Bulletin of the American Physical Society
APS March Meeting 2014
Volume 59, Number 1
Monday–Friday, March 3–7, 2014; Denver, Colorado
Session B27: Focus Session: Petascale Science and Beyond: Applications and Opportunities for Materials Science I |
Hide Abstracts |
Sponsoring Units: DCOMP Chair: Jack Wells, Oak Ridge National Laboratory Room: 501 |
Monday, March 3, 2014 11:15AM - 11:51AM |
B27.00001: Condensed Matter Applications of Quantum Monte Carlo at the Petascale Invited Speaker: David Ceperley Applications of the Quantum Monte Carlo method have a number of advantages allowing them to be useful for high performance computation. The method scales well in particle number, can treat complex systems with weak or strong correlation including disordered systems, and large thermal and zero point effects of the nuclei. The methods are adaptable to a variety of computer architectures and have multiple parallelization strategies. Most errors are under control so that increases in computer resources allow a systematic increase in accuracy. We will discuss a number of recent applications of Quantum Monte Carlo including dense hydrogen and transition metal systems and suggest future directions. [Preview Abstract] |
Monday, March 3, 2014 11:51AM - 12:03PM |
B27.00002: Peta-scale QMC simulations on DOE leadership computing facilities Jeongnim Kim Continuum quantum Monte Carlo (QMC) has proved to be an invaluable tool for predicting the properties of matter from fundamental principles. Even with numerous innovations in methods, algorithms and codes,~\footnote{Kim et al., J. Phys.: Conf. Ser., {\bf 402} 012008 (2012).} QMC simulations of realistic problems of 1000s and more electrons are demanding, requiring millions of core hours to achieve the target chemical accuracy. The multiple forms of parallelism afforded by QMC algorithms and high compute-to-communication ratio make them ideal candidates for acceleration in the multi/many-core paradigm. We have ported and tuned QMCPACK to recently deployed DOE doca-petaflop systems, Titan (Cray XK7 CPU/GPGPU) and Mira (IBM Blue Gene/Q). The efficiency gains through improved algorithms and architecture-specific tuning and, most importantly, the vast increase in computing powers have opened up opportunities to apply QMC at unprecedent scales, accuracy and time-to-solution. We present large-scale QMC simulations to study energetics of layered materials where vdW interactions play critical roles. [Preview Abstract] |
Monday, March 3, 2014 12:03PM - 12:15PM |
B27.00003: Quantum Monte Carlo simulations on Blue Gene/Q using QMCPACK: Performance and Applications Anouar Benali, Luke Shulenburger, Nichols A. Romero, Jeongnim Kim Quantum Monte Carlo (QMC) is the most accurate many-body method for computing ground-state properties in condensed-phase systems. QMC uses a stochastic sampling method to solve the many-body Schr\"{o}dinger equation. The advent of petascale supercomputing facilities and massively concurrent QMC algorithms has allowed us to study materials at unprecedented levels of accuracy. We will present the implementation and optimization of the QMCPACK [1-2] simulation package on the IBM Blue Gene/Q as well as results for a number systems including: van der Waals-dominated materials, transition metals and biological molecules. \\[4pt] [1] K. Esler, J. Kim, L. Shulenburger, and D. Ceperley, Computing in Science and Engineering 14, 40 (2012).\\[0pt] [2] J. Kim et al., Journal of Physics: Conference Series 402, 012008 (2012). [Preview Abstract] |
Monday, March 3, 2014 12:15PM - 12:27PM |
B27.00004: Diffusion Monte Carlo characterization of a methane molecule in a (H$_{2}$O)$_{20}$ dodecahedral cage Kenneth Jordan, Michael Deible The diffusion Monte Carlo method is used to investigate the interaction of a water molecule with a dodecahedral (H$_{2}$O)$_{20}$ cage as found in the methane hydrate crystal. The DMC value of the interaction energy between the methane molecule and the cage are compared with the results of MP2 and symmetry-adapted perturbation theory (SAPT) calculations [1]. In addition, the net interaction energy is decomposed into two- and three-, and $n \ge$ four-body contributions. The two- and three-body contributions are further analyzed in terms of SAPT calculations [1,2]. \\[4pt] [1] A. J. Misquitta, R. Podeszwa, B. Jeziorski, and K. Szalewicz, J. Chem. Phys. \textbf{123}, 214103 (2005); A. Hesselmann, G. Jansen, and M. Sch\"{u}tz, J. Chem. Phys., \textbf{122}, 014103 (2005).\\[0pt] [2] R. Podeszwa and K. Szalewicz, J. Chem. Phys., \textbf{126}, 194101 (2007). [Preview Abstract] |
Monday, March 3, 2014 12:27PM - 12:39PM |
