DOCSLIB.ORG

Dual Principal Component Pursuit and Filtrated Algebraic Subspace

Total Page：16

File Type：pdf, Size：1020Kb

Download full-text PDF Read full-text
Dual Principal Component Pursuit and Filtrated Algebraic Subspace Dual Principal Component Pursuit and Filtrated Algebraic Subspace Clustering by Manolis C. Tsakiris A dissertation submitted to The Johns Hopkins University in conformity with the requirements for the degree of Doctor of Philosophy. Baltimore, Maryland March, 2017 c Manolis C. Tsakiris 2017 All rights reserved Abstract Recent years have witnessed an explosion of data across scientific fields enabled by ad- vances in sensor technology and distributed hardware systems. This has given rise to the challenge of efficiently processing such data for performing various tasks, such as face recognition in images and videos. Towards that end, two operations on the data are of fun- damental importance, i.e., dimensionality reduction and clustering, and the idea of learning one or more linear subspaces from the data has proved a fruitful notion in that context. Nev- ertheless, state-of-the-art methods, such as Robust Principal Component Analysis (RPCA) or Sparse Subspace Clustering (SSC), operate under the hypothesis that the subspaces have dimensions that are small relative to the ambient dimension, and fail otherwise. This thesis attempts to advance the state-of-the-art of subspace learning methods in the regime where the subspaces have high relative dimensions. The first major contribution of this thesis is a single subspace learning method called Dual Principal Component Pursuit (DPCP), which solves the robust PCA problem in the presence of outliers. Contrary to sparse and low-rank state-of-the-art methods, the theoretical guarantees of DPCP do not place any constraints on the dimension of the subspace. In particular, DPCP computes the ii ABSTRACT orthogonal complement of the subspace, thus it is particularly suited for subspaces of low codimension. This is done by solving a non-convex cosparse problem on the sphere, whose global minimizers are shown to be vectors normal to the subspace. Algorithms for solving the non-convex problem are developed and tested on synthetic and real data, showing that DPCP is able to handle higher subspace dimensions and larger amounts of outliers than existing methods. Finally, DPCP is extended theoretically and algorithmically to the case of multiple subspaces. The second major contribution of this thesis is a subspace clustering method called Filtrated Algebraic Subspace Clustering (FASC), which builds upon algebraic geometric ideas of an older method, known as Generalized Principal Component Analysis (GPCA). GPCA is naturally suited for subspaces of large dimension, but it suffers from two weak- nesses: sensitivity to noise and large computational complexity. This thesis demonstrates that FASC addresses successfully the robustness to noise. This is achieved through an equivalent formulation of GPCA, which uses the idea of filtrations of unions of subspaces. An algebraic geometric analysis establishes the theoretical equivalence of the two meth- ods, while experiments on synthetic and real data reveal that FASC not only dramatically improves upon GPCA, but also upon existing methods on several occasions. Primary Reader: Rene´ Vidal Secondary Reader: Daniel P. Robinson iii Dedication This thesis is dedicated to my advisor, Rene´ Vidal, for marveling me with his genius, rigor, endurance and enthusiasm. iv Acknowledgments First of all, I am thankful to my advisor, Prof. Rene´ Vidal, whose influence has been enor- mous: working with him has been the best academic experience in my life. Secondly, I am thankful to Prof. Daniel P. Robinson, for many useful conversations, for always be- ing supportive and friendly, and, quite importantly, for not being easily convinced: this led to the identification of a few inaccuracies and the improvement of several arguments in this thesis. I also thank Prof. Aldo Conca of the mathematics department of the Uni- versity of Genova for enthusiastically answering many questions that i had regarding the Castelnuovo-Mumford regularity of subspace arrangements, as well as Prof. Glyn Harman for pointing out