Numerical Methods and Software for General and Structured Eigenvalue Problems

Numerical Methods and Software for General and Structured Eigenvalue Problems

\main" i i 2004/5/6 page i i i Numerical Methods and Software for General and Structured Eigenvalue Problems vorgelegt von Dipl.-Math. Daniel Kre¼ner von der FakultÄat II - Mathematik und Naturwissenschaften der Technischen UniversitÄat Berlin zur Erlangung des akademischen Grades Doktor der Naturwissenschaften - Dr. rer. nat. - genehmigte Dissertation Promotionsausschuss: Vorsitzender: Prof. Dr. GunÄ ter M. Ziegler Berichter: Prof. Dr. Volker Mehrmann Berichter: Prof. Dr. Ralph Byers Berichter: Prof. Dr. Bo KºagstrÄom Tag der wissenschaftlichen Aussprache: 14.05.2004 Berlin 2004 D 83 i i i i \main" i i 2004/5/6 page ii i i i i i i \main" i i 2004/5/6 page iii i i Zusammenfassung Die Dissertationsschrift beschÄaftigt sich mit numerischen Verfahren zur LÄosung allgemeiner und strukturierter Eigenwertprobleme. Weiterhin werden Software- Implementierungen der behandelten numerischen Verfahren vorgestellt. Die Arbeit enthÄalt zahlreiche BeitrÄage zu verschiedenen Aspekten des QR-Algorithmus, des QZ-Algorithmus, des periodischen QR-Algorithmus, verschiedener strukturerhal- tender Algorithmen furÄ (schief-)Hamiltonische Eigenwertprobleme und Arnoldi-Äahn- lichen Verfahren. Abstract In this thesis, we have investigated numerical methods for the solution of general and structured eigenvalue problems. Moreover, we have presented software imple- menting these methods. Contributions have been made to various aspects of the QR algorithm, the QZ algorithm, the periodic QR algorithm, structure-preserving methods for (skew-)Hamiltonian matrices, and the Krylov-Schur algorithm. i i i i \main" i i 2004/5/6 page iv i i i i i i \main" i i 2004/5/6 page v i i Contents Preface ix 1 The QR Algorithm 1 1 The Standard Eigenvalue Problem . 2 2 Perturbation Analysis . 4 2.1 Spectral Projectors and Separation . 4 2.2 Eigenvalues and Eigenvectors . 6 Eigenvalues. Eigenvectors. 2.3 Eigenvalue Clusters and Invariant Subspaces . 10 On the computation of sep. 2.4 Global Perturbation Bounds . 14 3 The Basic QR Algorithm . 17 3.1 Local Convergence . 17 Stationary shifts. Instationary shifts. 3.2 Hessenberg Form . 22 Reduction to Hessenberg form. 3.3 Implicit Shifted QR Iteration . 25 3.4 Deation . 27 3.5 The Overall Algorithm . 28 3.6 Failure of Global Converge . 30 4 Balancing . 32 4.1 Isolating Eigenvalues . 32 4.2 Scaling . 33 4.3 Merits of Balancing . 35 5 Block Algorithms . 36 5.1 Compact WY Representation . 36 5.2 Block Hessenberg Reduction . 37 5.3 Multishifts and Bulge Pairs . 39 5.4 Connection to Pole Assignment . 41 5.5 Tightly Coupled Tiny Bulges . 43 Introducing a chain of bulges. Chasing a chain of bulges. Getting rid o® a chain of bulges. Numerical results. 6 Advanced Deation Techniques . 48 6.1 Aggressive Early Deation . 48 Numerical results. 6.2 Aggressive Early Deation at Top Left Corner . 50 7 Computation of Invariant Subspaces . 52 7.1 Swapping Two Diagonal Blocks . 53 v i i i i \main" i i 2004/5/6 page vi i i vi Contents 7.2 Reordering . 54 7.3 Block Algorithm . 55 Numerical Results. 8 Case Study: Solution of an Optimal Control Problem . 57 2 The QZ Algorithm 61 1 The Generalized Eigenvalue Problem . 62 2 Perturbation Analysis . 63 2.1 Spectral Projectors and Dif . 64 2.2 Local Perturbation Bounds . 65 On the computation of dif. 2.3 Global Perturbation Bounds . 69 3 The Basic QZ Algorithm . 69 3.1 Hessenberg-Triangular Form . 70 3.2 Implicit Shifted QZ Iteration . 72 3.3 On the Use of Householder Matrices . 75 Early attempts. Opposite Householder matrices. The role of RQ decompositions. 3.4 Deation . 79 Deation of in¯nite eigenvalues. 3.5 The Overall Algorithm . 81 4 Balancing . 83 4.1 Isolating Eigenvalues . 84 4.2 Scaling . 84 5 Block Algorithms . 86 5.1 Reduction to Hessenberg-Triangular Form . 86 Stage 1. Stage 2. 5.2 Multishifts and Bulge Pairs . 92 5.3 Deation of In¯nite Eigenvalues Revisited . 93 5.4 Tightly Coupled Tiny Bulge Pairs . 94 Introducing a chain of bulge pairs. Chasing a chain of bulge pairs. Getting rid o® a chain of bulge pairs. Numerical results. 