## Transfer Functions and Resolvent Norm Approximation of Large Matrices (1998)

Venue: | Electron. Trans. Numer. Anal |

Citations: | 10 - 3 self |

### BibTeX

@ARTICLE{Simoncini98transferfunctions,

author = {V. Simoncini and E. Gallopoulos},

title = {Transfer Functions and Resolvent Norm Approximation of Large Matrices},

journal = {Electron. Trans. Numer. Anal},

year = {1998},

volume = {7},

pages = {190--201}

}

### OpenURL

### Abstract

. A unifying framework for methods employed in the approximation of the resolvent norm of nonnormal matrices is presented. This formulation uses specific transfer functions, and it provides new information about the approximation properties of these methods and their application in computing the pseudospectrum of matrices. Key words. Resolvent norm, transfer function, Arnoldi iteration, pseudospectrum. AMS subject classification. 65F15. 1. Introduction. We now know that the analysis of matrix-dependent algorithms is considerably more complicated when nonnormal matrices are involved; see for example [5]. In particular, several studies indicate that the eigenvalues of the matrix in question often provide insufficient or even misleading information [20]. This has been the motivation behind recent research on more reliable indicators as well as on methods for their practical computation. Several studies concur that a better accordance between theory and practice can be achieved by using ...

### Citations

237 | Users’ guide for the harwell-boeing sparse matrix collection (Release I
- Duff, Grimes, et al.
- 1992
(Show Context)
Citation Context ... with elements a kk = s k\Gamma1 and a kj = \Gammas k\Gamma1 c when j ? k, where s n\Gamma1 = 0:1 and s 2 + c 2 = 1 [19]; and iv) the Tolosa matrix of order n = 135 from the Harwell-Boeing collection =-=[5, 6]-=-. 2. Krylov subspace methods and the transfer function framework. We next recall some properties of Krylov subspace reduction and show how it naturally induces structured perturbations. We also show t... |

163 |
Hydrodynamic stability without eigenvalues
- Trefethen, Trefethen, et al.
- 1993
(Show Context)
Citation Context ... nonnormal matrices are involved; see for example [5]. In particular, several studies indicate that the eigenvalues of the matrix in question often provide insufficient or even misleading information =-=[20]-=-. This has been the motivation behind recent research on more reliable indicators as well as on methods for their practical computation. Several studies concur that a better accordance between theory ... |

92 | Pseudospectra of Matrices - Trefethen - 1991 |

50 | The Test Matrix Toolbox for Matlab, version 3.0, Numerical Analysis
- Higham
- 1995
(Show Context)
Citation Context ...value, the computation of kR(z)k relies on algorithms for singular values or eigenvalues. Standard techniques are adequate when the matrix dimension is small. For instance, the MATLAB function pscont =-=[11]-=-, plots level curves by first computing the complete singular value decomposition of A using the Golub-Reinsch SVD algorithm. An alternative approach that appears to lend itself to parallel computatio... |

35 | L.: Calculation of pseudospectra by the Arnoldi iteration
- Toh, Trefethen
- 1996
(Show Context)
Citation Context ...ense matrix" approaches. For that reason, projection-type methods for approximatingskR(z)k that reduce A and reference it only in matrix-vector products have been the subject of recent investigat=-=ions [3, 4, 10, 15, 16, 18]-=-. In this paper we introduce a unifying framework, based on transfer functions, for reduction based approximations of the resolvent norm of nonnormal matrices and apply it on Received February 2, 1998... |

30 |
A Block Lánczos Method for Computing the Singular Values of Corresponding Singular Vectors of a Matrix
- Golub, Luk, et al.
- 1981
(Show Context)
Citation Context ... ) jjsmax 2(T m ) jj j oe \Gamma1 . By starting the Lanczos recurrence with q 1 = [q 1 ; 0], only half of the elements of the basis Qm are nonzero. These correspond to the first or second n locations =-=[8]-=-. Hence, by splitting Qm and r as Qm = [U 1 ; U 2 ] and r = [r 1 ; r 2 ], respectively, we obtain (A \Gamma zI) \Gamma1 U 2 = U 1 Tm+r 1 e m with U i U i = I , i = 1; 2, so that G z (A; U 2 ; U 1 ) :=... |

