Abstract
This paper concentrates on perturbation theory concerning the tensor T-eigenvalues within the framework of tensor-tensor multiplication. Notably, it serves as a cornerstone for the extension of semidefinite programming into the domain of tensor fields, referred to as T-semidefinite programming. The analytical perturbation analysis delves into the sensitivity of T-eigenvalues for third-order tensors with square frontal slices, marking the first main part of this study. Three classical results from the matrix domain into the tensor domain are extended. Firstly, this paper presents the Gershgorin disc theorem for tensors, demonstrating the confinement of all T-eigenvalues within a union of Gershgorin discs. Afterward, generalizations of the Bauer-Fike theorem are provided, each applicable to different cases involving tensors, including those that are F-diagonalizable and those that are not. Lastly, the Kahan theorem is presented, addressing the perturbation of a Hermite tensor by any tensors. Additionally, the analysis establishes connections between the T-eigenvalue problem and various optimization problems. The second main part of the paper focuses on tensor pseudospectra theory, presenting four equivalent definitions to characterize tensor \(\varepsilon \)-pseudospectra. Accompanied by a thorough analysis of their properties and illustrative visualizations, this section also explores the application of tensor \(\varepsilon \)-pseudospectra in identifying more T-positive definite tensors.
Similar content being viewed by others
Data availability
No data was used for the research described in the article.
References
Bauer, F.L., Fike, C.T.: Norms and exclusion theorems. Numer. Math. 2, 137–141 (1960)
Beik, F., Saad, Y.: On the tubular eigenvalues of third-order tensors. arXiv preprint arXiv:2305.06323 (2023)
Braman, K.: Third-order tensors as linear operators on a space of matrices. Linear Algebra Appl. 433(7), 1241–1253 (2010)
Brazell, M., Li, N., Navasca, C., Tamon, C.: Solving multilinear systems via tensor inversion. SIAM J. Matrix Anal. Appl. 34(2), 542–570 (2013)
Cao, Z., Xie, P.: Perturbation analysis for t-product-based tensor inverse, Moore-Penrose inverse and tensor system. Commun. Appl. Math. Comput. 4(4), 1441–1456 (2022)
Cao, Z., Xie, P.: On some tensor inequalities based on the t-product. Linear Multilinear Algebra 71(3), 377–390 (2023)
Chang, S.Y., Wei, Y.: T-product tensors—part II: tail bounds for sums of random T-product tensors. Comput. Appl. Math. 41(3), Paper No. 99, 32 (2022)
Chang, S.Y., Wei, Y.: T-square tensors—Part I: inequalities. Comput. Appl. Math. 41(1), Paper No. 62, 27 (2022)
Chen, C., Surana, A., Bloch, A.M., Rajapakse, I.: Multilinear control systems theory. SIAM J. Control. Optim. 59(1), 749–776 (2021)
Chen, J., Ma, W., Miao, Y., Wei, Y.: Perturbations of Tensor-Schur decomposition and its applications to multilinear control systems and facial recognitions. Neurocomputing 547 Art. 126359, (2023)
Chu, K.-W.E.: Generalization of the Bauer-Fike theorem. Numer. Math. 49(6), 685–691 (1986)
Cui, Y.-N., Ma, H.-F.: The perturbation bound for the T-Drazin inverse of tensor and its application. Filomat 35(5), 1565–1587 (2021)
Davis, P.J.: Circulant Matrices, 2nd edn. Wiley, New York (1979)
Golub, G.H., Van Loan, C.F.: Matrix Computations, 4th edn. Johns Hopkins University Press, Baltimore (2013)
Greenbaum, A., Li, R.C., Overton, M.L.: First-order perturbation theory for eigenvalues and eigenvectors. SIAM Rev. 62(2), 463–482 (2020)