a Koksma-Hlawka inequality for integration on the unit sphere. Then I thank Miss Debbie Race of the ECE department for being extremely helpful and patient with several administrative issues. I thank the Center for Imaging Science (CIS) for being a cozy home for the last 4 years and the ECE department for always being supportive and admitting me as a PhD student to begin with. I thank all people who were nice to me during my stay in Baltimore. Special thanks go to sensei Ebon Phoenix for passionately teaching me martial arts, and to Tony Hatzigeor- v ACKNOWLEDGMENTS galis and Jose Torres for being like brothers to me. I also thank the JHU newbie, Christos Sapsanis, from whom, remarkably, the JHU graduate community has a lot to benefit. And Guilherme Franca and Tao Xiong for being great friends. Finally, i thank my best friend Dimitris Lountzis for being who he is, and most impor- tantly i thank my family, Chris, Evi and Michalis, for their infinite love. vi Contents Abstract ii Acknowledgments v List of Tables xiii List of Figures xv 1 Introduction 1 1.1 Modeling data with linear subspaces . 1 1.1.1 Modeling data with a single subspace . 1 1.1.2 Modeling data with multiple subspaces . 3 1.2 Challenges and the role of dimension . 4 1.3 Contributions of this thesis . 5 1.3.1 Dual principal component pursuit (DPCP) . 6 1.3.2 Filtrated algebraic subspace clustering . 8 1.4 Notation . 10 vii CONTENTS 2 Prior Art 13 2.1 Learning a single subspace in the presence of outliers . 15 2.2 Learning multiple subspaces . 20 2.3 Challenges in high relative dimensions . 25 3 Dual Principal Component Pursuit 29 3.1 Introduction . 29 3.2 Single subspace learning with outliers via DPCP . 32 3.2.1 Problem formulation . 32 3.2.1.1 Data model . 32 3.2.1.2 Conceptual formulation . 33 3.2.1.3 Hyperplane pursuit by `1 minimization . 34 3.2.2 Theoretical analysis of the continuous problem . 37 3.2.2.1 The underlying continuous problem . 38 3.2.2.2 Conditions for global optimality and convergence . 42 3.2.3 Theoretical analysis of the discrete problem . 48 3.2.3.1 Discrepancy bounds between continuous and discrete prob- lems . 49 3.2.3.2 Conditions for global optimality of the discrete problem 56 3.2.3.3 Conditions for convergence of the discrete recursive al- gorithm . 66 3.2.4 Algorithmic contributions . 71 viii CONTENTS 3.2.4.1 Relaxed DPCP and DPCA algorithms . 71 3.2.4.2 Relaxed and denoised DPCP . 74 3.2.4.3 Denoised DPCP . 76 3.2.4.4 DPCP via iteratively reweighted least-squares . 79 3.2.5 Experimental evaluation . 80 3.2.5.1 Computing a single dual principal component . 81 3.2.5.2 Outlier detection using synthetic data . 84 3.2.5.3 Outlier detection using real face and object images . 90 3.3 Learning a hyperplane arrangement via DPCP . 96 3.3.1 Problem overview . 96 3.3.2 Data model . 97 3.3.3 Theoretical analysis of the continuous problem . 98 3.3.3.1 Derivation, interpretation and basic properties of the con- tinuous problem . 99 3.3.3.2 The cases of i) two hyperplanes and ii) orthogonal hy- perplanes . 102 3.3.3.3 The case of three equiangular hyperplanes . 105 3.3.3.4 Conditions of global optimality for an arbitrary hyper- plane arrangement . 114 3.3.4 Theoretical analysis of the discrete problem . 120 3.3.5 Algorithms . 131 ix CONTENTS 3.3.5.1 Learning a hyperplane arrangement sequentially . 132 3.3.5.2 K-Hyperplanes via DPCP . 132 3.3.6 Experimental evaluation . 134 3.3.6.1 Synthetic data . 134 3.3.6.2 3D plane clustering of real Kinect data . 142 3.4 Conclusions . 148 4 Advances in Algebraic Subspace Clustering 155 4.1 Review of algebraic subspace clustering . 156 4.1.1 Subspaces of codimension 1 . 157 4.1.2 Subspaces of equal dimension . 160 4.1.3 Known number of subspaces of arbitrary dimensions . 161 4.1.4 Unknown number of subspaces of arbitrary dimensions . 165 4.1.5 Computational complexity and recursive ASC . 167 4.1.6 Instability in the presence of noise and spectral ASC . 168 4.1.7 The challenge . 170 4.2 Filtrated algebraic subspace clustering (FASC) . 171 4.2.1 Filtrations of subspace arrangements: geometric overview . 171 4.2.2 Filtrations of subspace arrangements: theory . 177 4.2.2.1 Data in general position in a subspace arrangement . 177 4.2.2.2 Constructing the first step of a filtration . 