DOCSLIB.ORG

AMS526: Numerical Analysis I (Numerical Linear Algebra) Lecture 16: Rayleigh Quotient Iteration

Total Page：16

File Type：pdf, Size：1020Kb

Download full-text PDF Read full-text
AMS526: Numerical Analysis I (Numerical Linear Algebra) Lecture 16: Rayleigh Quotient Iteration AMS526: Numerical Analysis I (Numerical Linear Algebra) Lecture 16: Rayleigh Quotient Iteration Xiangmin Jiao SUNY Stony Brook Xiangmin Jiao Numerical Analysis I 1 / 10 Solving Eigenvalue Problems All eigenvalue solvers must be iterative Iterative algorithms have multiple facets: 1 Basic idea behind the algorithms 2 Convergence and techniques to speed-up convergence 3 Efficiency of implementation 4 Termination criteria We will focus on first two aspects Xiangmin Jiao Numerical Analysis I 2 / 10 Simplification: Real Symmetric Matrices We will consider eigenvalue problems for real symmetric matrices, i.e. T m×m m A = A 2 R , and Ax = λx for x 2 R p ∗ T T I Note: x = x , and kxk = x x A has real eigenvalues λ1,λ2, ::: , λm and orthonormal eigenvectors q1, q2, ::: , qm, where kqj k = 1 Eigenvalues are often also ordered in a particular way (e.g., ordered from large to small in magnitude) In addition, we focus on symmetric tridiagonal form I Why? Because phase 1 of two-phase algorithm reduces matrix into tridiagonal form Xiangmin Jiao Numerical Analysis I 3 / 10 Rayleigh Quotient m The Rayleigh quotient of x 2 R is the scalar xT Ax r(x) = xT x For an eigenvector x, its Rayleigh quotient is r(x) = xT λx=xT x = λ, the corresponding eigenvalue of x For general x, r(x) = α that minimizes kAx − αxk2. 2 x is eigenvector of A() rr(x) = x T x (Ax − r(x)x) = 0 with x 6= 0 r(x) is smooth and rr(qj ) = 0 for any j, and therefore is quadratically accurate: 2 r(x) − r(qJ ) = O(kx − qJ k ) as x ! qJ for some J Xiangmin Jiao Numerical Analysis I 4 / 10 Power Iteration Simple power iteration for largest eigenvalue Algorithm: Power Iteration v (0) =some unit-length vector for k = 1; 2;::: w = Av (k−1) v (k) = w=kwk λ(k) = r(v (k)) = (v (k))T Av (k) Termination condition is omitted for simplicity Xiangmin Jiao Numerical Analysis I 5 / 10 Convergence of Power Iteration (0) k Expand initial v in orthonormal eigenvectors qi , and apply A : (0) v = a1q1 + a2q2 + ··· + amqm (k) k (0) v = ck A v k k k = ck (a1λ1 q1 + a2λ2 q2 + ··· + amλmqm) k k k = ck λ1 (a1q1 + a2(λ2/λ1) q2 + ··· + am(λm/λ1) qm) T (0) If jλ1j > jλ2j ≥ · · · ≥ jλmj ≥ 0 and q1 v 6= 0, this gives (k) k (k) 2k kv − (±q1)k = O jλ2/λ1j ; jλ − λ1j = O jλ2/λ1j T (k) where ± sign is chosen to be sign of q1 v It finds the largest eigenvalue (unless eigenvector is orthogonal to v (0)) Error reduces by only a constant factor (≈ jλ2/λ1j) each step, and very slowly especially when jλ2j ≈ jλ1j Xiangmin Jiao Numerical Analysis I 6 / 10 Inverse Iteration −1 −1 Apply power iteration on (A − µI ) , with eigenvalues f(λj − µ) g −1 If µ ≈ λJ for some J, then (λJ − µ) may be far larger than −1 (λj − µ) , j 6= J, so power iteration may converge rapidly Algorithm: Inverse Iteration v (0) =some unit-length vector for k = 1; 2;::: Solve (A − µI )w = v (k−1) for w v (k) = w=kwk λ(k) = r(v (k)) = (v (k))T Av (k) Converges to eigenvector qJ if parameter µ is close to λJ k ! 