skip to main content
OSTI.GOV title logo U.S. Department of Energy
Office of Scientific and Technical Information

Title: Multilevel domain decomposition for electronic structure calculations

Abstract

We introduce a new multilevel domain decomposition method (MDD) for electronic structure calculations within semi-empirical and density functional theory (DFT) frameworks. This method iterates between local fine solvers and global coarse solvers, in the spirit of domain decomposition methods. Using this approach, calculations have been successfully performed on several linear polymer chains containing up to 40,000 atoms and 200,000 atomic orbitals. Both the computational cost and the memory requirement scale linearly with the number of atoms. Additional speed-up can easily be obtained by parallelization. We show that this domain decomposition method outperforms the density matrix minimization (DMM) method for poor initial guesses. Our method provides an efficient preconditioner for DMM and other linear scaling methods, variational in nature, such as the orbital minimization (OM) procedure.

Authors:
 [1];  [2];  [3];  [4]
  1. EDF R and D, 1 avenue du General de Gaulle, 92141 Clamart Cedex (France) and CERMICS, Ecole Nationale des Ponts et Chaussees, 6 and 8, Avenue Blaise Pascal, Cite Descartes, 77455 Marne-La-Vallee Cedex 2 (France). E-mail: maxime.barrault@edf.fr
  2. CERMICS, Ecole Nationale des Ponts et Chaussees, 6 and 8, Avenue Blaise Pascal, Cite Descartes, 77455 Marne-La-Vallee Cedex 2 (France). E-mail: cances@cermics.enpc.fr
  3. Department of Mathematics, University of Florida, Gainesville, FL 32611-8105 (United States). E-mail: hager@math.ufl.edu
  4. CERMICS, Ecole Nationale des Ponts et Chaussees, 6 and 8, Avenue Blaise Pascal, Cite Descartes, 77455 Marne-La-Vallee Cedex 2 (France). E-mail: lebris@cermics.enpc.fr
Publication Date:
OSTI Identifier:
20991560
Resource Type:
Journal Article
Resource Relation:
Journal Name: Journal of Computational Physics; Journal Volume: 222; Journal Issue: 1; Other Information: DOI: 10.1016/j.jcp.2006.06.049; PII: S0021-9991(06)00318-4; Copyright (c) 2006 Elsevier Science B.V., Amsterdam, The Netherlands, All rights reserved; Country of input: International Atomic Energy Agency (IAEA)
Country of Publication:
United States
Language:
English
Subject:
71 CLASSICAL AND QUANTUM MECHANICS, GENERAL PHYSICS; COMPUTERIZED SIMULATION; DENSITY FUNCTIONAL METHOD; DENSITY MATRIX; ELECTRONIC STRUCTURE; MATHEMATICAL MODELS; MINIMIZATION; POLYMERS