182 4.2.2.3 Deciding whether to take a second step in a filtration . 185 x CONTENTS 4.2.2.4 Taking multiple steps in a filtration and terminating . 188 4.2.2.5 The FASC algorithm . 193 4.3 Filtrated spectral algebraic subspace clustering . 195 4.3.1 Implementing robust filtrations . 195 4.3.2 Combining multiple filtrations . 198 4.3.3 The FSASC algorithm . 199 4.3.4 A distance-based affinity . 200 4.3.5 Discussion on the computational complexity . 203 4.4 Experiments . 205 4.4.1 Experiments on synthetic data . 206 4.4.2 Experiments on real motion sequences . 217 4.5 Algebraic clustering of affine subspaces . 219 4.5.1 Motivation . 219 4.5.2 Problem statement and traditional approach . 221 4.5.3 Algebraic geometry of unions of affine subspaces . 224 4.5.3.1 Affine subspaces as affine varieties . 224 4.5.3.2 The projective closure of affine subspaces . 227 4.5.4 Correctness theorems for the homogenization trick . 232 4.6 Conclusions . 238 4.7 Appendix . 239 4.7.1 Notions from commutative algebra . 239 xi CONTENTS 4.7.2 Notions from algebraic geometry . 241 4.7.3 Subspace arrangements and their vanishing ideals .
Load more

Recommended publications
  • Orthogonal Complements (Revised Version)
    Orthogonal Complements (Revised Version) Math 108A: May 19, 2010 John Douglas Moore 1 The dot product You will recall that the dot product was discussed in earlier calculus courses. If n x = (x1: : : : ; xn) and y = (y1: : : : ; yn) are elements of R , we define their dot product by x · y = x1y1 + ··· + xnyn: The dot product satisfies several key axioms: 1. it is symmetric: x · y = y · x; 2. it is bilinear: (ax + x0) · y = a(x · y) + x0 · y; 3. and it is positive-definite: x · x ≥ 0 and x · x = 0 if and only if x = 0. The dot product is an example of an inner product on the vector space V = Rn over R; inner products will be treated thoroughly in Chapter 6 of [1]. Recall that the length of an element x 2 Rn is defined by p jxj = x · x: Note that the length of an element x 2 Rn is always nonnegative. Cauchy-Schwarz Theorem. If x 6= 0 and y 6= 0, then x · y −1 ≤ ≤ 1: (1) jxjjyj Sketch of proof: If v is any element of Rn, then v · v ≥ 0. Hence (x(y · y) − y(x · y)) · (x(y · y) − y(x · y)) ≥ 0: Expanding using the axioms for dot product yields (x · x)(y · y)2 − 2(x · y)2(y · y) + (x · y)2(y · y) ≥ 0 or (x · x)(y · y)2 ≥ (x · y)2(y · y): 1 Dividing by y · y, we obtain (x · y)2 jxj2jyj2 ≥ (x · y)2 or ≤ 1; jxj2jyj2 and (1) follows by taking the square root.
    [Show full text]
  • Does Geometric Algebra Provide a Loophole to Bell's Theorem?
    Discussion Does Geometric Algebra provide a loophole to Bell’s Theorem? Richard David Gill 1 1 Leiden University, Faculty of Science, Mathematical Institute; [email protected] Version October 30, 2019 submitted to Entropy Abstract: Geometric Algebra, championed by David Hestenes as a universal language for physics, was used as a framework for the quantum mechanics of interacting qubits by Chris Doran, Anthony Lasenby and others. Independently of this, Joy Christian in 2007 claimed to have refuted Bell’s theorem with a local realistic model of the singlet correlations by taking account of the geometry of space as expressed through Geometric Algebra. A series of papers culminated in a book Christian (2014). The present paper first explores Geometric Algebra as a tool for quantum information and explains why it did not live up to its early promise. In summary, whereas the mapping between 3D geometry and the mathematics of one qubit is already familiar, Doran and Lasenby’s ingenious extension to a system of entangled qubits does not yield new insight but just reproduces standard QI computations in a clumsy way. The tensor product of two Clifford algebras is not a Clifford algebra. The dimension is too large, an ad hoc fix is needed, several are possible. I further analyse two of Christian’s earliest, shortest, least technical, and most accessible works (Christian 2007, 2011), exposing conceptual and algebraic errors. Since 2015, when the first version of this paper was posted to arXiv, Christian has published ambitious extensions of his theory in RSOS (Royal Society - Open Source), arXiv:1806.02392, and in IEEE Access, arXiv:1405.2355.