2k ! (k) µ − λJ (k) µ − λJ kv − (±qJ )k = O ; jλ − λJ j = O µ − λK µ − λK where λJ and λK are closest and second closest eigenvalues to µ Standard method for determining eigenvector given eigenvalue Xiangmin Jiao Numerical Analysis I 7 / 10 Rayleigh Quotient Iteration Parameter µ is constant in inverse iteration, but convergence is better for µ close to the eigenvalue Improvement: At each iteration, set µ to last computed Rayleigh quotient Algorithm: Rayleigh Quotient Iteration v (0) =some unit-length vector λ(0) = r(v (0)) = (v (0))T Av (0) for k = 1; 2;::: Solve (A − λ(k−1)I )w = v (k−1) for w v (k) = w=kwk λ(k) = r(v (k)) = (v (k))T Av (k) Cost per iteration is linear for tridiagonal matrix Xiangmin Jiao Numerical Analysis I 8 / 10 Convergence of Rayleigh Quotient Iteration Cubic convergence in Rayleigh quotient iteration (k+1) (k) 3 kv − (±qJ )k = O(kv − (±qJ )k ) and (k+1) (k) 3 jλ − λJ j = O jλ − λJ j In other words, each iteration triples number of digits of accuracy (k) (k) Proof idea: If v is close to an eigenvector, kv − (±qJ )k ≤ , then accuracy of Rayleigh quotient estimate λ(k) is (k) 2 jλ − λJ j = O( ). One step of inverse iteration then gives (k+1) (k) (k) 3 kv − qJ k = O(jλ − λJ jkv − qJ k) = O( ) Rayleigh quotient is great in finding largest (or smallest) eigenvalue and its corresponding eigenvector. What if we want to find all eigenvalues? Xiangmin Jiao Numerical Analysis I 9 / 10 Operation Counts In Rayleigh quotient iteration, m×m (k−1) if A 2 R is full matrix, then solving (A − µI )w = v may take O(m3) flops per step m×m 2 if A 2 R is upper Hessenberg, then each step takes O(m ) flops m×m if A 2 R is tridiagonal, then each step takes O(m) flops Xiangmin Jiao Numerical Analysis I 10 / 10.
Load more

Recommended publications
  • Efficient Algorithms for High-Dimensional Eigenvalue
    Efficient Algorithms for High-dimensional Eigenvalue Problems by Zhe Wang Department of Mathematics Duke University Date: Approved: Jianfeng Lu, Advisor Xiuyuan Cheng Jonathon Mattingly Weitao Yang Dissertation submitted in partial fulfillment of the requirements for the degree of Doctor of Philosophy in the Department of Mathematics in the Graduate School of Duke University 2020 ABSTRACT Efficient Algorithms for High-dimensional Eigenvalue Problems by Zhe Wang Department of Mathematics Duke University Date: Approved: Jianfeng Lu, Advisor Xiuyuan Cheng Jonathon Mattingly Weitao Yang An abstract of a dissertation submitted in partial fulfillment of the requirements for the degree of Doctor of Philosophy in the Department of Mathematics in the Graduate School of Duke University 2020 Copyright © 2020 by Zhe Wang All rights reserved Abstract The eigenvalue problem is a traditional mathematical problem and has a wide appli- cations. Although there are many algorithms and theories, it is still challenging to solve the leading eigenvalue problem of extreme high dimension. Full configuration interaction (FCI) problem in quantum chemistry is such a problem. This thesis tries to understand some existing algorithms of FCI problem and propose new efficient algorithms for the high-dimensional eigenvalue problem. In more details, we first es- tablish a general framework of inexact power iteration and establish the convergence theorem of full configuration interaction quantum Monte Carlo (FCIQMC) and fast randomized iteration (FRI). Second, we reformulate the leading eigenvalue problem as an optimization problem, then compare the show the convergence of several coor- dinate descent methods (CDM) to solve the leading eigenvalue problem. Third, we propose a new efficient algorithm named Coordinate descent FCI (CDFCI) based on coordinate descent methods to solve the FCI problem, which produces some state-of- the-art results.