6 Aggressive Early Deation . 97 Numerical results. 3 The Periodic QR Algorithm 101 1 The Periodic Eigenvalue Problem . 102 2 Perturbation Theory . 104 2.1 Eigenvalues . 106 2.2 Invariant Subspaces . 108 On the computation of cA( ). 3 Derivation of the Periodic QR AlgorithmX . 110 3.1 The Perfect Shu²e . 110 The perfect shu²e version of the periodic Schur form. 3.2 Reduction to Hessenberg Form . 111 3.3 QR Iteration . 114 3.4 Deation . 115 3.5 Summary . 116 4 Computation of Invariant Subspaces . 116 5 Further Notes and References . 118 i i i i \main" i i 2004/5/6 page vii i i Contents vii 4 QR-Based Algorithms for (Skew-)Hamiltonian Matrices 121 1 Preliminaries . 123 1.1 Elementary Orthogonal Symplectic Matrices . 123 1.2 The Symplectic QR Decomposition . 125 2 The Skew-Hamiltonian Eigenvalue Problem . 127 2.1 Structured Decompositions . 127 2.2 Structured Condition Numbers for Eigenvalues . 130 2.3 Structured Condition Numbers for Invariant Subspaces . 132 Isotropic invariant subspaces and a quadratic matrix equation. Skew- symmetrizing the bottom part. Solving the decoupled linearized equation. Solving the coupled linearized equation. Solving the quadratic matrix equation. Perturbation bounds and a condition y number. On the computation of TW . Numerical Example. 2.4 An Algorithm for Computing thek Skew-Hamiltoniank Schur De- composition . 143 2.5 Computation of Invariant Subspaces . 145 2.6 Other Algorithms . 145 3 The Hamiltonian Eigenvalue Problem . 145 3.1 Structured Decompositions . 145 3.2 Structured Condition Numbers for Eigenvalues . 146 3.3 Structured Condition Numbers for Invariant Subspaces . 147 3.4 An Explicit Hamiltonian QR Algorithm . 148 3.5 Reordering a Hamiltonian Schur Decomposition . 149 3.6 Algorithms Based on H 2 . 150 3.7 Computation of Invariant Subspaces Based on H 2 . 154 3.8 Re¯nement of Stable Invariant Subspaces . 154 3.9 Other Algorithms . 155 4 Symplectic Balancing . 155 4.1 Isolating Eigenvalues . 156 4.2 Scaling . 157 5 Block Algorithms . 159 5.1 A WY-like representation for products of elementary orthogonal symplectic matrices . 159 5.2 Block Symplectic QR Decomposition . 161 5.3 Block Symplectic URV Decomposition . 162 5.4 Numerical Stability . 165 5.5 Numerical Results . 167 6 Numerical Experiments . 168 6.1 Accuracy of Eigenvalues . 169 6.2 Computing Times . 170 5 Krylov-Schur Algorithms 173 1 Basic Tools . 174 1.1 Krylov Subspaces . 174 1.2 The Arnoldi Method . 176 The Arnoldi Decomposition. The Basic Algorithm. 2 Restarting and the Krylov-Schur Algorithm . 179 2.1 Restarting an Arnoldi Decomposition . 179 2.2 The Krylov Decomposition . 180 2.3 Restarting a Krylov Decomposition . 182 i i i i \main" i i 2004/5/6 page viii i i viii Contents 2.4 Deating a Krylov Decomposition . 184 3 A Krylov-Schur Algorithm for Periodic Eigenproblems . 186 3.1 Periodic Arnoldi and Krylov Decompositions . 186 3.2 Restarting a Periodic Krylov Decomposition . 191 3.3 Deating a Periodic Krylov Decomposition . 192 Deation of singular factors. 4 Krylov-Schur Algorithms for (Skew-)Hamiltonian Eigenproblems . 195 4.1 SHIRA . 195 4.2 A Two-Sided Krylov-Schur Algorithm for Hamiltonian Matrices . 197 5 Balancing Sparse General Matrices . 198 5.1 Irreducible forms . 199 5.2 Krylov-Based Balancing . 200 6 Balancing Sparse Hamiltonian Matrices . 201 6.1 Irreducible forms . 202 6.2 Krylov-Based Balancing . ..

View Full Text

Details

  • File Type
    pdf
  • Upload Time
    -
  • Content Languages
    English
  • Upload User
    Anonymous/Not logged-in
  • File Pages
    267 Page
  • File Size
    -

Download

Channel Download Status
Express Download Enable

Copyright

We respect the copyrights and intellectual property rights of all users. All uploaded documents are either original works of the uploader or authorized works of the rightful owners.

  • Not to be reproduced or distributed without explicit permission.
  • Not used for commercial purposes outside of approved use cases.
  • Not used to infringe on the rights of the original creators.
  • If you believe any content infringes your copyright, please contact us immediately.

Support

For help with questions, suggestions, or problems, please contact us