Bulletin of the American Physical Society
APS March Meeting 2020
Volume 65, Number 1
Monday–Friday, March 2–6, 2020; Denver, Colorado
Session F40: Building the bridge to exascale: applications and opportunities for materials, chemistry, and biology IVFocus
|
Hide Abstracts |
Sponsoring Units: DCOMP DCMP DAMOP DMP Chair: James Belak, Lawrence Livermore Natl Lab Room: 705 |
Tuesday, March 3, 2020 8:00AM - 8:36AM |
F40.00001: Towards Exascale Electronic Structure and Quantum Transport Calculations Invited Speaker: Jerry Bernholc The development of robust, adaptive software and algorithms that can fully exploit exascale capabilities and future computing architectures is critical to designing advanced materials and devices with targeted properties. We have developed an open-source code that discretizes the DFT equations on real-space grids that are distributed over the nodes of a massively parallel system via domain decomposition. Multigrid techniques are used to dramatically accelerate convergence while only requiring nearest-neighbor communications. The real-space multigrid (RMG) code achieves full plane wave accuracy and scales from desktops and clusters to supercomputers consisting of ~200k cores and 20k GPUs, including the Cray XE-XK systems and the new IBM-NVIDIA pre-exascale Summit. Multilevel parallelization with MPI, threads and/or Cuda/HIP programming enables adaptation to future exascale supercomputers. RMG is distributed via www.rmgdft.org, with over 3,800 downloads do date. Advanced functionalities are provided through interfaces to other codes, including QMCPACK, BerkeleyGW, Phonopy, and ALAMODE. RMG is also being used for high throughput vibrational analysis of large systems at the Spallation Neutron Source. We will also describe the non-equilibrium Green’s function module based on variationally-optimized localized orbitals, by which quantum transport properties can be studied for devices containing tens of thousands of atoms with full DFT accuracy. For a system with ten thousand atoms, our initial implementation scales linearly from 100 to 1000 nodes on Summit, already gaining ~4x speed-up from GPUs over CPU-only calculations. Several applications will be described as time permits, including a nanocircuit that could potentially enable electrical sequencing of DNA, and a novel experimentally realizable graphene-nanoribbon-based negative differential resistance device. |
Tuesday, March 3, 2020 8:36AM - 8:48AM |
F40.00002: Towards first-principles design of quantum devices Jacek Jakowski, Jerry Bernholc, Mina Yoon, David Lingerfelt, Wenchang Lu, Emil Briggs With the Moore’s Law in high-performance computing approaching limits, future progress relies on harnessing quantum effects. Modeling quantum behavior at device scale requires an exascale computer and software that fully utilizes its extraordinary capabilities. Much of current research exploring future directions for nanoscale electronics is focused on two-dimensional materials. Evaluation of future nanodevices is very difficult and it it requires simulations involving thousands of atoms. Advanced quantum-mechanical simulations can be used to develop novel device concepts and identify the appropriate combinations of atomic constituents and nanoscale structures, thereby radically accelerating progress. We discuss the implementation of time-dependent electron dynamics in the real-space multigrid (RMG) density functional theory software package. The use of multiresolution grids permits real-space description of the electronic structure of systems comprised of hundreds-to-thousands of atoms, allowing atomistic description of systems approaching realistic device scales that can include arbitrary defects/impurities. Some results for quantum materials are presented that we believe could be the focus of future quantum computing and quantum information technologies. |
Tuesday, March 3, 2020 8:48AM - 9:00AM |
F40.00003: GPU-Acceleration of the ELPA2 Distributed Eigensolver for Applications in Electronic Structure Theory Victor Yu, Jonathan Moussa, Volker Blum The solution of eigenproblems is often a key computational bottleneck that limits the tractable system size of electronic structure theory. For large systems, these eigenproblems can easily exceed the capacity of a single computer, thus must be solved on distributed-memory parallel computers. The ELSI library facilitates large-scale electronic structure calculations by providing a unified interface to various fast and scalable eigensolvers and density matrix solvers, including the EigenExa, ELPA, libOMM, NTPoly, PEXSI, and SLEPc libraries. The ubiquitous adoption of hybrid CPU-GPU nodes in supercomputing opens up new opportunities to accelerate electronic structure calculations. We here present GPU-oriented optimizations of the ELPA two-stage tridiagonalization eigensolver (ELPA2). On top of its existing cuBLAS-based GPU offloading, we add a CUDA kernel to speed up the back-transformation of eigenvectors, which was known as the main bottleneck of the two-stage tridiagonalization algorithm. CPU, GPU, and MPI activities are overlapped wherever possible. Robust choices that maximize the GPU compute intensity are identified. We demonstrate the performance of this GPU-accelerated eigensolver by a set of benchmark calculations. |