    [Show full text]
  • Implicitly Restarted Arnoldi/Lanczos Methods for Large Scale Eigenvalue Calculations
    https://ntrs.nasa.gov/search.jsp?R=19960048075 2020-06-16T03:31:45+00:00Z NASA Contractor Report 198342 /" ICASE Report No. 96-40 J ICA IMPLICITLY RESTARTED ARNOLDI/LANCZOS METHODS FOR LARGE SCALE EIGENVALUE CALCULATIONS Danny C. Sorensen NASA Contract No. NASI-19480 May 1996 Institute for Computer Applications in Science and Engineering NASA Langley Research Center Hampton, VA 23681-0001 Operated by Universities Space Research Association National Aeronautics and Space Administration Langley Research Center Hampton, Virginia 23681-0001 IMPLICITLY RESTARTED ARNOLDI/LANCZOS METHODS FOR LARGE SCALE EIGENVALUE CALCULATIONS Danny C. Sorensen 1 Department of Computational and Applied Mathematics Rice University Houston, TX 77251 sorensen@rice, edu ABSTRACT Eigenvalues and eigenfunctions of linear operators are important to many areas of ap- plied mathematics. The ability to approximate these quantities numerically is becoming increasingly important in a wide variety of applications. This increasing demand has fu- eled interest in the development of new methods and software for the numerical solution of large-scale algebraic eigenvalue problems. In turn, the existence of these new methods and software, along with the dramatically increased computational capabilities now avail- able, has enabled the solution of problems that would not even have been posed five or ten years ago. Until very recently, software for large-scale nonsymmetric problems was virtually non-existent. Fortunately, the situation is improving rapidly. The purpose of this article is to provide an overview of the numerical solution of large- scale algebraic eigenvalue problems. The focus will be on a class of methods called Krylov subspace projection methods. The well-known Lanczos method is the premier member of this class.
    [Show full text]
  • A Geometric Theory for Preconditioned Inverse Iteration. III: a Short and Sharp Convergence Estimate for Generalized Eigenvalue Problems
    A geometric theory for preconditioned inverse iteration. III: A short and sharp convergence estimate for generalized eigenvalue problems. Andrew V. Knyazev Department of Mathematics, University of Colorado at Denver, P.O. Box 173364, Campus Box 170, Denver, CO 80217-3364 1 Klaus Neymeyr Mathematisches Institut der Universit¨atT¨ubingen,Auf der Morgenstelle 10, 72076 T¨ubingen,Germany 2 Abstract In two previous papers by Neymeyr: A geometric theory for preconditioned inverse iteration I: Extrema of the Rayleigh quotient, LAA 322: (1-3), 61-85, 2001, and A geometric theory for preconditioned inverse iteration II: Convergence estimates, LAA 322: (1-3), 87-104, 2001, a sharp, but cumbersome, convergence rate estimate was proved for a simple preconditioned eigensolver, which computes the smallest eigenvalue together with the corresponding eigenvector of a symmetric positive def- inite matrix, using a preconditioned gradient minimization of the Rayleigh quotient. In the present paper, we discover and prove a much shorter and more elegant, but still sharp in decisive quantities, convergence rate estimate of the same method that also holds for a generalized symmetric definite eigenvalue problem. The new estimate is simple enough to stimulate a search for a more straightforward proof technique that could be helpful to investigate such practically important method as the locally optimal block preconditioned conjugate gradient eigensolver. We demon- strate practical effectiveness of the latter for a model problem, where it compares favorably with two
