Skip to main content
Log in

A Riemannian inexact Newton-CG method for constructing a nonnegative matrix with prescribed realizable spectrum

  • Published:
Numerische Mathematik Aims and scope Submit manuscript

Abstract

This paper is concerned with the inverse eigenvalue problem of finding a nonnegative matrix such that it has the prescribed realizable spectrum. We reformulate the inverse eigenvalue problem as an under-determined constrained nonlinear matrix equation over several matrix manifolds. Then we propose a Riemannian inexact Newton-CG method for solving the nonlinear matrix equation. The global and quadratic convergence of the proposed method is established under some assumptions. We also extend the proposed method to the case of prescribed entries. Finally, numerical experiments are reported to illustrate the efficiency of the proposed method.

This is a preview of subscription content, log in via an institution to check access.

Access this article

Price excludes VAT (USA)
Tax calculation will be finalised during checkout.

Instant access to the full article PDF.

Institutional subscriptions

Similar content being viewed by others

References

  1. Absil, P.-A., Mahony, R., Sepulchre, R.: Optimization Algorithms on Matrix Manifolds. Princeton University Press, Princeton (2008)

    Book  Google Scholar 

  2. Baker, C.G.: Riemannian Manifold Trust-Region Methods with Applications to Eigenproblems. Ph.D. thesis. School of Computational Science, Florida State University, Tallahassee (2008)

  3. Bapat, R.B., Raghavan, T.E.S.: Nonnegative Matrices and Applications. Cambridge University Press, Cambridge (1997)

    Book  Google Scholar 

  4. Barrett, W.W., Johnson, C.R.: Possible spectra of totally positive matrices. Linear Algebra Appl. 62, 231–233 (1984)

    Article  MathSciNet  Google Scholar 

  5. Berman, A., Plemmons, R.J.: Nonnegative Matrices in the Mathematical Sciences. Academic Press, New York (1979)

    MATH  Google Scholar 

  6. Bernstein, D.: Matrix Mathematics—Theory, Facts, and Formulas, 2nd edn. Princeton University Press, Princeton (2009)

    MATH  Google Scholar 

  7. Borobia, A., Canogar, R.: The real nonnegative inverse eigenvalue problem is NP-hard. Linear Algebra Appl. 522, 127–139 (2017)

    Article  MathSciNet  Google Scholar 

  8. Boyle, M., Handelman, D.: The spectra of nonnegative matrices via symbolic dynamics. Ann. Math. 133, 249–316 (1991)

    Article  MathSciNet  Google Scholar 

  9. Chen, X., Liu, D.L.: Isospectral flow method for nonnegative inverse eigenvalue problem with prescribed structure. J. Comput. Appl. Math. 235, 3990–4002 (2011)

    Article  MathSciNet  Google Scholar 

  10. Chu, M.T.: Inverse eigenvalue problems. SIAM Rev. 40, 1–39 (1998)

    Article  MathSciNet  Google Scholar 

  11. Chu, M.T., Diele, F., Sgura, I.: Gradient flow method for matrix completion with prescribed eigenvalues. Linear Algebra Appl. 379, 85–112 (2004)

    Article  MathSciNet  Google Scholar 

  12. Chu, M.T., Driessel, K.R.: Constructing symmetric nonnegative matrices with prescribed eigenvalues by differential equations. SIAM J. Math. Anal. 22, 1372–1387 (1991)

    Article  MathSciNet  Google Scholar 

  13. Chu, M.T., Golub, G.H.: Structured inverse eigenvalue problems. Acta Numer. 11, 1–71 (2002)

    Article  MathSciNet  Google Scholar 

  14. Chu, M.T., Golub, G.H.: Inverse Eigenvalue Problems: Theory, Algorithms, and Applications. Oxford University Press, Oxford (2005)

    Book  Google Scholar 

  15. Chu, M.T., Guo, Q.: A numerical method for the inverse stochastic spectrum problem. SIAM J. Matrix Anal. Appl. 19, 1027–1039 (1998)

    Article  MathSciNet  Google Scholar 

  16. de Oliveira, G.N.: Nonnegative matrices with prescribed spectrum. Linear Algebra Appl. 54, 117–121 (1983)