    [Show full text]
  • Signing a Linear Subspace: Signature Schemes for Network Coding
    Signing a Linear Subspace: Signature Schemes for Network Coding Dan Boneh1?, David Freeman1?? Jonathan Katz2???, and Brent Waters3† 1 Stanford University, {dabo,dfreeman}@cs.stanford.edu 2 University of Maryland, [email protected] 3 University of Texas at Austin, [email protected]. Abstract. Network coding offers increased throughput and improved robustness to random faults in completely decentralized networks. In contrast to traditional routing schemes, however, network coding requires intermediate nodes to modify data packets en route; for this reason, standard signature schemes are inapplicable and it is a challenge to provide resilience to tampering by malicious nodes. Here, we propose two signature schemes that can be used in conjunction with network coding to prevent malicious modification of data. In particular, our schemes can be viewed as signing linear subspaces in the sense that a signature σ on V authenticates exactly those vectors in V . Our first scheme is homomorphic and has better performance, with both public key size and per-packet overhead being constant. Our second scheme does not rely on random oracles and uses weaker assumptions. We also prove a lower bound on the length of signatures for linear subspaces showing that both of our schemes are essentially optimal in this regard. 1 Introduction Network coding [1, 25] refers to a general class of routing mechanisms where, in contrast to tra- ditional “store-and-forward” routing, intermediate nodes modify data packets in transit. Network coding has been shown to offer a number of advantages with respect to traditional routing, the most well-known of which is the possibility of increased throughput in certain network topologies (see, e.g., [21] for measurements of the improvement network coding gives even for unicast traffic).
    [Show full text]
  • Quantification of Stability in Systems of Nonlinear Ordinary Differential Equations Jason Terry
    University of New Mexico UNM Digital Repository Mathematics & Statistics ETDs Electronic Theses and Dissertations 2-14-2014 Quantification of Stability in Systems of Nonlinear Ordinary Differential Equations Jason Terry Follow this and additional works at: https://digitalrepository.unm.edu/math_etds Recommended Citation Terry, Jason. "Quantification of Stability in Systems of Nonlinear Ordinary Differential Equations." (2014). https://digitalrepository.unm.edu/math_etds/48 This Dissertation is brought to you for free and open access by the Electronic Theses and Dissertations at UNM Digital Repository. It has been accepted for inclusion in Mathematics & Statistics ETDs by an authorized administrator of UNM Digital Repository. For more information, please contact [email protected]. Candidate Department This dissertation is approved, and it is acceptable in quality and form for publication: Approved by the Dissertation Committee: , Chairperson Quantification of Stability in Systems of Nonlinear Ordinary Differential Equations by Jason Terry B.A., Mathematics, California State University Fresno, 2003 B.S., Computer Science, California State University Fresno, 2003 M.A., Interdiscipline Studies, California State University Fresno, 2005 M.S., Applied Mathematics, University of New Mexico, 2009 DISSERTATION Submitted in Partial Fulfillment of the Requirements for the Degree of Doctor of Philosophy Mathematics The University of New Mexico Albuquerque, New Mexico December, 2013 c 2013, Jason Terry iii Dedication To my mom. iv Acknowledgments I would like
    [Show full text]
  • Lecture 6: Linear Codes 1 Vector Spaces 2 Linear Subspaces 3
    Error Correcting Codes: Combinatorics, Algorithms and Applications (Fall 2007) Lecture 6: Linear Codes January 26, 2009 Lecturer: Atri Rudra Scribe: Steve Uurtamo 1 Vector Spaces A vector space V over a field F is an abelian group under “+” such that for every α ∈ F and every v ∈ V there is an element αv ∈ V , and such that: i) α(v1 + v2) = αv1 + αv2, for α ∈ F, v1, v2 ∈ V. ii) (α + β)v = αv + βv, for α, β ∈ F, v ∈ V. iii) α(βv) = (αβ)v for α, β ∈ F, v ∈ V. iv) 1v = v for all v ∈ V, where 1 is the unit element of F. We can think of the field F as being a set of “scalars” and the set V as a set of “vectors”. If the field F is a finite field, and our alphabet Σ has the same number of elements as F , we can associate strings from Σn with vectors in F n in the obvious way, and we can think of codes C as being subsets of F n. 2 Linear Subspaces Assume that we’re dealing with a vector space of dimension n, over a finite field with q elements. n We’ll denote this as: Fq . Linear subspaces of a vector space are simply subsets of the vector space that are closed under vector addition and scalar multiplication: n n In particular, S ⊆ Fq is a linear subspace of Fq if: i) For all v1, v2 ∈ S, v1 + v2 ∈ S. ii) For all α ∈ Fq, v ∈ S, αv ∈ S. Note that the vector space itself is a linear subspace, and that the zero vector is always an element of every linear subspace.
