Bulletin of the American Physical Society
APS March Meeting 2015
Volume 60, Number 1
Monday–Friday, March 2–6, 2015; San Antonio, Texas
Session G23: Focus Session: Petascale Science and Beyond: Applications and Opportunities in Materials Science and Chemistry I |
Hide Abstracts |
Sponsoring Units: DCOMP Chair: Barry Schneider, National Institute for Standards and Technology Room: 202B |
Tuesday, March 3, 2015 11:15AM - 11:51AM |
G23.00001: Challenges for large scale ab initio Quantum Monte Carlo Invited Speaker: Paul Kent Ab initio Quantum Monte Carlo is an electronic structure method that is highly accurate, well suited to large scale computation, and potentially systematically improvable in accuracy. Due to increases in computer power, the method has been applied to systems where established electronic structure methods have difficulty reaching the accuracies desired to inform experiment without empiricism, a necessary step in the design of materials and a helpful step in the improvement of cheaper and less accurate methods. Recent applications include accurate phase diagrams of simple materials through to phenomena in transition metal oxides. Nevertheless there remain significant challenges to achieving a methodology that is robust and systematically improvable in practice, as well as capable of exploiting the latest generation of high-performance computers. In this talk I will describe the current state of the art, recent applications, and several significant challenges for continued improvement. [Preview Abstract] |
Tuesday, March 3, 2015 11:51AM - 12:03PM |
G23.00002: SILK QMC, sign-learning simulations of molecular systems Xiaoyao Ma, Frank Loffler, Karol Kowalski, Randall Hall, Juana Moreno, Mark Jarrell The Sign Learning Kink (SILK) based Quantum Monte Carlo (QMC) is used to calculate the ground state energies for H$_{2}$O, N$_2$ and F$_2$ molecules. This method is based on Feynman's path integral formalism and has two stages. The first, learning stage, reduces the minus sign problem by optimizing the Slater states which are used in the second, QMC stage. We test our method using different vector spaces and compare our results with other Quantum Chemical methods. We also perform exact diagonalization in those vector spaces as a benchmark. In each vector space and for each molecule, we perform SILK QMC for different bond lengths demonstrating that the SILK method is accurate for equilibrium and non-equilibrium geometries. [Preview Abstract] |
Tuesday, March 3, 2015 12:03PM - 12:15PM |
G23.00003: Development and Use of Quantum Chemistry Methods on Intel Many-Integrated Core Units Edoardo Apra We will describe the approach we have taken in porting quantum chemistry algorithms based on local basis functions to the Intel MIC hardware. The implementation completed in the NWChem code shows the feasibility of effectively combining the processing power of traditional CPU architecture and coprocessor hardware on Petascale class computers. Benchmarks of scientific applications will be presented to illustrate the performances of large scale calculations. [Preview Abstract] |
Tuesday, March 3, 2015 12:15PM - 12:27PM |
G23.00004: A scalable sparse eigensolver for petascale applications Murat Keceli, Hong Zhang, Peter Zapol, David Dixon, Albert Wagner Exploiting locality of chemical interactions and therefore sparsity is necessary to push the limits of quantum simulations beyond petascale. However, sparse numerical algorithms are known to have poor strong scaling. Here, we show that shift-and-invert parallel spectral transformations (SIPs) method can scale up to two-hundred thousand cores for density functional based tight-binding (DFTB), or semi-empirical molecular orbital (SEMO) applications. We demonstrated the robustness and scalability of the SIPs method on various kinds of systems including metallic carbon nanotubes, diamond crystals and water clusters. We analyzed how sparsity patterns and eigenvalue spectrums of these different type of applications affect the computational performance of the SIPs. The SIPs method enables us to perform simulations with more than five hundred thousands of basis functions utilizing more than hundreds of thousands of cores. SIPs has a better scaling for memory and computational time in contrast to dense eigensolvers, and it does not require fast interconnects. [Preview Abstract] |
Tuesday, March 3, 2015 12:27PM - 1:03PM |
G23.00005: Simulations of high-Tc superconductors using the DCA$^+$ algorithm Invited Speaker: Peter Staar For over three decades, the high Tc-cuprates have been a gigantic challenge for condensed matter theory. Even the simplest representation of these materials, i.e. the single band Hubbard model, is hard to solve quantitatively and its phase-diagram is therefore elusive. In this talk, we present the recent algorithmic and implementation advances [1,2] to the Dynamical Cluster Approximation (DCA). The algorithmic advances allow us to determine self-consistently a continuous self-energy in momentum space, which in turn reduces the cluster-shape dependency of the superconducting transition temperature and thus accelerates the convergence of the latter versus cluster-size. Furthermore, the introduction of the smooth self-energy suppresses artificial correlations and thus reduces the fermionic sign-problem, allowing us to simulate larger clusters at much lower temperatures. By combining these algorithmic improvements with a very efficient GPU accelerated QMC-solver [3], we are now able to determine the superconducting transition temperature accurately and show that the Cooper-pairs have indeed a d-wave structure, as was predicted by Zhang and Rice. \\[4pt] [1] Peter Staar, Thomas A. Maier and Thomas C. Schulthess ( Phys. Rev. B 88, 115101 (2013) )\\[0pt] [2] Peter Staar, Thomas A. Maier and Thomas C. Schulthess ( Phys. Rev. B 89, 195133 (2014) )\\[0pt] [3] Peter Staar, Thomas A. Maier, Michael S. Summers, Gilles Fourestey, Raffaele Solca, and Thomas C. Schulthess. ``Taking a quantum leap in time to solution for simulations of high-Tc superconductors.'' (In Proceedings of SC13: International Conference for High Performance Computing, Networking, Storage and Analysis, SC '13, pages 1:1-1:11, New York, NY, USA, 2013. ACM.) [Preview Abstract] |
Tuesday, March 3, 2015 1:03PM - 1:15PM |