B27.00005: Quantum Monte Carlo for Materials Design Tim Mueller, Lucas Wagner, Jeffrey Grossman The accurate calculation of formation energies is critical to evaluating the stability and chemical reactivity of newly designed materials. Comprehensive databases of formation energies can be used to screen materials for stability before they have been synthesized, but the reliability of such databases depends on the accuracy of the data they contain. Quantum Monte Carlo (QMC) is a highly accurate method that can calculate the formation energies a wide variety of chemical substances, including molecules and metals. The wide applicability of QMC calculations is made possible in part by the fact that the cost of a QMC calculation scales roughly linearly with system size when calculating energies per atom. The accuracy of QMC comes at significant computational cost, but it scales nearly linearly with the number of processors up to a large numbers of processing cores, making it well-suited for large, highly parallel computers. As the initial step towards developing a database of accurate formation energies calculated using QMC, we demonstrate how automated QMC calculations can be used to accurately calculate the formation energies of a variety of different materials. [Preview Abstract] |
Monday, March 3, 2014 12:39PM - 12:51PM |
B27.00006: Detecting phase-transitions in electronic lattice-models with DCA$^+$ Peter Staar, Thomas Maier, Thomas Schulthess The DCA$^+$ algortihm was recently introduced to extend the dynamic cluster approximation (DCA) by introducing a self-energy with continuous momentum dependence. This removes artificial long-range correlations and thereby reduces the fermion sign problem as well as cluster shape dependencies. Here, we extend the DCA$^+$ algorithm to the calculation of two-particle quantities by introducing irreducible vertex functions with continuous momentum dependence compatible with the DCA$^+$ self-energy. This enables the study of phase transitions within the DCA$^+$ framework in a much more controlled fashion than with the DCA. We validate the new method using a calculation of the superconducting transition temperature $T_c$ in the attractive Hubbard model by reproducing previous high-precision finite size quantum Monte Carlo results. We then calculate $T_c$ in the doped repulsive Hubbard model, for which previous DCA calculations could only access the weak-coupling ($U=4t$) regime for large clusters. We show that the new algorithm provides access to much larger clusters and thus asymptotic converged results for $T_c$ for both the weak ($U=4t$) and intermediate ($U=7t$) coupling regimes, and thereby enables the accurate determination of the exact infinite cluster size result. [Preview Abstract] |
Monday, March 3, 2014 12:51PM - 1:03PM |
B27.00007: Implementation of continuous-time QMC impurity solver for Dynamical Mean Field Theory Mancheon Han, Choong-Ki Lee, Hyoung Joon Choi Dynamical mean field theory maps an interacting lattice problem to an interacting impurity problem in non-interacting bath. Continuoustime Quantum Monte Carlo (CT-QMC) method is numerically exact way to obtain a solution for such an impurity problem. We developed hybridization-expansion CT-QMC (CT-HYB) impurity solver and tested its validity by studying infinite neighbor Bethe lattice which has semicircular density of states and only local Hubbard interaction. This work is supported by the NRF of Korea (Grant No. 2011-0018306). Computational resources have been provided by KISTI Supercomputing Center (Project No. KSC-2013-C3-008). [Preview Abstract] |
Monday, March 3, 2014 1:03PM - 1:15PM |
B27.00008: Quantum Chemistry on Petascale Machines: Done! What's Next? Edoardo Apra We will illustrate some recent application of parallel code NWChem on Petascale hardware related to material science topics. For example, we will report of the use of embedded cluster approach to model the ground state and excited state properties of crystalline compounds. The methods analyzed in the talk will range from Density-Functional based to wave-function based (e.g. Coupled Cluster). In the second half of the talk we will describe on-going software and algorithmic developments geared towards exploiting the aggregate resources available in upcoming 100 petaflops architectures. [Preview Abstract] |
Monday, March 3, 2014 1:15PM - 1:27PM |
B27.00009: Including short length scale correlations in quantum chemistry methods K. Bhaskaran-Nair, K. Kowalski, J. Moreno, W. Shelton, M. Jarrell Many aspects of computational chemistry and computational material science require accuracies that can only be obtained by a small class of highly accurate computational methods that appropriately account for instantaneous interactions between electrons in molecules or in materials. To aid in addressing the sign problem associated with DMFT based methods we use accurate quantum chemistry methods to treat short length scale correlations within DMFT type formulations and its cluster extensions. The local Green function is obtained from truncated variants of Configuration Interaction and Coupled Cluster methods, which efficiently describe the electron correlation effects. This work is supported by the National Science Foundation award NSF EPS-1003897 with additional support from the Louisiana Board of Regents. [Preview Abstract] |
Monday, March 3, 2014 1:27PM - 1:39PM |
