relevant accuracy for materials design requires petascale computational access.
Ceperley and his colleagues have devised algorithms (similar to genetic algorithms) where the computation is shared across processors within a node using Open-MP and across nodes with MPI. In addition, parts of the calculation can be done very effectively on GPUS with speed-ups over the CPUs of more than an order of magnitude. The QMCPACK software, developed in Ceperley’s group, and one of the benchmark algorithms for Blue Waters, is one of the few codes that can take full advantage of its architecture.