Tuesday, March 3, 2020 9:00AM - 9:12AM |
F40.00004: Driving exascale computational science with AiiDA Sebastiaan Huber, Giovanni Pizzi, Nicola Marzari The ever-growing availability of computational power and sustained development of advanced computational methods have contributed much to recent scientific progress. |
Tuesday, March 3, 2020 9:12AM - 9:24AM |
F40.00005: DMFTwDFT: A free-license DMFT package interfaced with various DFT implementations Hyowon Park, Vijay Singh, Uthpala Herath, Benny Wah, Xingyu Liao, Aldo H Romero The electronic and structural characterization of strongly correlated materials (SCMs) is one of the most difficult problems in computational materials science. Dynamical Mean Field Theory (DMFT) has been successful in treating local many-body effects of correlated orbitals in SCMs, particularly when it is combined with density functional theory (DFT). In this talk, we present an open-source computational package (DMFTwDFT) combining DMFT with various DFT codes interfaced to a Wannier90 package for adopting maximally localized Wannier functions as local orbitals to describe a correlated subspace. We also provide the library mode for computing a DMFT density matrix such that our package can be efficiently linked to various DFT codes and achieves the charge-self-consistency within DFT+DMFT loops. We used our code for the study of well-known correlated materials, namely LaNiO3, SrVO3, and NiO to compute the density of states, the band structure, the total energy, the atomic force, and the Fermi surface within DFT+DMFT, and also compared our results to those obtained from other DFT+DMFT codes. |
Tuesday, March 3, 2020 9:24AM - 9:36AM |
F40.00006: Scaling and Performance of Real-Space Electronic Structure Calculations on Exascale Architectures Emil Briggs, Wenchang Lu, Jerry Bernholc Electronic structure calculations are hard to scale on massively parallel systems, and these challenges are compounded on CPU-GPU exascale architectures. Real-space formulations enable easy parallelization via domain decomposition, which has been implemented in our open-source real-space multigrid code (RMG). RMG has been designed to perform well on leading-edge supercomputers from inception. However, scaling to extreme sizes and to distributed multi-CPU-GPU architectures requires careful consideration of data distribution and flow, including inter-node transfers as well as between CPUs and GPUs located on the same node. The large mismatches between CPU and GPU clock speeds and FLOP rates provides additional constraints as well as optimization opportunities. We will discuss efficient and scalable implementation of distributed data flow and key electronic structure algorithms on exascale class machines in RMG, as well as methods for addressing MPI scalability constraints and data bottlenecks. RMG source code and build scripts for pre-exascale Summit, Cray XE-XK, clusters, Linux, Windows, and MacOS workstations are available at www.rmgdft.org together with help files and examples. |
Tuesday, March 3, 2020 9:36AM - 9:48AM |
F40.00007: From LSMS to MuST: Large scale first principles materials calculations at the exascale Markus Eisenbach, Xianglin Liu, Khorgolkhuu Odbadrakh, Yang Wang We present recent advances in our Locally-selfconsistent Multiple Scattering (LSMS) code for scalable first principles density functional calculations of materials. A fundamental science drivers for large scale calculations is the need to understand materials beyond periodic crystalline lattices. Due to the large simulation cells of many thousands of atoms needed to describe complex electronic and magnetic ordering, defect states or disorder in alloys, the cubic scaling of traditional first principles methods has prevented direct first principles calculations. The real space formalism of LSMS enables calculations for O(100,000) atom. In preparation for exascale systems, we are extending the use of accelerators to the calculation of forces and embedding methods for disordered systems. We will present results and performance measurements for defects in high entropy alloys and non-collinear magnetism in disordered systems. The computational capabilities will be available in our Multiple Scattering Theory suite (MuST) [https://github.com/mstsuite] |
Tuesday, March 3, 2020 9:48AM - 10:00AM |
F40.00008: Mixed precision sampling of quantum states of matter Thomas Maier, Giovanni Balduzzi, Urs R Haehner, Ying Wai Li, Arghya Chatterjee, Ed D'Azevedo Monte Carlo simulations are widely used throughout all areas of science. In materials science, they provide an important framework to unravel the mechanisms that give rise to complex behavior and different quantum states of matter. The DCA++ code is a high-performance research application that solves quantum many-body, materials problems with a cutting edge quantum Monte Carlo based dynamic cluster approximation (DCA). Here we discuss how mixed precision arithmetics finds a natural home in statistical sampling based Monte Carlo, and how it is being used in DCA++ to boost performance on ORNL’s Summit supercomputer. |
