The Quadratic Eigenvalue Problem
Summary (3 min read)
2. Applications of QEPs.
- A wide variety of applications require the solution of a QEP, most of them arising in the dynamic analysis of structural mechanical, and acoustic systems, in electrical circuit simulation, in fluid mechanics, and, more recently, in modeling microelectronic mechanical systems (MEMS) [28] , [155] .
- QEPs also have interesting applications in linear algebra problems and signal processing.
- The list of applications discussed in this section is by no means exhaustive, and, in fact, the number of such applications is constantly growing as the methodologies for solving QEPs improve.
C
- Figure 2 .2 illustrates the behavior of the solution for different values of the viscous damping factor ζ.
- The properties of systems of second-order differential equations (2.1) for n ≥ 1 have been analyzed in some detail by Lancaster [87] and more recently by Gohberg, Lancaster, and Rodman [62] .
- If an eigenvalue has positive real part, q(t) can grow exponentially toward infinity.
- For the Millennium Bridge it is unwanted.
- By this means, eigenvalues corresponding to unstable modes or yielding large vibrations can be relocated or damped.
2.2. Vibration Analysis of Structural Systems-Modal Superposition
- The futuristic footbridge, spanning the Thames by St Paul's Cathedral, will be fitted with shock absorbers to reduce the alarming 10cm swaying first noticed when a huge crowd surged across at its opening in June.
- The bridge, designed by Lord Foster and closed three days after its opening, will be fitted with "viscous dampers" and "tuned mass dampers"-likened to shock absorbers-by the bridge's engineers, Ove Arup.
2.3. Vibro-Acoustics.
- This application raises the problem of truncation error due to the discretization of operators in an infinite-dimensional space.
- The authors will not treat this aspect in this paper but will restrict ourselves to the study of QEPs of finite dimension.
2.4. Fluid Mechanics.
- When studying the temporal stability, β is a real and fixed wavenumber.
- The imaginary part of ω represents the temporal growth rate.
3.8.1. Spectrum Location.
- Symmetry alone does not guarantee that the eigenvalues are real and that the pencil is diagonalizable by congruence transformations.
- Veselić [147] shows that the overdamping condition (3.17) is equivalent to the definiteness of the symmetric linearization obtained from L1 or L2.
- Hence, for overdamped problems (see section 3.9), the symmetric linearization fully reflects the spectral properties of the QEP (real eigenvalues).
3.8.3. Self-Adjoint Triple, Sign Characteristic, and Factorization.
- All the eigenvalues lie in the left half-plane as shown in Figure 3 .3.
- Figure 3 .4 displays the eigenvalue distribution, with the characteristic gap between the n smallest eigenvalues and the n largest.
4.1. Conditioning.
- Condition numbers for the eigenvalues can be derived without assuming that the eigenvalue is finite.
- Moreover, by working in projective spaces the problem of choosing a normalization for the eigenvectors is avoided.
4.2. Backward Error. A natural definition of the normwise backward error of an approximate eigenpair (
- It is often not appreciated that in finite precision arithmetic these two choices are not equivalent (see the example in section 5.1).
- One approach, adopted by MATLAB 6's polyeig function for solving the polynomial eigenvalue problem and illustrated in Algorithm 5.1 below, is to use whichever part of ξ yields the smallest backward error (4.7).
4.3. Pseudospectra.
- The eigenvalues of the unperturbed QEP, marked by dots, all lie in the left half-plane, so the unperturbed system is stable.
- The plot shows that relative perturbations of order 10 −10 (corresponding to the yellow region of the plot) can move the eigenvalues to the right half-plane, making the system unstable.
5. Numerical Methods for Dense Problems.
- Is the backward error of the GEP solution and, as expected, is of the order of the unit roundoff because the QZ algorithm is a backward stable algorithm for the solution of the GEP.
- For the smallest eigenvalue in modulus, the second choice ξ 2 yields a smaller backward error.
- This example shows that even if the algorithm chooses the part of the vector ξ that yields the smallest backward error, these backward errors can be much larger than the unit roundoff.
- Even though backward stability is not guaranteed, Algorithm 5.1 is the preferred method if M , C, and K have no particular structure and are not too large.
5.2. Symmetric Linearization.
- If the pencil is indefinite, the HR [21] , [23] , LR [124] , and Falk-Langemeyer [45] algorithms can be employed in order to take advantage of the symmetry of the GEP, but all these methods can be numerically unstable and can even break down completely.
- There is a need to develop efficient and reliable numerical methods for the solution of symmetric indefinite GEPs.
5.3. Hamiltonian/Skew-Hamiltonian
- Figure 5 .2 compares the spectrum of the QEP associated with (5.3) when computed by the MATLAB 6 function polyeig and their implementation of Van Loan's square-reduced algorithm [146] .
- Not surprisingly, the eigenvalues computed by polyeig (which uses a companion linearization and the QZ algorithm) do not have a Hamiltonian structure, and some of them have positive real part, suggesting incorrectly that the system described by (5.3) is unstable.
- In contrast, Van Loan's squarereduced algorithm together with the Hamiltonian/skew-Hamiltonian linearization preserves the Hamiltonian structure, yielding pure imaginary computed eigenvalues that confirm the system's stability.
5.4. Sensitivity of the Linearization.
- This leads to the Lanczos method when S is Hermitian and the Arnoldi or non-Hermitian Lanczos method when S is non-Hermitian.
- The authors aim is to explain how existing Krylov subspace methods can be applied to the QEP.