    [Show full text]
  • Worksheet 1, for the MATLAB Course by Hans G. Feichtinger, Edinburgh, Jan
    Worksheet 1, for the MATLAB course by Hans G. Feichtinger, Edinburgh, Jan. 9th Start MATLAB via: Start > Programs > School applications > Sci.+Eng. > Eng.+Electronics > MATLAB > R2007 (preferably). Generally speaking I suggest that you start your session by opening up a diary with the command: diary mydiary1.m and concluding the session with the command diary off. If you want to save your workspace you my want to call save today1.m in order to save all the current variable (names + values). Moreover, using the HELP command from MATLAB you can get help on more or less every MATLAB command. Try simply help inv or help plot. 1. Define an arbitrary (random) 3 × 3 matrix A, and check whether it is invertible. Calculate the inverse matrix. Then define an arbitrary right hand side vector b and determine the (unique) solution to the equation A ∗ x = b. Find in two different ways the solution to this problem, by either using the inverse matrix, or alternatively by applying Gauss elimination (i.e. the RREF command) to the extended system matrix [A; b]. In addition look at the output of the command rref([A,eye(3)]). What does it tell you? 2. Produce an \arbitrary" 7 × 7 matrix of rank 5. There are at east two simple ways to do this. Either by factorization, i.e. by obtaining it as a product of some 7 × 5 matrix with another random 5 × 7 matrix, or by purposely making two of the rows or columns linear dependent from the remaining ones. Maybe the second method is interesting (if you have time also try the first one): 3.
    [Show full text]
  • 3. Hilbert Spaces
    3. Hilbert spaces In this section we examine a special type of Banach spaces. We start with some algebraic preliminaries. Definition. Let K be either R or C, and let Let X and Y be vector spaces over K. A map φ : X × Y → K is said to be K-sesquilinear, if • for every x ∈ X, then map φx : Y 3 y 7−→ φ(x, y) ∈ K is linear; • for every y ∈ Y, then map φy : X 3 y 7−→ φ(x, y) ∈ K is conjugate linear, i.e. the map X 3 x 7−→ φy(x) ∈ K is linear. In the case K = R the above properties are equivalent to the fact that φ is bilinear. Remark 3.1. Let X be a vector space over C, and let φ : X × X → C be a C-sesquilinear map. Then φ is completely determined by the map Qφ : X 3 x 7−→ φ(x, x) ∈ C. This can be see by computing, for k ∈ {0, 1, 2, 3} the quantity k k k k k k k Qφ(x + i y) = φ(x + i y, x + i y) = φ(x, x) + φ(x, i y) + φ(i y, x) + φ(i y, i y) = = φ(x, x) + ikφ(x, y) + i−kφ(y, x) + φ(y, y), which then gives 3 1 X (1) φ(x, y) = i−kQ (x + iky), ∀ x, y ∈ X. 4 φ k=0 The map Qφ : X → C is called the quadratic form determined by φ. The identity (1) is referred to as the Polarization Identity.