Hachimi, A.E., Jbilou, K., Ratnani, A., Reichel, L.: Spectral computation with third-order tensors using the t-product. Appl. Numer. Math. 193, 1–21 (2023)
Han, F., Miao, Y., Sun, Z., Wei, Y.: T-ADAF: adaptive data augmentation framework for image classification network based on tensor T-product operator. Neural Process. Lett. 55, 10993–11016 (2023)
Hao, N., Kilmer, M.E., Braman, K., Hoover, R.C.: Facial recognition using tensor-tensor decompositions. SIAM J. Imaging Sci. 6(1), 437–463 (2013)
Horn, R.A., Johnson, C.R.: Matrix Analysis, 2nd edn. Cambridge University Press, Cambridge (2013)
Kato, T.: Perturbation Theory for Linear Operators. Springer-Verlag, New York (1966)
Kilmer, M.E., Braman, K., Hao, N.: Third-order tensors as operators on matrices: A theoretical and computational framework with applications in imaging. Technical Report 2011-01, Tufts University (2011). https://www.cs.tufts.edu/t/tr/techreps/TR-2011-01
Kilmer, M.E., Braman, K., Hao, N., Hoover, R.C.: Third-order tensors as operators on matrices: a theoretical and computational framework with applications in imaging. SIAM J. Matrix Anal. Appl. 34(1), 148–172 (2013)
Kilmer, M.E., Horesh, L., Avron, H., Newman, E.: Tensor-tensor algebra for optimal representation and compression of multiway data. Proc. Natl. Acad. Sci. USA 118(28), Paper No. e2015851,118, 12 (2021)
Kilmer, M.E., Martin, C.D.: Factorization strategies for third-order tensors. Linear Algebra Appl. 435(3), 641–658 (2011)
Kilmer, M.E., Martin, C.D., Perrone, L.: A third-order generalization of the matrix SVD as a product of third-order tensors. Technical Report 2008-4, Tufts University (2008). https://www.cs.tufts.edu/t/tr/techreps/TR-2008-4
Kostić, V. R., Cvetković, Lj., Cvetković, D. Lj.: Pseudospectra localizations and their applications. Numer. Linear Algebra Appl. 23(2), 356–372 (2016)
Li, C., Liu, Q., Wei, Y.: Pseudospectra localizations for generalized tensor eigenvalues to seek more positive definite tensors. Comput. Appl. Math. 38(4), Paper No. 183, 22 (2019)
Liu, W.-H., Jin, X.-Q.: A study on T-eigenvalues of third-order tensors. Linear Algebra Appl. 612, 357–374 (2021)
Liu, Y., Chen, L., Zhu, C.: Improved robust tensor principal component analysis via low-rank core matrix. IEEE J. Sel. Top. Signal Process. 12(6), 1378–1389 (2018)
Liu, Y., Ma, H.: Weighted generalized tensor functions based on the tensor-product and their applications. Filomat 36(18), 6403–6426 (2022)
Lu, C., Feng, J., Chen, Y., Liu, W., Lin, Z., Yan, S.: Tensor robust principal component analysis with a new tensor nuclear norm. IEEE Trans. Pattern Anal. Mach. Intell. 42(4), 925–938 (2019)
Lund, K.: The tensor t-function: a definition for functions of third-order tensors. Numer. Linear Algebra Appl. 27(3), e2288, 17 (2020)
Lund, K., Schweitzer, M.: The Fréchet derivative of the tensor t-function. Calcolo 60(3), Paper No. 35, 34 (2023)
Luo, Y.S., Zhao, X.L., Jiang, T.X., Chang, Y., Ng, M.K., Li, C.: Self-supervised nonlinear transform-based tensor nuclear norm for multi-dimensional image recovery. IEEE Trans. Image Process. 31, 3793–3808 (2022)
Miao, Y., Qi, L., Wei, Y.: Generalized tensor function via the tensor singular value decomposition based on the T-product. Linear Algebra Appl. 590, 258–303 (2020)
Miao, Y., Qi, L., Wei, Y.: T-Jordan canonical form and T-Drazin inverse based on the T-product. Commun. Appl. Math. Comput. 3(2), 201–220 (2021)
Miao, Y., Wang, T., Wei, Y.: Stochastic conditioning of tensor functions based on the tensor-tensor product. Pac. J. Optim. 19(2), 205–235 (2023)