Citation Formats

Barrault, M., Cances, E., Hager, W.W., and Le Bris, C.. Multilevel domain decomposition for electronic structure calculations. United States: N. p., 2007. Web. doi:10.1016/j.jcp.2006.06.049.
Barrault, M., Cances, E., Hager, W.W., & Le Bris, C.. Multilevel domain decomposition for electronic structure calculations. United States. doi:10.1016/j.jcp.2006.06.049.
Barrault, M., Cances, E., Hager, W.W., and Le Bris, C.. Thu . "Multilevel domain decomposition for electronic structure calculations". United States. doi:10.1016/j.jcp.2006.06.049.
@article{osti_20991560,
title = {Multilevel domain decomposition for electronic structure calculations},
author = {Barrault, M. and Cances, E. and Hager, W.W. and Le Bris, C.},
abstractNote = {We introduce a new multilevel domain decomposition method (MDD) for electronic structure calculations within semi-empirical and density functional theory (DFT) frameworks. This method iterates between local fine solvers and global coarse solvers, in the spirit of domain decomposition methods. Using this approach, calculations have been successfully performed on several linear polymer chains containing up to 40,000 atoms and 200,000 atomic orbitals. Both the computational cost and the memory requirement scale linearly with the number of atoms. Additional speed-up can easily be obtained by parallelization. We show that this domain decomposition method outperforms the density matrix minimization (DMM) method for poor initial guesses. Our method provides an efficient preconditioner for DMM and other linear scaling methods, variational in nature, such as the orbital minimization (OM) procedure.},
doi = {10.1016/j.jcp.2006.06.049},
journal = {Journal of Computational Physics},
number = 1,
volume = 222,
place = {United States},
year = {Thu Mar 01 00:00:00 EST 2007},
month = {Thu Mar 01 00:00:00 EST 2007}
}
  • This study investigates algebraic multilevel domain decomposition preconditioners of the Schwarz type for solving linear systems associated with Newton-Krylov methods. The key component of the preconditioner is a coarse approximation based on algebraic multigrid ideas to approximate the global behavior of the linear system. The algebraic multilevel preconditioner is based on an aggressive coarsening graph partitioning of the non-zero block structure of the Jacobian matrix. The scalability of the preconditioner is presented as well as comparisons with a two-level Schwarz preconditioner using a geometric coarse grid operator. These comparisons are obtained on large-scale distributed-memory parallel machines for systems arising frommore » incompressible flow and transport using a stabilized finite element formulation. The results demonstrate the influence of the smoothers and coarse level solvers for a set of 3D example problems. For preconditioners with more than one level, careful attention needs to be given to the balance of robustness and convergence rate for the smoothers and the cost of applying these methods. For properly chosen parameters, the two- and three-level preconditioners are demonstrated to be scalable to 1024 processors.« less
  • In this paper, we investigate a novel approach based on the combination of Tucker-type and canonical tensor decomposition techniques for the efficient numerical approximation of functions and operators in electronic structure calculations. In particular, we study applicability of tensor approximations for the numerical solution of Hartree-Fock and Kohn-Sham equations on 3D Cartesian grids. We show that the orthogonal Tucker-type tensor approximation of electron density and Hartree potential of simple molecules leads to low tensor rank representations. This enables an efficient tensor-product convolution scheme for the computation of the Hartree potential using a collocation-type approximation via piecewise constant basis functions onmore » a uniform nxnxn grid. Combined with the Richardson extrapolation, our approach exhibits O(h{sup 3}) convergence in the grid-size h=O(n{sup -1}). Moreover, this requires O(3rn+r{sup 3}) storage, where r denotes the Tucker rank of the electron density with r=O(logn), almost uniformly in n. For example, calculations of the Coulomb matrix and the Hartree-Fock energy for the CH{sub 4} molecule, with a pseudopotential on the C atom, achieved accuracies of the order of 10{sup -6} hartree with a grid-size n of several hundreds. Since the tensor-product convolution in 3D is performed via 1D convolution transforms, our scheme markedly outperforms the 3D-FFT in both the computing time and storage requirements.« less
  • We present a fully grid-based approach for solving Hartree–Fock and all-electron Kohn–Sham equations based on low-rank approximation of three-dimensional electron orbitals. Due to the low-rank structure the total complexity of the algorithm depends linearly with respect to the one-dimensional grid size. Linear complexity allows for the usage of fine grids, e.g. 8192{sup 3} and, thus, cheap extrapolation procedure. We test the proposed approach on closed-shell atoms up to the argon, several molecules and clusters of hydrogen atoms. All tests show systematical convergence with the required accuracy.
  • Methods are proposed for the efficient parallel solution of nonlinear nodal kinetics equations. Because the two-node calculation in the nonlinear nodal method is naturally parallelizable, the majority of the effort is devoted to the development of parallel methods for solving the coarse-mesh finite difference (CMFD) problem. A preconditioned Krylov subspace method (biconjugate gradient stabilized) is chosen as the iterative algorithm for the CMFD problem, and an efficient parallel preconditioning scheme is developed based on domain decomposition techniques. An incomplete lower-upper triangle factorization method is first formulated for the coefficient matrices representing each three-dimensional subdomain, and coupling between subdomains is thenmore » approximated by incorporating only the effect of the nonleakage terms of neighboring subdomains. The methods are applied to fixed-source problems created from the International Atomic Energy Agency three-dimensional benchmark problem. The effectiveness of the incomplete domain decomposition preconditioning on a multiprocessor is evidenced by the small increase in the number of iterations as the number of subdomains increases. Through the application to both CMFD-only and nodal calculations, it is demonstrated that speedups as large as 49 with 96 processors are attainable in the nonlinear nodal kinetics calculations.« less