G23.00006: A New Class of $J_{eff}=1/2$ Mott Insulators Turan Birol, Kristjan Haule We predict a novel class of Jeff=1/2 Mott insulators in a family of Ir and Rh fluoride compounds with the K$_2$GeF$_6$ crystal structure that are previously synthesized, but not characterized extensively. First principles calculations in the level of all electron Density Functional Theory + Dynamical Mean Field Theory (DFT+DMFT) indicate that these compounds have large Mott gaps and some of them exhibit unprecedented proximity to the ideal, SU(2) symmetric Jeff=1/2 limit. [Preview Abstract] |
Tuesday, March 3, 2015 1:15PM - 1:27PM |
G23.00007: Petascale electronic structure code with a new parallel eigensolver Emil Briggs, Wenchang Lu, Miroslav Hodak, Yan Li, CT Kelley, Jerzy Bernholc We describe recent developments within the Real Space Multigrid (RMG) electronic structure code. RMG uses real-space grids, a multigrid pre-conditioner, and subspace diagonalization to solve the Kohn-Sham equations. It is designed for use on massively parallel computers and has shown excellent scalability and performance, reaching 6.5 PFLOPS on 18k Cray compute nodes with 288k CPU cores and 18k GPUs. For large problems, the diagonalization becomes computationally dominant and a novel, highly parallel eigensolver was developed that makes efficient use of a large number of nodes. Test results for a range of problem sizes are presented, which execute up to 3.5 times faster than standard eigensolvers such as Scalapack. RMG is now an open source code, running on Linux, Windows and MacIntosh systems. It may be downloaded at . [Preview Abstract] |
Tuesday, March 3, 2015 1:27PM - 1:39PM |
G23.00008: Full potential KKR approach to the calculation of Hellmann-Feynman force and total energy Yang Wang, G.M. Stocks The Korringa-Kohn-Rostoker (KKR) method is an ab initio electronic structure calculation method based on multiple scattering theory. Unlike the traditional approach, the full-potential KKR method, as well as its linear scaling approach, namely the full-potential LSMS method, does not make a spherical geometry assumption for the LDA potential and the charge density, i.e., the the muffin-tin approximation. Consequently, these full-potential methods allow to calculate the Hellmann-Feynman force acting on each ion in the unit cell. In this presentation, we show an implementation of the full-potential KKR and LSMS methods, discuss the force and total energy calculation in the framework of multiple scattering theory, and finally discuss our approach to overcoming the major computational bottleneck in a full-potential calculation by employing GPGPU acceleration technique. [Preview Abstract] |
Tuesday, March 3, 2015 1:39PM - 1:51PM |
G23.00009: Linear-scaling density-functional theory with wavelets: challenges and opportunities for petascale and beyond Laura Ratcliff, Luigi Genovese, Stephan Mohr, Thierry Deutsch Density-functional theory (DFT) has been used to study a wide range of materials in simulations with a moderate level of parallelism. A common approach divides the electronic orbitals between MPI tasks, however this limits the number of tasks that can be used for a given system. The most straightforward path to exploiting petascale machines is therefore to increase the size of the system being studied. However, standard implementations of DFT scale cubically with the number of atoms so that the time rapidly increases for large systems. Algorithms must therefore be designed with reduced scaling, such as the linear-scaling approach in BigDFT, which uses an adaptive localized basis set that is itself represented in an underlying wavelet basis set. It thus retains all the benefits of wavelets, such as systematic convergence, while also presenting some new advantages, e.g. the definition of a fragment approach. Nonetheless, as we move towards the exascale, there remain a number of challenges associated both with increasing parallelism and the treatment of large systems. We will outline the algorithms and parallelization used in BigDFT and present some recent results which have been facilitated by this approach, as well as discussing some of the future challenges. [Preview Abstract] |
Tuesday, March 3, 2015 1:51PM - 2:03PM |
G23.00010: Functional derivatives for multi-scale modeling Samuel Reeve, Alejandro Strachan As we look beyond petascale computing and towards the exascale, effectively utilizing computational resources by using multi-fidelity and multi-scale materials simulations becomes increasingly important. Determining when and where to run high-fidelity simulations in order to have the most effect on a given quantity of interest (QoI) is a difficult problem. This work utilizes functional uncertainty quantification (UQ) for this task. While most UQ focuses on uncertainty in output from uncertainty in input parameters, we focus on uncertainty from the function itself (e.g. from using a specific functional form for an interatomic potential or constitutive law). In the case of a multi-scale simulation with a given constitutive law, calculating the functional derivative of the QoI with respect to that constitutive law can determine where a fine-scale model evaluation will maximize the increase in accuracy of the predicted QoI. Additionally, for a given computational budget the optimal set of coarse and fine-scale simulations can be determined. Numerical calculation of the functional derivative has been developed and methods of including this work within existing multi-fidelity and multi-scale orchestrators are explored. [Preview Abstract] |
Tuesday, March 3, 2015 2:03PM - 2:15PM |
G23.00011: DMRG study of Many-Body Localization Xiongjie Yu, Bryan Clark, David Pekker Numerical studies on many-body localization (MBL) problems have heavily relied on exact diagonalization (ED) techniques so far which has severely limited the system size that can be studied. Here we report a density matrix renormalization group (DMRG) based method for simulations in the many-body localized phase allowing us to reach system sizes inaccessible to ED. We describe our techniques and report on our results applying DMRG to larger systems. [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. |
© 2020 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
1 Research Road, Ridge, NY 11961-2701
(631) 591-4000
Office of Public Affairs
529 14th St NW, Suite 1050, Washington, D.C. 20045-2001
(202) 662-8700