Tuesday, March 3, 2020 10:00AM - 10:12AM |
F40.00009: Towards finite-temperature tensor network simulations of the two-dimensional Hubbard model Alexander Wietek, Miles Stoudenmire The phase diagram of the two-dimensional Hubbard model at finite temperature poses one of the most interesting conundrums in contemporary condensed matter physics. Tensor network techniques, such as matrix-product based approaches as well as 2D tensor networks (PEPS), yield state-of-the-art unbiased simulations of the 2D Hubbard model at zero temperature and are capable of giving unbiased results at finite temperature as well. A promising approach for applying tensor networks to study finite-temperature quantum systems is the minimally entangled typical thermal state (METTS) algorithm, which is a Monte Carlo technique that samples from a family of entangled wavefunctions, and which offers favorable scaling and parallelism. We demonstrate how the METTS algorithm in combination with modern time-evolution algorithms for matrix-product states, like the time-dependent variational principle (TDVP) method, allows simulating the Hubbard model at finite temperature for cylinder geometries approaching the two-dimensional limit. |
Tuesday, March 3, 2020 10:12AM - 10:24AM |
F40.00010: Comparison of systematically improvable DMC and AFQMC for condensed matter Anouar Benali, Fionn Malone, Miguel A Morales, Luke Shulenburger Diffusion Monte Carlo (DMC) and Auxiliary Field Quantum Monte Carlo (AFQMC) can both have their approximations systematically improved by applying successively more accurate trial wavefunctions. In this work we assess the cost and feasibility of determining exact total energies for solid state Hamiltonians by studying primitive cells of four representative materials, Al, LiF, C and TiO2. Specifically, we utilize multideterminant trial wavefunctions generated via selective CI techniques with various sizes of single particle basis. Attention is paid to the rate at which the error decreases as the trial wavefunction includes more determinants and also to the cost as the basis increases in size. In this way, we are able to compare both DMC and AFQMC on equal terms at identical levels of approximation. |
Tuesday, March 3, 2020 10:24AM - 10:36AM |
F40.00011: Quantum Monte Carlo with orbital optimization applied on solids Ye Luo The accuracy and efficiency of quantum Monte Carlo (QMC) methods can be improved directly by improving many body wavefunction. Full wavefunction optimization introduced a decade ago enabled solving scientific challenges beyond chemical accuracy. Recently algorithmic development pushed the computational efficiency of wavefunction derivatives and thus paved the way for simulation with large electron counts although demonstrations are mostly limited to molecular systems. When applying QMC in solids, using fixed orbitals calculated by density functional theory (DFT) together with variationally optimized Jastrow factors is still the common practice. Recent study on bandgaps shows qualitative improvement by simply adding limited multi determinant expansion. This hints the necessity of improving single particle orbitals restricted by DFT. Thus, we enable orbital optimization schemes like mixing occupied and unoccupied orbitals or directly optimizing orbital shapes and study their strength and weakness on solid state systems. |
Tuesday, March 3, 2020 10:36AM - 10:48AM |
F40.00012: Auxiliary-Field Quantum Monte Carlo Applied to Correlated Materials Fionn Malone, Joonho Lee, Shuai Zhang, Miguel A Morales In this talk we investigate the application of auxiliary-field quantum Monte Carlo (AFQMC) to correlated materials. Focusing on elemental transition metals and their oxides we systematically investigate the effect of trial wavefunction and basis set quality on the energetics and static properties computed from AFQMC. We show how recent algorithmic developments allow us to reliably reach the thermodynamic limit and compare directly to experiment. |
Tuesday, March 3, 2020 10:48AM - 11:00AM |
F40.00013: Magnetic and charge orders in the ground state of the 2D repulsive Hubbard model Hao Xu, Mingpu Qin, Hao Shi, Yuan-Yao He, Shiwei Zhang Using the auxiliary field quantum Monte-Carlo (AFQMC) method combined with a self-consistent constraint [1], we systematically study the ground state of the two-dimensional repulsive Hubbard model as a function of doping and interaction strengths. We focus on the spin and charge correlations, applying pinning fields to break translational symmetry and computing the spin and charge densities to probe the emergence of long-range order and collective modes. A phase diagram of spin-density and charge-density waves (including stripes) is obtained. |
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