    Article  Google Scholar 

  17. Dedieu, J.P., Priouret, P., Malajovich, G.: Newton’s method on Riemannian manifolds: covariant alpha theory. IMA J. Numer. Anal. 23, 395–419 (2003)

    Article  MathSciNet  Google Scholar 

  18. Dennis, J.E., Schnabel, R.B.: Numerical Methods for Unconstrained Optimization and Nonlinear Equations. SIAM, Philadelphia (1996)

    Book  Google Scholar 

  19. Egleston, P.D., Lenker, T.D., Narayan, S.K.: The nonnegative inverse eigenvalue problem. Linear Algebra Appl. 379, 475–490 (2004)

    Article  MathSciNet  Google Scholar 

  20. Eisenstat, S.C., Walker, H.F.: Globally convergent inexact Newton methods. SIAM J. Optim. 4, 392–422 (1994)

    Article  MathSciNet  Google Scholar 

  21. Ellard, R., Migoc, H.: Connecting sufficient conditions for the symmetric nonnegative inverse eigenvalues problem. Linear Algebra Appl. 498, 521–552 (2016)

    Article  MathSciNet  Google Scholar 

  22. Fiedler, M.: Eigenvalues of nonnegative symmetric matrices. Linear Algebra Appl. 9, 119–142 (1974)

    Article  MathSciNet  Google Scholar 

  23. Friedland, S., Melkman, A.A.: On the eigenvalues of nonnegative Jacobi matrices. Linear Algebra Appl. 25, 239–254 (1979)

    Article  MathSciNet  Google Scholar 

  24. Golub, G.H., Van Loan, C.F.: Matrix Computations, 4th edn. Johns Hopkins University Press, Baltimore (2013)

    MATH  Google Scholar 

  25. Johnson, C.R., Marijuán, C., Paparella, P., Pisonero, M.: The NIEP. arXiv:1703.10992 (2017)

  26. Johnson, C.R., Paparella, P.: Perron spectratopes and the real nonnegative inverse eigenvalue problem. Linear Algebra Appl. 493, 281–300 (2016)

    Article  MathSciNet  Google Scholar 

  27. Karpelevič, F.I.: On the characteristic roots of matrices with nonnegative elements. Izv. Akad. Nauk SSSR Ser. Mat. 15, 361–383 (1951). (in Russian)

    MathSciNet  Google Scholar 

  28. Laffey, T.J., Šmigoc, H.: Nonnegative realization of spectra having negative real parts. Linear Algebra Appl. 416, 148–159 (2006)

    Article  MathSciNet  Google Scholar 

  29. Lin, M.M.: Fast recursive algorithm for constructing nonnegative matrices with prescribed real eigenvalues. Appl. Math. Comput. 256, 582–590 (2015)

    MathSciNet  MATH  Google Scholar 

  30. Loewy, R., London, D.: A note on an inverse problems for nonnegative matrices. Linear Multilinear Algebra 6, 83–90 (1978)

    Article  MathSciNet  Google Scholar 

  31. Luenberger, D.G.: Optimization by Vector Space Methods. Wiley, New York (1969)

    MATH  Google Scholar 

  32. Marijuán, C., Pisonero, M., Soto, R.L.: A map of sufficient conditions for the symmetric nonnegative inverse eigenvalue problem. Linear Algebra Appl. 530, 344–365 (2017)

    Article  MathSciNet  Google Scholar 

  33. Minc, H.: Nonnegative Matrices. Wiley, New York (1988)

    MATH  Google Scholar 

  34. Orsi, R.: Numerical methods for solving inverse eigenvalue problems for nonnegative matrices. SIAM J. Matrix Anal. Appl. 28, 190–212 (2006)

    Article  MathSciNet  Google Scholar 

  35. Paparella, P.: Realizing Suleimanova-type spectra via permutative matrices. Electron. J. Linear Algebra. 31, 306–312 (2016)

    Article  MathSciNet  Google Scholar 

  36. Perfect, H.: Methods of constructing certain stochastic matrices. Duke Math. J. 20, 395–404 (1953)

    Article  MathSciNet  Google Scholar 

  37. Perfect, H.: Methods of constructing certain stochastic matrices. II. Duke Math. J. 22, 305–311 (1955)

    Article  MathSciNet  Google Scholar 

  38. Reams, R.: An inequality for nonnegative matrices and the inverse eigenvalue problem. Linear Multilinear Algebra 41, 367–375 (1996)