    [Show full text]
  • Accelerated Stochastic Power Iteration
    Accelerated Stochastic Power Iteration CHRISTOPHER DE SAy BRYAN HEy IOANNIS MITLIAGKASy CHRISTOPHER RE´ y PENG XU∗ yDepartment of Computer Science, Stanford University ∗Institute for Computational and Mathematical Engineering, Stanford University cdesa,bryanhe,[email protected], [email protected], [email protected] July 11, 2017 Abstract Principal component analysis (PCA) is one of the most powerful tools in machine learning. The simplest method for PCA, the power iteration, requires O(1=∆) full-data passes to recover the principal component of a matrix withp eigen-gap ∆. Lanczos, a significantly more complex method, achieves an accelerated rate of O(1= ∆) passes. Modern applications, however, motivate methods that only ingest a subset of available data, known as the stochastic setting. In the online stochastic setting, simple 2 2 algorithms like Oja’s iteration achieve the optimal sample complexity O(σ =p∆ ). Unfortunately, they are fully sequential, and also require O(σ2=∆2) iterations, far from the O(1= ∆) rate of Lanczos. We propose a simple variant of the power iteration with an added momentum term, that achieves both the optimal sample and iteration complexity.p In the full-pass setting, standard analysis shows that momentum achieves the accelerated rate, O(1= ∆). We demonstrate empirically that naively applying momentum to a stochastic method, does not result in acceleration. We perform a novel, tight variance analysis that reveals the “breaking-point variance” beyond which this acceleration does not occur. By combining this insight with modern variance reduction techniques, we construct stochastic PCAp algorithms, for the online and offline setting, that achieve an accelerated iteration complexity O(1= ∆).
    [Show full text]
  • Numerical Linear Algebra Program Lecture 4 Basic Methods For
    Program Lecture 4 Numerical Linear Algebra • Basic methods for eigenproblems. Basic iterative methods • Power method • Shift-and-invert Power method • QR algorithm • Basic iterative methods for linear systems • Richardson’s method • Jacobi, Gauss-Seidel and SOR • Iterative refinement Gerard Sleijpen and Martin van Gijzen • Steepest decent and the Minimal residual method October 5, 2016 1 October 5, 2016 2 National Master Course National Master Course Delft University of Technology Basic methods for eigenproblems The Power method The eigenvalue problem The Power method is the classical method to compute in modulus largest eigenvalue and associated eigenvector of a Av = λv matrix. can not be solved in a direct way for problems of order > 4, since Multiplying with a matrix amplifies strongest the eigendirection the eigenvalues are the roots of the characteristic equation corresponding to the in modulus largest eigenvalues. det(A − λI) = 0. Successively multiplying and scaling (to avoid overflow or underflow) yields a vector in which the direction of the largest Today we will discuss two iterative methods for solving the eigenvector becomes more and more dominant. eigenproblem. October 5, 2016 3 October 5, 2016 4 National Master Course National Master Course Algorithm Convergence (1) The Power method for an n × n matrix A. Let the n eigenvalues λi with eigenvectors vi, Avi = λivi, be n ordered such that |λ1| ≥ |λ2|≥ . ≥ |λn|. u0 ∈ C is given • Assume the eigenvectors v ,..., vn form a basis. for k = 1, 2, ... 1 • Assume |λ1| > |λ2|. uk = Auk−1 Each arbitrary starting vector u0 can be written as: uk = uk/kukk2 e(k) ∗ λ = uk−1uk u0 = α1v1 + α2v2 + ..
