Fast Sparse Cholesky Decomposition and Inversion using Nested Dissection Matrix Reordering

被引:15
|
作者
Brandhorst, Kai [1 ]
Head-Gordon, Martin [1 ,2 ]
机构
[1] Univ Calif Berkeley, Dept Chem, Berkeley, CA 94720 USA
[2] Lawrence Berkeley Natl Lab, Div Chem Sci, Berkeley, CA USA
关键词
ELECTRONIC-STRUCTURE CALCULATIONS; PLESSET CORRELATION-ENERGY; DEGREE ORDERING ALGORITHM; DENSITY-MATRIX; NULL SPACE; DIRECT OPTIMIZATION; TENSOR FORMULATION; ELIMINATION TREES; MOLECULAR-SYSTEMS; HARTREE-FOCK;
D O I
10.1021/ct100618s
中图分类号
O64 [物理化学(理论化学)、化学物理学];
学科分类号
070304 ; 081704 ;
摘要
Here we present an efficient, yet nonlinear scaling, algorithm for the computation of Cholesky factors of sparse symmetric positive definite matrices and their inverses. The key feature of this implementation is the separation of the task into an algebraic and a numeric part. The algebraic part of the algorithm attempts to find a reordering of the rows and columns which preserves at least some degree of sparsity and afterward determines the exact nonzero structure of both the Cholesky factor and its corresponding inverse. It is based on graph theory and does not involve any kind of numerical thresholding. This preprocessing then allows for a very efficient implementation of the numerical factorization step. Furthermore this approach even allows use of highly optimized dense linear algebra kernels which leads to yet another performance boost. We will show some illustrative timings of our sparse code and compare it to the standard library implementation and a recent sparse implementation using thresholding. We conclude with some comments on how to deal with positive semidefinite matrices.
引用
收藏
页码:351 / 368
页数:18
相关论文
共 50 条
  • [1] Matrix Inversion Using Cholesky Decomposition
    Krishnamoorthy, Aravindh
    Menon, Deepak
    2013 SIGNAL PROCESSING: ALGORITHMS, ARCHITECTURES, ARRANGEMENTS, AND APPLICATIONS (SPA), 2013, : 70 - 72
  • [2] A Reconfigurable Processing Element Implementation for Matrix Inversion Using Cholesky Decomposition
    Happonen, Aki
    Burian, Adrian
    Hemming, Erwin
    PROCEEDINGS OF WORLD ACADEMY OF SCIENCE, ENGINEERING AND TECHNOLOGY, VOL 3, 2005, 3 : 114 - 117
  • [3] A fixed-point implementation of matrix inversion using Cholesky decomposition
    Burian, A
    Takala, J
    Ylinen, M
    PROCEEDINGS OF THE 46TH IEEE INTERNATIONAL MIDWEST SYMPOSIUM ON CIRCUITS & SYSTEMS, VOLS 1-3, 2003, : 1431 - 1434
  • [4] SYMINV - ALGORITHM FOR INVERSION OF A POSITIVE DEFINITE MATRIX BY CHOLESKY DECOMPOSITION
    SEAKS, T
    ECONOMETRICA, 1972, 40 (05) : 961 - 962
  • [5] Fast convolutional sparse coding using matrix inversion lemma
    Sorel, Michal
    Sroubek, Filip
    DIGITAL SIGNAL PROCESSING, 2016, 55 : 44 - 51
  • [6] Fast and accurate pseudoinverse with sparse matrix reordering and incremental approach
    Jung, Jinhong
    Sael, Lee
    MACHINE LEARNING, 2020, 109 (12) : 2333 - 2347
  • [7] Fast and accurate pseudoinverse with sparse matrix reordering and incremental approach
    Jinhong Jung
    Lee Sael
    Machine Learning, 2020, 109 : 2333 - 2347
  • [8] COMMENT ON SYMINV - ALGORITHM FOR INVERSION OF POSITIVE DEFINITE MATRIX BY CHOLESKY DECOMPOSITION
    STEWART, J
    ECONOMETRICA, 1974, 42 (04) : 771 - 771
  • [9] A new nested Cholesky decomposition and estimation for the covariance matrix of bivariate longitudinal data
    Feng, Sanying
    Lian, Heng
    Xue, Liugen
    COMPUTATIONAL STATISTICS & DATA ANALYSIS, 2016, 102 : 98 - 109
  • [10] Enhancing the Sparse Matrix Storage Using Reordering Techniques
    Freire, Manuel
    Marichal, Raul
    Gonzaga de Oliveira, Sanderson L.
    Dufrechou, Ernesto
    Ezzatti, Pablo
    HIGH PERFORMANCE COMPUTING, CARLA 2023, 2024, 1887 : 66 - 76