    Article  MathSciNet  Google Scholar 

  39. Senata, E.: Non-negative Matrices and Markov Chains, 2nd edn. Springer, New York (2006)

    Google Scholar 

  40. Soto, R.L.: Existence and construction of nonnegative matrices with prescribed spectrum. Linear Algebra Appl. 369, 169–184 (2003)

    Article  MathSciNet  Google Scholar 

  41. Soto, R.L.: Realizability criterion for the symmetric nonnegative inverse eigenvalue problem. Linear Algebra Appl. 416, 783–794 (2006)

    Article  MathSciNet  Google Scholar 

  42. Soto, R.L.: A family of realizability criteria for the real and symmetric nonnegative inverse eigenvalue problem. Numer. Linear Algebra Appl. 20, 336–348 (2013)

    Article  MathSciNet  Google Scholar 

  43. Soules, G.W.: Constructing symmetric nonnegative matrices. Linear Multilinear Algebra 13, 241–251 (1983)

    Article  MathSciNet  Google Scholar 

  44. Suleǐmanova, H.R.: Stochastic matrices with real characteristic numbers. Doklady Akad. Nauk SSSR (NS) 66, 343–345 (1949)

    MathSciNet  MATH  Google Scholar 

  45. Simonis, J.P.: Inexact Newton Methods Applied to Under-Determined Systems. Ph.D. thesis. Department of Mathematical Science, Worcester Polytechnic Institute (2006)

  46. Smith, S.T.: Optimization techniques on Riemannian manifolds. Fields Inst. Commun. 3, 113–136 (1994)

    MathSciNet  MATH  Google Scholar 

  47. Xu, S.F.: An Introduction to Inverse Algebraic Eigenvalue Problems. Beijing: Friedr. Vieweg & Sohn, Braunschweig (1998)

    MATH  Google Scholar 

  48. Yao, T.T., Bai, Z.J., Zhao, Z., Ching, W.K.: A Riemannian Fletcher-Reeves conjugate gradient method for doubly stochastic inverse eigenvalue problems. SIAM J. Matrix Anal. Appl. 37, 215–234 (2016)

    Article  MathSciNet  Google Scholar 

  49. Zhao, Z., Bai, Z.J., Jin, X.Q.: A Riemannian Newton algorithm for nonlinear eigenvalue problems. SIAM J. Matrix Anal. Appl. 36, 752–774 (2015)

    Article  MathSciNet  Google Scholar 

  50. Zhao, Z., Jin, X.Q., Bai, Z.J.: A geometric nonlinear conjugate gradient method for stochastic inverse eigenvalue problems. SIAM J. Numer. Anal. 54, 2015–2035 (2016)

    Article  MathSciNet  Google Scholar 

Download references

Acknowledgements

We are very grateful to the editor and the referees for their valuable comments and suggestions, which have considerably improved this paper.

Author information

Authors and Affiliations

Authors

Corresponding author

Correspondence to Zheng-Jian Bai.

Additional information

The research of Z. Zhao is supported by the National Natural Science Foundation of China (No. 11601112). The research of Z.-J. Bai is partially supported by the National Natural Science Foundation of China (No. 11671337), the Natural Science Foundation of Fujian Province of China (No. 2016J01035), and the Fundamental Research Funds for the Central Universities (No. 20720180008). The research of X.-Q. Jin is supported by the research Grant MYRG2016-00077-FST from University of Macau.