    [Show full text]
  • The Godunov–Inverse Iteration: a Fast and Accurate Solution to the Symmetric Tridiagonal Eigenvalue Problem
    The Godunov{Inverse Iteration: A Fast and Accurate Solution to the Symmetric Tridiagonal Eigenvalue Problem Anna M. Matsekh a;1 aInstitute of Computational Technologies, Siberian Branch of the Russian Academy of Sciences, Lavrentiev Ave. 6, Novosibirsk 630090, Russia Abstract We present a new hybrid algorithm based on Godunov's method for computing eigenvectors of symmetric tridiagonal matrices and Inverse Iteration, which we call the Godunov{Inverse Iteration. We use eigenvectors computed according to Go- dunov's method as starting vectors in the Inverse Iteration, replacing any non- numeric elements of Godunov's eigenvectors with random uniform numbers. We use the right-hand bounds of the Ritz intervals found by the bisection method as Inverse Iteration shifts, while staying within guaranteed error bounds. In most test cases convergence is reached after only one step of the iteration, producing error estimates that are as good as or superior to those produced by standard Inverse Iteration implementations. Key words: Symmetric eigenvalue problem, tridiagonal matrices, Inverse Iteration 1 Introduction Construction of algorithms that enable to find all eigenvectors of the symmet- ric tridiagonal eigenvalue problem with guaranteed accuracy in O(n2) arith- metic operations has become one of the most pressing problems of the modern numerical algebra. QR method, one of the most accurate methods for solv- ing eigenvalue problems, requires 6n3 arithmetic operations and O(n2) square root operations to compute all eigenvectors of a tridiagonal matrix [Golub and Email address: [email protected] (Anna M. Matsekh ). 1 Present address: Modeling, Algorithms and Informatics Group, Los Alamos Na- tional Laboratory. P.O. Box 1663, MS B256, Los Alamos, NM 87544, USA Preprint submitted to Elsevier Science 31 January 2003 Loan (1996)].
    [Show full text]
  • A Note on Inverse Iteration
    A Note on Inverse Iteration Klaus Neymeyr Universit¨atRostock, Fachbereich Mathematik, Universit¨atsplatz 1, 18051 Rostock, Germany; SUMMARY Inverse iteration, if applied to a symmetric positive definite matrix, is shown to generate a sequence of iterates with monotonously decreasing Rayleigh quotients. We present sharp bounds from above and from below which highlight inverse iteration as a descent scheme for the Rayleigh quotient. Such estimates provide the background for the analysis of the behavior of the Rayleigh quotient in certain approximate variants of inverse iteration. key words: Symmetric eigenvalue problem; Inverse iteration; Rayleigh quotient. 1. Introduction Inverse iteration is a well-known iterative procedure to compute approximations of eigenfunctions and eigenvalues of linear operators. It was introduced by Wielandt in 1944 in a sequence of five papers, see [1], to treat the matrix eigenvalue problem Axi = λixi for a real or complex square matrix A. The scalar λi is the ith eigenvalue and the vector xi denotes a corresponding eigenvector. Given a nonzero starting vector x(0), inverse iteration generates a sequence of iterates x(k) by solving the linear systems (A − σI)x(k+1) = x(k), k = 0, 1, 2,..., (1) where σ denotes an eigenvalue approximation and I is the identity matrix. In practice, the iterates are normalized after each step. If A is a symmetric matrix, then the iterates x(k) converge to an eigenvector associated with an eigenvalue closest to σ if the starting vector x(0) is not perpendicular to that vector. For non-symmetric matrices the issue of starting vectors is discussed in Sec.
