<inline-formula><tex-math id="M1">\begin{document}$ Z $\end{document}</tex-math></inline-formula>-eigenvalue exclusion theorems for tensors

Gang Wang, Yuan ZhangJournal of Industrial and Management Optimization 2019 · DOI

TL;DR

This paper introduces new "exclusion sets" for Z-eigenvalues of tensors, which are regions guaranteed not to contain any Z-eigenvalues. By defining these exclusion sets, the authors derive tighter inclusion sets, improving the precision with which Z-eigenvalues can be located. This is crucial for applications where accurate Z-eigenvalue computation is essential.

What Problem Does This Paper Address?

Locating Z-eigenvalues of tensors is a challenging problem in tensor analysis, more complex than finding H-eigenvalues due to the non-homogeneous nature of the Z-eigenvalue problem. While existing methods provide "inclusion sets" that bound the possible locations of Z-eigenvalues, these sets can be quite broad. The paper aims to refine these bounds by establishing "exclusion sets" – regions that are guaranteed not to contain any Z-eigenvalues. By subtracting these exclusion sets from existing inclusion sets, the authors create tighter, more precise regions for Z-eigenvalue localization, which is vital for various applications like medical resonance imaging and data analysis.

Methods

The core contribution of this paper lies in the development of three novel Z-eigenvalue exclusion theorems for tensors. These theorems are derived by analyzing the properties of Z-eigenvectors and their corresponding Z-eigenvalues. Specifically, the authors leverage the definition of Z-eigenpairs and manipulate inequalities derived from the Z-eigenvalue equation.

The general approach involves:

  1. Defining Z-eigenpairs: A pair $(\lambda, x)$ is a Z-eigenpair if $\lambda$ is a scalar and $x$ is a non-zero vector such that $Ax^{m-1} = \lambda x^m$ and $x^Tx = 1$, where $A$ is an $m$-order $n$-dimensional tensor and $x^m$ is a vector with components $x_i^m$.
  2. Exploiting eigenvector properties: The proofs utilize the fact that an eigenvector $x$ with $\|x\|_2 = 1$ must have at least one non-zero component. The authors identify a component with the largest modulus, denoted $x_t$, and use its properties to derive inequalities.
  3. Constructing exclusion sets: By manipulating these inequalities, the authors derive conditions that a Z-eigenvalue $\lambda$ cannot satisfy if it belongs to certain regions. These regions form the exclusion sets.
  4. Forming tighter inclusion sets: The new inclusion sets are constructed by taking the difference between existing, broader inclusion sets (like Gersgorin-type or Brauer-type sets) and the newly derived exclusion sets.

The paper builds upon existing work on Z-eigenvalue inclusion sets, such as those by Chang et al. [2] and Wang et al. [24], and aims to improve upon them.

Datasets

Not discussed in the paper. The paper focuses on theoretical development and uses numerical examples to illustrate the effectiveness of the proposed theorems.

Key Insights and Ideas

  • The power of exclusion: The fundamental insight is that by precisely identifying regions where Z-eigenvalues cannot exist, we can significantly narrow down the search space for those that do exist. This is a more constructive approach than simply expanding existing inclusion sets.
  • Leveraging eigenvector structure: The proofs cleverly exploit the properties of the eigenvector with the largest modulus to derive crucial inequalities. This highlights the importance of understanding the geometric and algebraic structure of eigenvectors in tensor analysis.
  • Generalizability: The theorems are established for general tensors, making them applicable to a wide range of problems where tensor Z-eigenvalues are relevant.
  • Tighter bounds mean better computation: The resulting tighter inclusion sets are not just theoretically interesting; they have practical implications for numerical algorithms that rely on eigenvalue localization for convergence or efficiency.

Key Results

The paper establishes three main Z-eigenvalue exclusion theorems, leading to three new, tighter Z-eigenvalue inclusion sets for general tensors:

  1. Theorem 3.1: Introduces the inclusion set $\Theta(A) = \bigcup_{i \in N} \Theta_i(A)$, where $\Theta_i(A) = K_i(A) \setminus \Delta_i(A)$. Here, $K_i(A)$ are Gersgorin-type disks, and $\Delta_i(A)$ is a newly defined exclusion set.
  2. Theorem 3.2: Proposes the inclusion set $\Phi(A) = \bigcup_{i \in N} \bigcup_{j \in N, j \neq i} \Phi_{i,j}(A)$, where $\Phi_{i,j}(A) = L_{i,j}(A) \setminus N_{i,j}(A)$. $L_{i,j}(A)$ and $N_{i,j}(A)$ are Brauer-type inclusion sets from previous work, and $N_{i,j}(A)$ is a new exclusion set.
  3. Theorem 3.3: Introduces the inclusion set $\Psi(A) = \bigcup_{i,j \in N, i \neq j} \Psi_{i,j}(A)$, where $\Psi_{i,j}(A) = N_{i,j}(A) \setminus \Omega_{i,j}(A)$. $N_{i,j}(A)$ are sets from Lemma 2.3, and $\Omega_{i,j}(A)$ is a new exclusion set.

The paper demonstrates through two numerical examples (Example 3.1 and Example 3.2) that the newly derived inclusion sets $\Theta(A)$ and $\Phi(A)$ are indeed tighter than existing results like $K(A)$ and $L(A)$. The exclusion sets $\Delta(A)$ and $\Omega(A)$ are shown to be disjoint from the Z-eigenvalues.

Limitations and Future Work

  • Comparison of new sets: The paper notes that in Example 3.1, the sets $\Theta(A)$ and $\Phi(A)$ do not include each other, suggesting that these new inclusion sets offer different perspectives and might be complementary.
  • Further exploration of inclusion sets: The authors suggest that it is "worth further considerations to find their proper subsets which do not include any Z-eigenvalues of tensors to exclude them." This implies that there is potential for discovering even more refined exclusion and inclusion sets.
  • Computational complexity: While the paper focuses on theoretical bounds, the practical computational efficiency of Z-eigenvalue algorithms using these new bounds is not explicitly analyzed. Future work could explore how these tighter bounds impact the performance of numerical methods.

Summary updated

Abstract

To locate all $ Z$-eigenvalues of a tensor more precisely, we establish three $Z$-eigenvalue exclusion sets such that all $ Z$-eigenvalues do not belong to them and get three tighter $Z$-eigenvalue inclusion sets of tensor by using these $Z$-eigenvalue exclusion sets. Furthermore, we show that the new inclusion sets are tighter than the existing results via two running examples.
BibTeX
@article{wang2019lt,
  author = {Gang Wang and Yuan Zhang},
  title = {{\<inline-formula\>\<tex-math id="M1"\>\textbackslash\{\}begin\{document\}$ Z $\textbackslash\{\}end\{document\}\</tex-math\>\</inline-formula\>-eigenvalue exclusion theorems for tensors}},
  journal = {Journal of Industrial and Management Optimization},
  year = {2019},
  doi = {10.3934/jimo.2019039},
}
Figure