6.2. Projection Methods Applied
- The goal is to add a new vector v k+1 so that the Ritz pair becomes more accurate.
- The subspace is extended by a quadratic Cayley transform applied to the Ritz vector.
7.3. Hamiltonian/Skew-Hamiltonian Linearization-Direct Methods.
- As the authors have stressed throughout this survey, QEPs are of growing importance and many open questions are associated with them.
- The authors hope that their unified overview of applications, theory, and numerical methods for QEPs will stimulate further research in this area.
Did you find this useful? Give us your feedback
Citations
613 citations
606 citations
Cites background from "The Quadratic Eigenvalue Problem"
...Mathematical theory and numerical techniques for quadratic eigenvalue problems can be found in the recent survey [85] and references therein....
[...]
595 citations
Cites methods from "The Quadratic Eigenvalue Problem"
...Gathering the three row vectors into three design matrices, we obtain the following quadratic eigenvalue problem (QEP) [17]:...
[...]
408 citations
Additional excerpts
...Tisseur and Meerbergen [324]....
[...]
References
9,334 citations
8,608 citations
7,422 citations
Related Papers (5)
Frequently Asked Questions (17)
Q2. What is the projection method for L()?
The projection method approximates an eigenvector x of L(λ) by a vector x̃ = V ξ ∈ Kk with corresponding approximate eigenvalue λ̃. AsW ∗L(λ̃)x̃ = W ∗L(λ̃)V ξ = Lk(λ̃)ξ = 0, the projection method forces the residual r = L(λ̃)x̃ to be orthogonal to Lk.
Q3. What is the eigenvalue of the skew-Hamiltonian matrix?
The reduction to a Hamiltonian eigenproblem uses the fact that when the skew-Hamiltonian matrix B is nonsingular, it can be written in factored form asB = B1B2 = [ The author12C 0 M ] [ M 12C 0 The author] with BT2 J = JB1.(5.2)Then H = B−11 AB −1 2 is Hamiltonian.
Q4. What is the simplest way to construct a Krylov subspace?
It produces a non-Hermitian tridiagonal matrix Tk and a pair of matrices Vk and Wk such that W ∗kVk = The authorand whose columns form bases for the Krylov subspaces Kk(S, v) and Kk(S∗, w), where v and w are starting vectors such that w∗v = 1.
Q5. What is the simplest way to measure the robustness of a numerical algorithm?
To measure the robustness of the system, one can take as a global measureν2 = 2n∑ k=1 ω2kκ(λk) 2,where the ωk are positive weights.
Q6. What is the advantage of working with a linearization of Q() and a?
One important advantage of working with a linearization of Q(λ) and a Krylov subspace method is that one can get at the same time the partial Schur decomposition of the single matrix S that is used to define the Krylov subspaces.
Q7. What is the spectral transformation used to approximate eigenvalues?
The shift-and-invert spectral transformation f(λ) = 1/(λ − σ) and the Cayley spectral transformation f(λ) = (λ− β)/(λ− σ) (for β = σ), used to approximate eigenvalues λ closest to the shift σ, are other possible spectral transformations that are discussed in [7], for example.
Q8. What is the way to deflate a Ritz eigenpair?
If (λ̃, x̃) is a converged Ritz eigenpair that belongs to the set of desired eigenvalues, one may want to lock it and then continue to compute the remaining eigenvalues without altering (λ̃, x̃).
Q9. What is the main drawback of the pseudo-Lanczos algorithm?
Parlett and Chen [114] introduced a pseudo-Lanczos algorithm for symmetric pencils that uses an indefinite inner product and respects the symmetry of the problem.
Q10. What is the distribution of the eigenvalues of G() in the complex?
G(λ)∗ = G(−λ̄),(3.19)the distribution of the eigenvalues of G(λ) in the complex plane is symmetric with respect to the imaginary axis.
Q11. What is the common method for finding eigenpairs of large QEPs?
Most algorithms for large QEPs proceed by generating a sequence of subspaces {Kk}k≥0 that contain increasingly accurate approximations to the desired eigenvectors.
Q12. What is the way to solve the eigenvalue problem?
One approach, adopted by MATLAB 6’s polyeig function for solving the polynomial eigenvalue problem and illustrated in Algorithm 5.1 below, is to use whichever part of ξ̃ yields the smallest backward error (4.7).
Q13. What can be done to solve the symmetric indefinite GEP?
If the pencil is indefinite, the HR [21], [23], LR [124], and Falk–Langemeyer [45] algorithms can be employed in order to take advantage of the symmetry of the GEP, but all these methods can be numerically unstable and can even break down completely.
Q14. What is the disadvantage of the Arnoldi and Lanczos methods?
(Note that the matrix Hk is the Galerkin projection of S and not of A− λB.)A major disadvantage of the shift-and-invert Arnoldi and Lanczos methods is that a change of shift σ requires building a new Krylov subspace: all information built with the old σ is lost.
Q15. What is the way to deflate a converged eigenvalue?
If the converged (λ̃, x̃) does not belong to the set of wanted eigenvalues, one may want to remove it from the current subspace Kk.
Q16. What is the oblique projection of L() onto Kk?
In this case Lk is the orthogonal projection of L(λ) onto Kk. When W = V , Lk is the oblique projection of L(λ) onto Kk along Lk.
Q17. What is the reorthogonalization of columns of Vk?
This process does not guarantee orthogonality of the columns of Vk in floating point arithmetic, so reorthogonalization is recommended to improve the numerical stability of the method [32], [135].