Abstract
Monte Carlo simulations are inherently compute-bound. Although short computations may provide order-of-magnitude estimates, long CPU times are generally required to achieve the accuracy needed for reliable comparison of Monte Carlo results with experiment or theory. The advent of supercomputers, which have made possible significantly increased computer speeds for those applications which are amenable to vector or parallel processing, thus offers promise for Monte Carlo applications. In fact, Monte Carlo codes are often highly parallel, and offer multiple avenues for both parallelization and vectorization.We explore the gains to be obtained with supercomputers for the quantum Monte Carlo (QMC) method. The QMC algorithm treated here is used in quantum mechanical molecular calculations, to obtain solutions to the Schrödinger equation. This approach has recently been shown to achieve high accuracy in electronic structure computations. QMC is here demonstrated to fully take advantage of parallel and vector processor systems. Levels of parallelism are discussed, and an overview of parallel computer architectures, as well as present vector supercomputers is given. We also discuss how one adapts QMC to these machines. Performance ratios (versus scalar operation) for a number of supercomputer systems are given.KeywordsVector LengthEnsemble SizeSingle PrecisionMonte Carlo CodeQuantum Monte CarloThese keywords were added by machine and not by the authors. This process is experimental and the keywords may be updated as the learning algorithm improves.
Submitted Version (
Free)
Published Version
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call