Appendix

Appendix

In this appendix, we establish some basic properties of the product manifold \(\mathbb {R}^{n\times n}\times \mathcal {O}(n)\times \mathcal {V}\) and the differential of G defined in (2.1). We first show that the nonlinear matrix Eq. (2.1) is under-determined for all \(n\ge 2\). The dimension of \(\mathbb {R}^{n\times n}\times \mathcal {O}(n)\times \mathcal {V}\) is given by

$$\begin{aligned} \dim (\mathbb {R}^{n\times n}\times \mathcal {O}(n) \times \mathcal {V}) = n^2 + \frac{n(n-1)}{2} + |\mathcal {J}|, \end{aligned}$$

where \(\mathcal {J}\) is the complementary index set of \(\mathcal {I}\) with respect to the index set \(\mathcal {N}\), and \(|\mathcal {J}|\) is the cardinality of \(\mathcal {J}\). Thus

$$\begin{aligned} \dim (\mathbb {R}^{n\times n}\times \mathcal {O}(n) \times \mathcal {V}) > \dim \mathbb {R}^{n\times n}\quad \text{ for } n\ge 2. \end{aligned}$$

Hence, (2.1) is under-determined for all \(n\ge 2\).

The tangent space of \(\mathbb {R}^{n\times n}\times \mathcal {O}(n) \times \mathcal {V}\) at a point \((S,Q,V)\in \mathbb {R}^{n\times n}\times \mathcal {O}(n) \times \mathcal {V}\) is given by

$$\begin{aligned} \begin{array}{c} T_{(S,Q,V)}\big ( \mathbb {R}^{n\times n}\times \mathcal {O}(n)\times \mathcal {V}\big ) = T_S \mathbb {R}^{n\times n}\times T_Q \mathcal {O}(n)\times T_V \mathcal {V}. \end{array} \end{aligned}$$

Here, \(T_S \mathbb {R}^{n\times n}\), \(T_Q \mathcal {O}(n)\), and \(T_V \mathcal {V}\) are the tangent spaces of \(\mathbb {R}^{n\times n}\), \(\mathcal {O}(n)\), and \(\mathcal {V}\) at \(S\in \mathbb {R}^{n\times n}\), \(Q\in \mathcal {O}(n)\), and \(V\in \mathcal {V}\) accordingly, which are given by [1, p. 42]:

$$\begin{aligned} T_S \mathbb {R}^{n\times n}=\mathbb {R}^{n\times n},\quad T_Q \mathcal {O}(n)= \big \{ Q\varOmega \ | \ \varOmega ^T = -\varOmega ,\; \varOmega \in \mathbb {R}^{n\times n}\big \}, \quad T_V \mathcal {V}= \mathcal {V}. \end{aligned}$$

A retraction R on \(\mathbb {R}^{n\times n}\times \mathcal {O}(n)\times \mathcal {V}\) is given by

$$\begin{aligned} R_{(S,Q,V)} (\xi _S,\zeta _Q,\eta _V) = \big (R_S(\xi _S), R_Q(\zeta _Q),R_{V}(\eta _V)\big ) \end{aligned}$$

for all \((S,Q,V) \in \mathbb {R}^{n\times n}\times \mathcal {O}(n) \times \mathcal {V}\) and \((\xi _S,\eta _Q,\gamma _V)\in T_{(S,Q,V)}\big ( \mathbb {R}^{n\times n}\times \mathcal {O}(n)\times \mathcal {V}\big )\), where \(R_S\), \(R_Q\), and \(R_{V}\) are the retractions on \(\mathbb {R}^{n\times n}\), \(\mathcal {O}(n)\), and \(\mathcal {V}\) accordingly, which may take the following form:

$$\begin{aligned} \left\{ \begin{array}{ccl} R_S(\xi _S) &{}=&{} S + \xi _S, \quad \mathrm{for} \; \xi _S \in T_S\mathbb {R}^{n\times n},\\ R_Q(\zeta _Q) &{}=&{} \mathrm{qf} (Q + \zeta _Q), \quad \mathrm{for} \; \zeta _Q \in T_Q\mathcal {O}(n),\\ R_{V}(\eta _V) &{}=&{} V+\eta _V, \quad \mathrm{for} \; \eta _V \in T_{V} \mathcal {V}. \end{array} \right. \end{aligned}$$

Here, \(\mathrm{qf}(A)\) means the Q factor of the QR decomposition of a nonsingular matrix \(A\in \mathbb {R}^{n\times n}\) in the form of \(A=Q\widetilde{R}\) with \(Q\in \mathcal {O}(n)\) and \(\widetilde{R}\) being an upper triangular matrix with strictly positive diagonal entries. For other choices of retractions on \(\mathcal {O}(n)\), one may refer to [1, pp. 58–59].