Mo, C., Li, C., Wang, X., Wei, Y.: \(Z\)-eigenvalues based structured tensors: \(\cal{M}_z\)-tensors and strong \(\cal{M}_z\)-tensors. Comput. Appl. Math. 38(4), Paper No. 175, 25 (2019)
Mo, C., Wang, X., Wei, Y.: Time-varying generalized tensor eigenanalysis via Zhang neural networks. Neurocomputing 407, 465–479 (2020)
Newman, E., Kilmer, M.E.: Nonnegative tensor patch dictionary approaches for image compression and deblurring applications. SIAM J. Imaging Sci. 13(3), 1084–1112 (2020)
Olson, B.J., Shaw, S.W., Shi, C., Pierre, C., Parker, R.G.: Circulant matrices and their application to vibration analysis. Appl. Mech. Rev. 66(4), 040803 (2014)
Pakmanesh, M., Afshin, H.: \(M\)-numerical ranges of odd-order tensors based on operators. Ann. Funct. Anal. 13(3), Paper No. 37, 22 (2022)
Qi, L.: Eigenvalues of a real supersymmetric tensor. J. Symbolic Comput. 40(6), 1302–1324 (2005)
Qi, L., Zhang, X.: T-quadratic forms and spectral analysis of T-symmetric tensors. arXiv preprint arXiv:2101.10820 (2021)
Rayleigh, L.: The Theory of Sound, vol. I. Macmillan, London (1927)
Rellich, F.: Perturbation Theory of Eigenvalue Problems. Gordon and Breach Science Publishers, New York-London-Paris (1969)
Schrödinger, E.: Quantisierung als Eigenwertproblem. Annalen Phys. 386(18), 109–139 (1926)
Shi, X., Wei, Y.: A sharp version of Bauer-Fike’s theorem. J. Comput. Appl. Math. 236(13), 3218–3227 (2012)
Stewart, G.W., Sun, J.G.: Matrix Perturbation Theory. Computer Science and Scientific Computing. Academic Press Inc, Boston, MA (1990)
Sun, J.: Matrix Perturbation Analysis (In Chinese). Academic Press, Beijing (1987)
Tang, L., Yu, Y., Zhang, Y., Li, H.: Sketch-and-project methods for tensor linear systems. Numer. Linear Algebra Appl. 30(2), Paper No. e2470, 32 (2023)
Trefethen, L.N., Embree, M.: Spectra and Pseudospectra: The Behavior of Nonnormal Matrices and Operators. Princeton University Press, Princeton, NJ (2005)
Turatti, E.: On tensors that are determined by their singular tuples. SIAM J. Appl. Algebra Geom. 6(2), 319–338 (2022)
Wang, X., Che, M., Mo, C., Wei, Y.: Solving the system of nonsingular tensor equations via randomized Kaczmarz-like method. J. Comput. Appl. Math. 421, Paper No. 114,856, 15 (2023)
Wang, X., Wei, P., Wei, Y.: A fixed point iterative method for third-order tensor linear complementarity problems. J. Optim. Theory Appl. 197(1), 334–357 (2023)
Wang, Y., Yang, Y.: Hot-SVD: higher order t-singular value decomposition for tensors based on tensor-tensor product. Comput. Appl. Math. 41(8), Paper No. 394, 33 (2022)
Wei, P., Wang, X., Wei, Y.: Neural network models for time-varying tensor complementarity problems. Neurocomputing 523, 18–32 (2023)
Wu, T.: Graph regularized low-rank representation for submodule clustering. Pattern Recognit. 100, Art. 107145, (2020)
Yang, Y., Zhang, J.: Perron-Frobenius type theorem for nonnegative tubal matrices in the sense of \(t\)-product. J. Math. Anal. Appl. 528(2), Paper No. 127, 541, 17 (2023)
Zhao, X.L., Xu, W.H., Jiang, T.X., Wang, Y., Ng, M.K.: Deep plug-and-play prior for low-rank tensor completion. Neurocomputing 400, 137–149 (2020)
Zheng, M.M., Huang, Z.H., Wang, Y.: T-positive semidefiniteness of third-order symmetric tensors and T-semidefinite programming. Comput. Optim. Appl. 78(1), 239–272 (2021)
Acknowledgements