    [Show full text]
  • Inverse, Shifted Inverse, and Rayleigh Quotient Iteration As Newton’S Method⇤
    INVERSE, SHIFTED INVERSE, AND RAYLEIGH QUOTIENT ITERATION AS NEWTON’S METHOD⇤ JOHN DENNIS JR.† AND RICHARD TAPIA‡ Abstract. Two-norm normalized inverse, shifted inverse, and Rayleigh quotient iteration are well-known algorithms for approximating an eigenvector of a symmetric matrix. In this work we establish rigorously that each one of these three algorithms can be viewed as a standard form of Newton’s method from the nonlinear programming literature, followed by the normalization. This equivalence adds considerable understanding to the formal structure of inverse, shifted inverse, and Rayleigh quotient iteration and provides an explanation for their good behavior despite the possible need to solve systems with nearly singular coefficient matrices; the algorithms have what can be viewed as removable singularities. A thorough historical development of these eigenvalue algorithms is presented. Utilizing our equivalences we construct traditional Newton’s method theory analysis in order to gain understanding as to why, as normalized Newton’s method, inverse iteration and shifted inverse iteration are only linearly convergent and not quadratically convergent, and why a new linear system need not be solved at each iteration. We also investigate why Rayleigh quotient iteration is cubically convergent and not just quadratically convergent. Key words. inverse power method, inverse iteration, shifted inverse iteration, Rayleigh quotient iteration, Newton’s method AMS subject classifications. 65F15, 49M37, 49M15, 65K05 1. Introduction. The initial
    [Show full text]
  • Arxiv:1105.1185V1 [Math.NA] 5 May 2011 Ento 2.2
    ITERATIVE METHODS FOR COMPUTING EIGENVALUES AND EIGENVECTORS MAYSUM PANJU Abstract. We examine some numerical iterative methods for computing the eigenvalues and eigenvectors of real matrices. The five methods examined here range from the simple power iteration method to the more complicated QR iteration method. The derivations, procedure, and advantages of each method are briefly discussed. 1. Introduction Eigenvalues and eigenvectors play an important part in the applications of linear algebra. The naive method of finding the eigenvalues of a matrix involves finding the roots of the characteristic polynomial of the matrix. In industrial sized matrices, however, this method is not feasible, and the eigenvalues must be obtained by other means. Fortunately, there exist several other techniques for finding eigenvalues and eigenvectors of a matrix, some of which fall under the realm of iterative methods. These methods work by repeatedly refining approximations to the eigenvectors or eigenvalues, and can be terminated whenever the approximations reach a suitable degree of accuracy. Iterative methods form the basis of much of modern day eigenvalue computation. In this paper, we outline five such iterative methods, and summarize their derivations, procedures, and advantages. The methods to be examined are the power iteration method, the shifted inverse iteration method, the Rayleigh quotient method, the simultaneous iteration method, and the QR method. This paper is meant to be a survey over existing algorithms for the eigenvalue computation problem. Section 2 of this paper provides a brief review of some of the linear algebra background required to understand the concepts that are discussed. In section 3, the iterative methods are each presented, in order of complexity, and are studied in brief detail.
    [Show full text]
  • Preconditioned Inverse Iteration and Shift-Invert Arnoldi Method
    Preconditioned inverse iteration and shift-invert Arnoldi method Melina Freitag Department of Mathematical Sciences University of Bath CSC Seminar Max-Planck-Institute for Dynamics of Complex Technical Systems Magdeburg 3rd May 2011 joint work with Alastair Spence (Bath) 1 Introduction 2 Inexact inverse iteration 3 Inexact Shift-invert Arnoldi method 4 Inexact Shift-invert Arnoldi method with implicit restarts 5 Conclusions Outline 1 Introduction 2 Inexact inverse iteration 3 Inexact Shift-invert Arnoldi method 4 Inexact Shift-invert Arnoldi method with implicit restarts 5 Conclusions Problem and iterative methods Find a small number of eigenvalues and eigenvectors of: Ax = λx, λ ∈ C,x ∈ Cn A is large, sparse, nonsymmetric Problem and iterative methods Find a small number of eigenvalues and eigenvectors of: Ax = λx, λ ∈ C,x ∈ Cn A is large, sparse, nonsymmetric Iterative solves Power method Simultaneous iteration Arnoldi method Jacobi-Davidson method repeated application of the matrix A to a vector Generally convergence to largest/outlying eigenvector Problem and iterative methods Find a small number of eigenvalues and eigenvectors of: Ax = λx, λ ∈ C,x ∈ Cn A is large, sparse, nonsymmetric Iterative solves Power method Simultaneous iteration Arnoldi method Jacobi-Davidson method The first three of these involve repeated application of the matrix A to a vector Generally convergence to largest/outlying eigenvector Shift-invert strategy Wish to find a few eigenvalues close to a shift σ σ λλ λ λ λ 3 1 2 4 n Shift-invert strategy Wish to find a few eigenvalues close to a shift σ σ λλ λ λ λ 3 1 2 4 n Problem becomes − 1 (A − σI) 1x = x λ − σ each step of the iterative method involves repeated application of A =(A − σI)−1 to a vector Inner iterative solve: (A − σI)y = x using Krylov method for linear systems.