We now establish the differential of G. By simple calculation, the differential \(\mathrm {D}G(S,Q,V): T_{(S,Q,V)}\big (\mathbb {R}^{n\times n}\times \mathcal {O}(n) \times \mathcal {V}\big ) \rightarrow T_{G(S,Q,V)}\mathbb {R}^{n\times n}\simeq \mathbb {R}^{n\times n}\) of G at \((S,Q,V) \in \mathbb {R}^{n\times n}\times \mathcal {O}(n) \times \mathcal {V}\) is determined by

$$\begin{aligned} \mathrm {D}G(S,Q,V) [(\varDelta S,\varDelta Q, \varDelta V)] = 2S\odot \varDelta S + [ Q(\varLambda +V)Q^T , \varDelta QQ^T]- Q\varDelta VQ^T \end{aligned}$$

for all \((\varDelta S,\varDelta Q, \varDelta V) \in T_{(S,Q,V)}(\mathbb {R}^{n\times n}\times \mathcal {O}(n) \times \mathcal {V})\). On the other hand, with respect to the Riemannian metric defined in (2.2), the adjoint \((\mathrm {D}G(S,Q,V))^* : T_{G(S,Q,V)}\mathbb {R}^{n\times n}\rightarrow T_{(S,Q,V)}(\mathbb {R}^{n\times n}\times \mathcal {O}(n)\times \mathcal {V})\) of \(\mathrm {D}G(S,Q,V)\) is determined by

$$\begin{aligned}&(\mathrm {D}G(S,Q,V))^* [ \varDelta Z ]\\&\quad = ((\mathrm {D}G(S,Q,V))_1^* [ \varDelta Z ],(\mathrm {D}G(S,Q,V))_2^* [ \varDelta Z ],(\mathrm {D}G(S,Q,V))_3^* [ \varDelta Z ]) \end{aligned}$$

for all \(\varDelta Z\in T_{G(S,Q,V)}\mathbb {R}^{n\times n}\), where for each \(\varDelta Z\in T_{G(S,Q,V)}\mathbb {R}^{n\times n}\),

$$\begin{aligned} \left\{ \begin{array}{rcl} (\mathrm {D}G(S,Q,V))_1^* [ \varDelta Z ] &{}=&{} 2S\odot \varDelta Z, \\ (\mathrm {D}G(S,Q,V))_2^* [ \varDelta Z ] &{}=&{} \displaystyle \frac{1}{2}\big ( [ Q(\varLambda + V)Q^T, (\varDelta Z)^T ] + [ Q(\varLambda + V)^TQ^T, \varDelta Z ]\big )Q, \\ (\mathrm {D}G(S,Q,V))_3^* [ \varDelta Z ] &{}=&{} -W\odot \big (Q^T\varDelta Z Q\big ). \end{array} \right. \end{aligned}$$

Here, \(W\in \mathbb {R}^{n\times n}\) is defined by

$$\begin{aligned} W_{ij}=\left\{ \begin{array}{ll} 0,&{} \text{ if } (i,j) \in \mathcal {I},\\ 1, &{} \text{ otherwise }. \end{array} \right. \end{aligned}$$
(A.1)

Analogously, we can establish the tangent space of the product manifold \(\mathcal {Z}\times \mathcal {O}(n)\times \mathcal {V}\), a retraction on \(\mathcal {Z}\times \mathcal {O}(n)\times \mathcal {V}\), and the differential of H defined in (4.2) and its adjoint. Here, \(\mathcal {Z}\times \mathcal {O}(n)\times \mathcal {V}\) is equipped with the Riemannian metric defined as in (2.2).

Rights and permissions

Reprints and permissions

About this article

Check for updates. Verify currency and authenticity via CrossMark

Cite this article

Zhao, Z., Bai, ZJ. & Jin, XQ. A Riemannian inexact Newton-CG method for constructing a nonnegative matrix with prescribed realizable spectrum. Numer. Math. 140, 827–855 (2018). https://doi.org/10.1007/s00211-018-0982-2

Download citation

  • Received:

  • Revised:

  • Published:

  • Issue Date:

  • DOI: https://doi.org/10.1007/s00211-018-0982-2

Mathematics Subject Classification

Navigation