The authors would like to thank the handling editor and two referees for their very detailed comments. Changxin Mo acknowledges support from the National Natural Science Foundation of China (Grant No. 12201092), the Natural Science Foundation Project of CQ CSTC (Grant No. CSTB2022NSCQ-MSX0896), the Science and Technology Research Program of Chongqing Municipal Education Commission (Grant No. KJQN202200512), the Chongqing Talents Project (Grant No. cstc2022ycjh-bgzxm0040), and the Research Foundation of Chongqing Normal University (Grant No. 21XLB040), P. R. of China. Weiyang Ding’s research is supported by the Science and Technology Commission of Shanghai Municipality under grants 23ZR1403000, 20JC1419500, and 2018SHZDZX0. Yimin Wei is supported by the National Natural Science Foundation of China under Grant 12271108, the Ministry of Science and Technology of China under grant G2023132005L and the Science and Technology Commission of Shanghai Municipality under grant 23JC1400501.
Author information
Authors and Affiliations
Corresponding author
Ethics declarations
Conflict of interest
The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.
Additional information
Communicated by Guoyin Li.
Publisher's Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Appendix
Appendix
Proof of Theorem 3
The theorem is evident when \(\mu \in \varLambda (\mathcal {A})\), as the left-hand sides of (12) and (13) vanish. Therefore, we assume that \(\mu \notin \varLambda (\mathcal {A})\). By Lemma 1, we can see that
and moreover \(\mu I_{mn} - {\text {bcirc}}(\mathcal {A} + \varepsilon \mathcal {B}) \) is singular. This means that
is also singular since the matrices multiplied on the left and right sides are nonsingular. Notice that
Therefore (16) can be rewritten as
and thus the following matrix
is singular.
By the assumption that \(|N|^q = 0\) and note that \(\mu I_{mn} - D \) is a nonsingular diagonal matrix, it follows that \(((\mu I_{mn} - D)^{-1}N)^q = 0\). Hence,
and
holds under the 1-, 2- and \(\infty \)-norms cases.
If \(\min _{\lambda \in \varLambda (\mathcal {A})}|\lambda -\mu |\ge 1\), then
In the case of \(\min _{\lambda \in \varLambda (\mathcal {A})}|\lambda -\mu |<1\), then
By (17), we can obtain
Combining the above theoretical analyses, in the spectral norm case, we can see
or
for
respectively. Let \(\theta = \Vert {\text {bcirc}}(\varepsilon \mathcal {B}) \Vert _2 \sum _{k=0}^{q-1}\Vert N\Vert _2^{k} \). Then we get the result (12) for the spectral norm. Similar to the proof given in Theorem 2, the Frobenius norm case can be obtained easily.
For the 1- and \(\infty \)-norms, by using (17) again we get
where
The proof is completed. \(\square \)
Proof of Theorem 4
We only need to consider the case that \(\mu \) is not a T-eigenvalue of \(\mathcal {A}\). Hence \(\mu I_{mn} - \tilde{D}-\tilde{N}\) is nonsingular. Similar as (17), the matrix
is singular. By using a similar proof process given for Theorem 3 in the above, we could get the conclusion. \(\square \)
Proof of Theorem 8
To prove the assertion of (I), we utilize (14) and Remark 5, which allow us to exploit the properties of \(\varLambda _{\varepsilon }(A^{(i)})\) for each matrix \(A^{(i)}\). It is well-known that \(\varLambda _{\varepsilon }(A^{(i)})\) is nonempty, open, and bounded, with at most m connected components, each containing one or more eigenvalues of \(A^{(i)}\) [52, Theorem 2.4]. Consequently, these same properties also hold for the given tensor \(\mathcal {A}\) by the above analysis. Additionally, the number of connected components is bounded by nm due to the relationship expressed by
Now, we proceed to part (II). Denote \(\left( F_{n}^{\textrm{H}} \otimes I_{m}\right) {\text {bcirc}}(\mathcal {A}) \left( F_{n}\otimes I_{m}\right) :=A\). First, note that
Therefore, for any \(c\in \mathbb {C}\), we have
We complete the proof of this part.