    [Show full text]
  • 10. Vector Subspaces the Solution to a Homogeneous Equation a X = 0
    10. Vector subspaces The solution to a homogeneous equation A~x = ~0 in R3 is one of • The origin. • A line through the origin. • A plane through the origin. • The whole of R3. These are all examples of linear subspaces. Definition 10.1. Let H be a subset of Rn. H is called a linear subspace if (1) ~0 2 H. (2) H is closed under addition: If ~u and ~v 2 H then ~u + ~v 2 H. (3) H is closed under scalar multiplication: If ~u and λ is a scalar then λ~u 2 H. Geometrically H is closed under scalar multiplication if and only if H is a union of lines through the origin. H is then closed under addition if and only if it contains every plane containing ever pair of lines. Example 10.2. Let H = f~0g. Then H is a linear subspace. Indeed, ~0 2 H. ~0 + ~0 = ~0 2 H. Similarly λ~0 = ~0. Example 10.3. Let H = Rn. Then H is a linear subspace. Indeed, ~0 2 H. H is obviously closed under addition and scalar multiplication. Now consider lines in R3. Example 10.4. Let H be the x-axis. Then H is a linear subspace. Indeed, ~0 2 H. If ~u and ~v belong to H then ~u and ~v are multiples of (1; 0; 0) and the sum is a multiple of (1; 0; 0). Similarly if λ is a scalar then λ~u is a multiple of (1; 0; 0). Example 10.5. Let H be a line in R3 through the origin.
    [Show full text]
  • Linear Algebra Review Vectors
    Linear Algebra Review By Tim K. Marks UCSD Borrows heavily from: Jana Kosecka [email protected] http://cs.gmu.edu/~kosecka/cs682.html Virginia de Sa Cogsci 108F Linear Algebra review UCSD Vectors The length of x, a.k.a. the norm or 2-norm of x, is 2 2 2 x = x1 + x2 +L+ xn e.g., x = 32 + 22 + 52 = 38 ! ! 1 Good Review Materials http://www.imageprocessingbook.com/DIP2E/dip2e_downloads/review_material_downloads.htm (Gonzales & Woods review materials) Chapt. 1: Linear Algebra Review Chapt. 2: Probability, Random Variables, Random Vectors Online vector addition demo: http://www.pa.uky.edu/~phy211/VecArith/index.html 2 Vector Addition u+v v u Vector Subtraction u-v u v 3 Example (on board) Inner product (dot product) of two vectors # 6 & "4 % % ( $ ' a = % 2 ( b = $1 ' $%" 3'( #$5 &' a" b = aT b ! ! #4 & % ( = [6 2 "3] %1 ( ! $%5 '( = 6" 4 + 2"1+ (#3)" 5 =11 ! ! ! 4 Inner (dot) Product The inner product is a SCALAR. v α u 5 Transpose of a Matrix Transpose: Examples: If , we say A is symmetric. Example of symmetric matrix 6 7 8 9 Matrix Product Product: A and B must have compatible dimensions In Matlab: >> A*B Examples: Matrix Multiplication is not commutative: 10 Matrix Sum Sum: A and B must have the same dimensions Example: Determinant of a Matrix Determinant: A must be square Example: 11 Determinant in Matlab Inverse of a Matrix If A is a square matrix, the inverse of A, called A-1, satisfies AA-1 = I and A-1A = I, Where I, the identity matrix, is a diagonal matrix with all 1’s on the diagonal.
    [Show full text]
  • Linear Subspaces
    LINEAR SUBSPACES 1. Subspaces of Rn We wish to generalize the notion of lines and planes. To that end, say a subset W ⊂ Rn is a (linear) subspace if it has the following three properties: (1) (Non-empty): ~0 2 W ; (2) (Closed under addition): ~v1;~v2 2 W ) ~v1 + ~v2 2 W ; (3) (Closed under scaling): ~v 2 W and k 2 R ) k~v 2 W . n o EXAMPLE: ~0 and Rn are subspaces of Rn for any n ≥ 1. 1 EXAMPLE: Line t : t 2 given by x = x is a subspace. 1 R 1 2 x 1 1 NON-EXAMPLE: W = : x ≥ 0; y ≥ 0 . As 2 W; but − 62 W: y 1 1 EXAMPLE: If T : Rm ! Rn, then ker(T ) is a subspace and Rm and Im (T ) is a subspace of Rn. For instance, (1) T (~0) = ~0 ) ~0 2 ker(T ); (2) ~v1;~v2 2 ker(T ) ) T (~v1 + ~v2) = T (~v1) + T (~v2) = ~0 ) ~v1 + ~v2 2 ker(T ); (3) ~v 2 ker(T ); k 2 R ! T (k~v) = kT (~v) = k~0 = ~0 ) k~v 2 ker(T ): 2. Span of vectors n Given a set of vectors ~v1; : : : ;~vm 2 R we define the span of these vectors to be W = span(~v1; : : : ;~vk) = fc1~v1 + ::: + ck~vk : c1; : : : ; cm 2 Rg In other words, ~w 2 W means ~w is a linear combination of the ~vi. If A = ~v1 j · · · j ~vm is the n × m matrix with columns ~vi, then span(~v1; : : : ;~vk) = Im (A) n and so span(~v1; : : : ;~vk) is a subspace of R .