28 | Computing the field of values and pseudospectra using the Lanczos method with continuation
- Braconnier, Higham
- 1996
(Show Context)
Citation Context ...ense matrix" approaches. For that reason, projection-type methods for approximatingskR(z)k that reduce A and reference it only in matrix-vector products have been the subject of recent investigat=-=ions [3, 4, 10, 15, 16, 18]-=-. In this paper we introduce a unifying framework, based on transfer functions, for reduction based approximations of the resolvent norm of nonnormal matrices and apply it on Received February 2, 1998... |

23 |
Spectral portraits of matrices and criteria of spectrum dichotomy
- Godunov
- 1992
(Show Context)
Citation Context ...omputation. Several studies concur that a better accordance between theory and practice can be achieved by using the resolvent function R(z) = (A \Gamma zI) \Gamma1 , for A 2 C n\Thetan and z 2 C ` C =-=[7, 19]. The-=- following, well-known, result shows that studying the variation of the resolvent norm corresponds to analyzing how the eigenvalues change under perturbations of A: For any " ? 0, the locus of po... |

15 |
Spectral portrait for nonHermitian large sparse matrices
- Carpraux, Erhel, et al.
- 1994
(Show Context)
Citation Context ...ense matrix" approaches. For that reason, projection-type methods for approximatingskR(z)k that reduce A and reference it only in matrix-vector products have been the subject of recent investigat=-=ions [3, 4, 10, 15, 16, 18]-=-. In this paper we introduce a unifying framework, based on transfer functions, for reduction based approximations of the resolvent norm of nonnormal matrices and apply it on Received February 2, 1998... |

10 |
Operator coefficient methods for linear equations
- Grcar
- 1989
(Show Context)
Citation Context ...e size will become clear from the context. For the numerical experiments we use the following nonnormal matrices: i) the pentadiagonal Toeplitz Grcar matrix matrix A = Toeplitz([\Gamma1; 1; 1; 1; 1]) =-=[9]-=-, where unless indicated otherwise, the Grcar matrix will be of order n = 100; ii) the bidiagonal matrix of order n = 64 with elements a kk = \Gamma0:3k in the diagonal and a k;k+1 = 1 in the first up... |

9 |
Spectral portraits computation by a Lanczos method, normal equation version
- Marques, Toumazou
- 1995
(Show Context)
Citation Context |

9 | The rational Krylov algorithm for large nonsymmetric eigen-values—mapping the resolvent norms (pseudospectrum). Unpublished note
- RUHE
(Show Context)
Citation Context |

8 | Parallel computation of spectral portrait of large matrices by Davidson type methods
- Heuveline, Philippe, et al.
- 1997
(Show Context)
Citation Context |

8 | Matrix Approximation Problems and Nonsymmetric Iterative Methods - Toh - 1996 |

2 | Complete iterative method for computing pseudospectra
- Braconnier
- 1997
(Show Context)
Citation Context ...spectrum, where the largest singular value of (A \Gamma zI) \Gamma1 is less dominant. This property is also common in projection type approximations, which are most successful for z near the spectrum =-=[2]-=-. If A has special structure, E and D could be constructed to focus on parts of A that best capture the nonnormality. One such example is when A is upper triangular with large off--diagonal elements, ... |

2 |
Une m'ethode stable de bloc-diagonalisation de matrices: Application au calcul de portrait spectral
- Lavall'ee, Sadkane
- 1997
(Show Context)
Citation Context ...sition of A using the Golub-Reinsch SVD algorithm. An alternative approach that appears to lend itself to parallel computation is to first block diagonalize A and then approximate its singular values =-=[14]. Many app-=-lications of interest, however, lead to matrices that are large, sparse and possibly structured, properties that are not exploited in the aforementioned "dense matrix" approaches. For that r... |

1 |
GALLOPOULOS,On transfer functions and approximation of resolvent norms applied to the computation of matrix pseudospectra
- BEKAS, KOUTIS, et al.
(Show Context)
Citation Context ...basis of an n \Theta (m + 1) matrix with real random entries uniformly distributed ETNA Kent State University etna@mcs.kent.edu 200 Transfer functions and resolvent norm approximation in the interval =-=[0; 1]-=- and m = 10; here D = [I ; 0]E 2 R n\Thetam . Note that a less accurate approximation is obtained away from the spectrum, where the largest singular value of (A \Gamma zI) \Gamma1 is less dominant. Th... |