For part (III), by Lemma 1, we know that
which implies that
since for any nonzero \(c\in \mathbb {C}\) and matrix \(A\in \mathbb {C}^{m\times m}\), the following equality
holds [52, Theorem 2.4]. Thus we get the result that \(\varLambda _{|c| \varepsilon }(c \mathcal {A})=c \varLambda _{\varepsilon }(\mathcal {A})\) for any nonzero \(c \in \mathbb {C}\).
Now, we prove the last part of this theorem. By Lemma 1, we know that
Therefore,
where the conclusion \(\varLambda _{\varepsilon }(A^{\textrm{H}}) = \overline{ \varLambda _{\varepsilon }(A)}\) under the two-norm case for any matrix \(A\in \mathbb {C}^{m\times m}\) [52, Theorem 2.4] is applied in the second equality. \(\square \)
Proof of Theorem 9
If \(\lambda \) is a T-eigenvalue of tensor \(\mathcal {A}\), then it is an eigenvalue of the matrix \({\text {bcirc}}(\mathcal {A})\). Therefore, for any \(\mu \in \mathbb {C}\), \(\lambda + \mu \) is an eigenvalue of \({\text {bcirc}}(\mathcal {A}) + \mu I\). Note that \(\Vert \mu I\Vert = |\mu |\), and by the definition of pseudospectra on tensors, we obtain \(\lambda + \mu \in \varLambda _{\varepsilon }(\mathcal {A})\) for any \(|\mu | < \varepsilon \). Thus, we have completed the proof of (15).
For the normal tensor case, by Lemmas 4 and 1, we obtain
and
in which \(D^{(i)}\) is diagonal for \(i = 1, \cdots , n\) by Lemma 3. Also note that \(\Vert \cdot \Vert =\Vert \cdot \Vert _{2}\), we may assume directly that \(\mathcal {A}\) is F-diagonal. Therefore, the diagonal entries of \({\text {bcirc}}(\mathcal {A})\) are exactly the T-eigenvalues. As we all know, the \(\varepsilon \)-pseudospectra is just the union of the open \(\varepsilon \)-balls about the points of the spectra for any normal matrix; equivalently, we have
which implies
by the \(\varepsilon \)-pseudospectra of tensors. We get the conclusion since \(\varLambda (\mathcal {A})+\varDelta _{\varepsilon }\) is the same as \(\{z: {\text {dist}}(z, \varLambda (\mathcal {A}))<\varepsilon \}\). \(\square \)
Rights and permissions
Springer Nature or its licensor (e.g. a society or other partner) holds exclusive rights to this article under a publishing agreement with the author(s) or other rightsholder(s); author self-archiving of the accepted manuscript version of this article is solely governed by the terms of such publishing agreement and applicable law.
About this article
Cite this article
Mo, C., Ding, W. & Wei, Y. Perturbation Analysis on T-Eigenvalues of Third-Order Tensors. J Optim Theory Appl (2024). https://doi.org/10.1007/s10957-024-02444-z
Received:
Accepted:
Published:
DOI: https://doi.org/10.1007/s10957-024-02444-z
Keywords
- Perturbation analysis
- T-eigenvalues
- Tensor-tensor multiplication
- T-positive semidefiniteness
- Pseudospectra theory
- Gershgorin disc theorem
- Bauer-Fike theorem
- Kahan theorem