    [Show full text]
  • Lecture 12 — February 26 and 28 12.1 Introduction
    EE 381V: Large Scale Learning Spring 2013 Lecture 12 | February 26 and 28 Lecturer: Caramanis & Sanghavi Scribe: Karthikeyan Shanmugam and Natalia Arzeno 12.1 Introduction In this lecture, we focus on algorithms that compute the eigenvalues and eigenvectors of a real symmetric matrix. Particularly, we are interested in finding the largest and smallest eigenvalues and the corresponding eigenvectors. We study two methods: Power method and the Lanczos iteration. The first involves multiplying the symmetric matrix by a randomly chosen vector, and iteratively normalizing and multiplying the matrix by the normalized vector from the previous step. The convergence is geometric, i.e. the `1 distance between the true and the computed largest eigenvalue at the end of every step falls geometrically in the number of iterations and the rate depends on the ratio between the second largest and the largest eigenvalue. Some generalizations of the power method to compute the largest k eigenvalues and the eigenvectors will be discussed. The second method (Lanczos iteration) terminates in n iterations where each iteration in- volves estimating the largest (smallest) eigenvalue by maximizing (minimizing) the Rayleigh coefficient over vectors drawn from a suitable subspace. At each iteration, the dimension of the subspace involved in the optimization increases by 1. The sequence of subspaces used are Krylov subspaces associated with a random initial vector. We study the relation between Krylov subspaces and tri-diagonalization of a real symmetric matrix. Using this connection, we show that an estimate of the extreme eigenvalues can be computed at each iteration which involves eigen-decomposition of a tri-diagonal matrix.
    [Show full text]
  • Randomized Numerical Linear Algebra: Foundations & Algorithms
    Randomized Numerical Linear Algebra: Foundations & Algorithms Per-Gunnar Martinsson, University of Texas at Austin Joel A. Tropp, California Institute of Technology Abstract: This survey describes probabilistic algorithms for linear algebra computations, such as factorizing matrices and solving linear systems. It focuses on techniques that have a proven track record for real-world problems. The paper treats both the theoretical foundations of the subject and practical computational issues. Topics include norm estimation; matrix approximation by sampling; structured and un- structured random embeddings; linear regression problems; low-rank approximation; sub- space iteration and Krylov methods; error estimation and adaptivity; interpolatory and CUR factorizations; Nystrom¨ approximation of positive semidefinite matrices; single- view (“streaming”) algorithms; full rank-revealing factorizations; solvers for linear sys- tems; and approximation of kernel matrices that arise in machine learning and in scientific computing. CONTENTS 1. Introduction 2 2. Linear algebra preliminaries 6 3. Probability preliminaries 9 4. Trace estimation by sampling 10 5. Schatten p-norm estimation by sampling 16 6. Maximum eigenvalues and trace functions 18 7. Matrix approximation by sampling 24 8. Randomized embeddings 30 9. Structured random embeddings 35 10. How to use random embeddings 40 arXiv:2002.01387v2 [math.NA] 15 Aug 2020 11. The randomized rangefinder 43 12. Error estimation and adaptivity 52 13. Finding natural bases: QR, ID, and CUR 57 14. Nystrom¨ approximation 62 15. Single-view algorithms 64 16. Factoring matrices of full or nearly full rank 68 17. General linear solvers 76 18. Linear solvers for graph Laplacians 79 19. Kernel matrices in machine learning 85 20. High-accuracy approximation of kernel matrices 94 References 106 Randomized Numerical Linear Algebra P.G.
    [Show full text]