B27.00010: Some lessons learned on the simulation of atomic-scale stochastic processes in carbon systems Vincent Meunier, Colin Daniels, Zachary Bullard The behaviors of many materials are rooted in stochastic processes due to spatial and temporal fluctuations in their nano- and micro- structures. This talk will be the opportunity to present preliminary results on attempts to shed light on the role played by disorder on the dynamical appearance of atomic-scale defects and how these build their way up to mesoscopic length scales and over macroscopic time scales. I will present a simple algorithm that allows translating atomic level properties into scales relevant to devices and materials systems. The algorithm enables the random introduction of elementary mutations in low-dimensional systems and leads to the investigation of the emergence of structures with new functionality and to novel nanostructures resulting from the coalescence of elementary building blocks. The mutations are introduced by local modifications to the connectivity table and are accepted based on a Metropolis algorithm. Externally imposed constraints can be introduced as needed, depending on the actual conditions to be simulated. In addition, the fast prototyping of the effect of mutations on electronic properties is made possible by the ability to enact mutations as perturbation potentials using Dyson equation to update Green functions as mutations are accepted. Results applied to the coalescence, annealing, and phase separation in a number of carbon nanostructures will be shown and compared to experiments when available. [Preview Abstract] |
Monday, March 3, 2014 1:39PM - 1:51PM |
B27.00011: Parallel in time simulations using high level quantum chemistry methods and complex empirical potentials Eric Bylaska, Jonathan Weare, John Weare Algorithms that support parallel decomposition in the time dimension are presented and applied to conventional molecular dynamics (MD) models and {\em ab initio} molecular dynamics (AIMD) models of realistic complexity. The algorithms support convenient parallel implementation to achieve significant improvement in the simulation of high level (e.g. MP2) and excited state dynamics. Given a forward time integrator propagating the system from time $t_i$ (trajectory position and velocity $\mathbf{x}_i =(\mathbf{r}_i,\mathbf{v}_i)$) to $t_{i+1}$ as $\mathbf{x}_{i+1} = \mathbf{f}_i(\mathbf{x}_i)$, the dynamics problem is transformed into a root finding problem, $\mathbf{F} = \mathbf{[x_{i}-f(x_{(i-1})]_{i}}=\mathbf{0}$, for the trajectory variables. The fixed point problem is unconditionally convergent and is solved iteratively using a variety of optimization techniques, including quasi-Newton and preconditioned quasi-Newton methods. The algorithm is parallelized by assigning a processor to each time-step entry in the columns of $\mathbf{F(\mathbf{X})}$. Less accurate but more efficient dynamical models based on simplified interactions or coarsening time-steps provide preconditioners for the root finding problem and lead to an algorithm similar to the parareal algorithm. [Preview Abstract] |
Monday, March 3, 2014 1:51PM - 2:03PM |
B27.00012: Computational Science at the Argonne Leadership Computing Facility Nichols Romero The goal of the Argonne Leadership Computing Facility (ALCF) is to extend the frontiers of science by solving problems that require innovative approaches and the largest-scale computing systems. ALCF's most powerful computer -- Mira, an IBM Blue Gene/Q system -- has nearly one million cores. How does one program such systems? What software tools are available? Which scientific and engineering applications are able to utilize such levels of parallelism? This talk will address these questions and describe a sampling of projects that are using ALCF systems in their research, including ones in nanoscience, materials science, and chemistry. Finally, the ways to gain access to ALCF resources will be presented. [Preview Abstract] |
Monday, March 3, 2014 2:03PM - 2:15PM |
B27.00013: Computational Science with the Titan Supercomputer: Early Outcomes and Lessons Learned Jack Wells Modeling and simulation with petascale computing has supercharged the process of innovation and understanding, dramatically accelerating time-to-insight and time-to-discovery. This presentation will focus on early outcomes from the Titan supercomputer at the Oak Ridge National Laboratory. ~Titan has over 18,000 hybrid compute nodes consisting of both CPUs and GPUs. In this presentation, I will discuss the lessons we have learned in deploying Titan and preparing applications to move from conventional CPU architectures to a hybrid machine. I will present early results of materials applications running on Titan and the implications for the research community as we prepare for exascale supercomputer in the next decade. Lastly, I will provide an overview of user programs at the Oak Ridge Leadership Computing Facility with specific information how researchers may apply for allocations of computing resources. [Preview Abstract] |
Follow Us |
Engage
Become an APS Member |
My APS
Renew Membership |
Information for |
About APSThe American Physical Society (APS) is a non-profit membership organization working to advance the knowledge of physics. |
© 2024 American Physical Society
| All rights reserved | Terms of Use
| Contact Us
Headquarters
1 Physics Ellipse, College Park, MD 20740-3844
(301) 209-3200
Editorial Office
100 Motor Pkwy, Suite 110, Hauppauge, NY 11788
(631) 591-4000
Office of Public Affairs
529 14th St NW, Suite 1050, Washington, D.C. 20045-2001
(202) 662-8700