Quantum Monte Carlo Endstation for Petascale Computing
Quantum Monte Carlo (QMC) Over the last 25 years, simulation methods have had a significant impact on condensed matter, materials and chemical physics. In particular, simulations applied to quantum systems, known collectively as Quantum Monte Carlo (QMC) methods, have become a key contender to predict the properties of systems of correlated electrons in real materials. QMC methodology, which maps the many-particle Schrödinger equation onto a random walk process, represents a new way to solve electronic structure problems. This is because QMC can easily handle the inherent multi-dimensionality of many-electron problems, so that the calculations scale well with the number of particles. Monte Carlo is inherently flexible: it can be parallelized in a variety of independent ways so that calculations show near machine limit efficiencies on virtually any parallel computer. The research undertaken in this project will develop the QMC computational infrastructure so that it can be applied at the petaflop level, providing an unprecedented opportunity for breakthrough science projects in nanoscience, materials science, physics, and chemistry.
More details of this emerging field can be found on the QMC wiki
.
Endstation What is an 'endstation'? When a large particle accelerator is funded, experimental groups are also funded with a fraction of the overall budget to build the necessary detectors to carry out the experimental program. The challenge of scientific computing at the petascale is analogous. If the power of the computational facility is to be realized to achieve scientific breakthroughs, one needs to carefully design, code and test software to run on a machine with tens of thousands of processors, so that the machine can be used as soon as it is available.
Software The goal of this project is to develop a QMC "Endstation" for breakthrough calculations of quantum many-electron systems on petaflop computational platforms; it is focused on building the computational infrastructure for QMC codes to the petaflop scale. Some of the projects are:
- optimized kernels for parallel orbital evaluation and propagation, and local orbital representations that enable sparse matrix techniques to reduce the computer time and memory requirements for very large systems;
- optimized kernels for efficient evaluation of one- and two- body wave functions;
- parallel random number generators to provide millions of independent uncorrelated streams and tests to ensure that their quality is sufficient;
- hierarchical interfaces for load-balancing, monitoring and checkpointing of QMC calculations;
- enabling tools such as analysis and statistical post-processing, interfaces between various QMC codes, web-friendly inputs and outputs and standardized data storage.
The QMC codes to be developed for the endstation are QMCPACK, QWALK, CHAMP and xxxx.
Applications The capability of this endstation will allow a new level in predicting and understanding the behavior of materials systems at the microscopic level. These techniques are applicable to many materials problems. By performing accurate benchmark calculations, the results will validate less computer-intensive methods such as those based on density functional theory. Tests of the methods have been chosen on some of the most challenging application areas with the potential to deliver breakthrough science at the petascale. For example:
- Calculations of transition metal oxide systems with the goal to reveal the quantum origin of their structures, energy differences between various phases and magnetic orderings;
- Calculations related to hydrogen storage systems and photo-voltaics;
- Finite temperature QMC calculations of liquids and crystals with ionic degrees of freedom as dynamical quantum variables, particularly those containing hydrogen.
Collaborating Investigators and Institutions
The team consists of forefront scientists in QMC, developers of major codes and practitioners of these methods in major application projects.
*PICTURES of 9 PIs *
Support
The Endstation QMC collaboration is supported until December 2009 by a $1.8M grant from the DOE xxxx, administered by Oak Ridge National Laboratory.