    [Show full text]
  • Clifford Algebra and the Projective Model of Homogeneous Metric Spaces
    Clifford algebra and the projective model of homogeneous metric spaces: Foundations Andrey Sokolov July 12, 2013 Contents 1 Introduction 3 2 2-dimensional geometry 5 2.1 Projective foundations . .5 2.1.1 Projective duality . .5 2.1.2 Top-down view of geometry . .7 2.1.3 Points at infinity . .8 2.1.4 Extending lines with points at infinity . 10 2.1.5 The line at infinity . 10 2.1.6 Orientation of finite points (sheaves) . 12 2.1.7 Orientation of points at infinity (stacks) and lines . 13 2.1.8 Embedding . 14 2.2 Grassmann algebra . 16 2.3 Visualising bivectors and trivectors . 18 2.4 The metric and Clifford algebra . 20 2.5 Orthogonality in R3∗ with elliptic metric . 22 arXiv:1307.2917v1 [math.MG] 8 Jul 2013 2.6 Duality transformation on multivectors and the join . 24 2.7 Generalising duality transformation to R(n+1)∗ ...................... 26 2.8 Representing orientation and weight with blades . 28 2.9 Euclidean plane E2 ..................................... 32 2.10 Linear functions . 40 3 3-dimensional geometry 46 3.1 Points and planes . 46 3.2 Lines . 47 1 3.3 Points, lines, and the plane at infinity . 49 3.4 Embedding and Grassmann algebra . 50 3.5 Orientation . 53 3.6 The metric and Clifford algebra . 57 3.7 Euclidean space E3 ..................................... 59 4 4-dimensional geometry 72 4.1 Introduction . 72 4.2 Embedding and Grassmann algebra . 73 4.3 The metric and Clifford algebra . 80 4.4 Euclidean hyperspace E4 .................................. 81 5 1-dimensional geometry 85 5.1 Euclidean line E1 .....................................
    [Show full text]
  • Lecture 13 – Subspace Methods for Eigenvalue Problem
    Select a b Cn, b = 0 ∈ 6 %% Initiation: v = b/ b 2, w = Av k k V = [v], W = [w], H = [v∗w] repeat %% Extraction: 1) Compute pre Ritz pairs (ϑ,y): Hy = ϑy 2) Selected an appropriate pre Ritz pair, say, (ϑ,y). 3) u = Vy, r = Wy ϑu − %% Stopping criterion: 4) if r 2 < tol, x = u, λ = ϑ, break, end k k %% Expansion: 5) Select an expansion vector t Cn, t = 0. ∈ 6 6) [v,~h]= Orth(V, t) 7) V [V, v], H [H; v∗W] ← ← 8) w = Av, W [W, w], H = [H, V∗w] ← end repeat Algorithm 13.1. Subspace method for computing one eigenpair (λ, x) with residual accuracy tol for a general square matrix A. March 26, 2018 Lecture 13 – Subspace Methods for Eigenvalue Problem In this lecture, A is an n n matrix. × A pseudo code for any subspace method for eigenvalue problems basically is as in Alg. 13.1. A (hopefully well-conditioned) basis V for a search space = span(V) is expanded in each step (as in the substeps 5 – 8 of Alg. 13.1): an expansionV vector t is selected (in substep 5) and is orthogonalised against V (in 6) before being included in the basis V. An appropriate approximate eigenpair is extracted from the search subspace in each step (the substeps 1 – 3): H is the projection of A onto the search subspace . An appropriate solution (ϑ, y) of the projected problem Hy = ϑy (in 1) is lifted to an approximateV solution in n-space (in substep 3).
    [Show full text]