- Research
- Open access
- Published:
The eigenvalue product bounds of the Lyapunov matrix differential equation and the stability of a class of time-varying nonlinear system
Journal of Inequalities and Applications volume 2019, Article number: 172 (2019)
Abstract
The Lyapunov matrix differential equation plays an important role in many scientific and engineering fields. In this paper, we first give a class relation between the eigenvalue of functional matrix derivative and the derivative of functional matrix eigenvalue. Using this relation, we convert the Lyapunov matrix differential equation into an eigenvalue differential equation. Further, by the Schur theorem, combining Hölder’s integral inequality with arithmetic–geometric average inequality, we provide several lower and upper bounds on the eigenvalue product for the solution of the Lyapunov matrix differential equation. As an application in control and optimization, we show that our bounds can be used to discuss the stability of a class of time-varying nonlinear systems. Finally, we illustrate the superiority and effectiveness of the derived bounds by a numerical example.
1 Introduction
Consider the following Lyapunov matrix differential equation [1]:
where \(A(t)\in \mathbb{C}^{n\times n}\), \(Q(t) = Q^{H}(t)\in \mathbb{C} ^{n\times n}\), and \(Q(t)\geq 0\) are continuous functions of t, and \(P(t)\in \mathbb{C}^{n\times n}\) is the Hermitian positive semidefinite solution of (1).
It is well known that the linear matrix differential equation (1) has many express styles, and it usually can be found in time-varying nonlinear and linear systems. For one thing, consider the time-varying nonlinear system
where \(x(t), r(t)\in \mathbb{C}^{n}\), and \(P(t)\) is the positive definite solution of the Lyapunov matrix differential equation (1). Here \(\Delta f(x(t),\varPhi (t))\) is a nondeterministic term. In Sect. 5, we will show that the eigenvalue bounds of the Lyapunov matrix differential equation (1) can be used to discuss the stability of the time-varying nonlinear system (2).
For another, sometimes we need to consider the time-varying linear system
where \(A,B:[t_{0},\infty )\rightarrow \mathbb{C}^{n\times n}\) are given time-varying matrices. Based on (3), we will also consider the optimal control problem
where \(Q(t)\in \mathbb{C}^{n\times n}\) is Hermitian positive semidefinite, and \(R(t),S\in \mathbb{C}^{n\times n}\) are Hermitian positive definite.
Kucera [2] showed that the optimal value of (3) can be obtained by taking \(u(t)\) of thye form
where \(P(t)\) is the unique Hermitian positive definite solution of the following Riccati matrix differential equation:
with terminal condition \(P(t_{0})=S\).
Observe that when \(B(t)=0\) and \(P(t_{0})=P_{0}\), (4) reduces to (1).
In addition, (1) is also expressed in the following form [3]:
where \(A,Q \in \mathbb{C}^{n\times n}\), \(Q \geq 0\). Obviously, (1) is more general than (6).
Considering their applications, the authors of [1, 3,4,5,6,7,8,9,10,11,12] have paid their attention to the bounds of the solution for the Lyapunov and Riccati matrix differential equations. These include eigenvalue bounds (Mori et al. [3], Hmamed [4], Dai and Bai [5], Zhang and Liu [6], Liu et al. [7]), trace bounds (Liu and He [8], Zhu and Pagilla [9]), norm bounds (Patel and Toda [10]), and matrix bounds (Jodar and Ponsoda [1], Lee [11], Chen et al. [12]).
The paper is organized as follows. In Sect. 2, we show some notation and lemmas. In Sect. 3, we derive lower bounds on the eigenvalue product for (1). In Sect. 4, we present upper bounds on the eigenvalue product for (1). In Sect. 5, we show that our bounds can be used to discuss the stability of a class of time-varying nonlinear systems. In Sect. 6, we give a corresponding numerical example to show the superiority and effectiveness of the derived bounds.
2 Notation and lemmas
Throughout this paper, let \(\mathbb{C}\) denote the set of complex numbers and \(\mathbb{C}^{m\times n}\) denote the set of \(m\times n\) complex numbers. For \(a\in \mathbb{C}\), \(\operatorname{Re}(a)\) is the real part of a. Let \(x=(x_{1},x_{2},\ldots ,x_{n})\) be a real n-element array reordered in nonincreasing order, that is, \(x_{[1]}\geq x_{[2]} \geq \cdots \geq x_{[n]}\). For \(X\in \mathbb{C}^{n\times n}\), we denote by \(X^{H}\), \(X^{-1}\), \(\operatorname{tr}(X)\), and \(\operatorname{det}(X)\) the conjugate transpose, the inverse, the trace, and the determinant of X, respectively. Further, suppose the diagonal elements and eigenvalues of X are \(d(X)=(d_{1}(X),d_{2}(X),\ldots ,d_{n}(X))\) and \(\lambda (X)=( \lambda _{1}(X),\lambda _{2}(X),\ldots ,\lambda _{n}(X))\), respectively. Assume that \(\operatorname{Re}(d_{[1]}(X))\geq \operatorname{Re}(d_{[2]}(X)]\geq \cdots \geq \operatorname{Re}(d _{[n]}(X))\), \(\operatorname{Re}(\lambda _{[1]}(X))\geq \operatorname{Re}(\lambda _{[2]}(X))\geq \cdots \geq \operatorname{Re}(\lambda _{[n]}(X))\). The inequality \(X>(\geq )\ 0\) means that X is a Hermitian positive (nonnegative) matrix. The identity matrix of appropriate dimensions is denoted I.
Let k be an integer with \(1\leq k\leq n\). We denote by \(Q_{k,n}\) the set of all increasing sequences, that is,
Let \(A=(a_{ij})\in \mathbb{C}^{m\times n}\), and let k and r be integers satisfying \(1\leq k\leq m\) and \(1\leq r\leq n\), respectively. For \(\alpha \in Q_{k,m}\) and \(\beta \in Q_{r,n}\), we denote by \(A[\alpha ,\beta ]\) the \(k\times r\) matrix whose \((i,j)\)th entry is \(a_{\alpha _{i},\beta _{j}}\). If α is equal to β, then we simplify the notation to \(A[\alpha ]\).
Definition 2.1
([13, p. 502])
Let \(A\in \mathbb{C}^{n \times n}\) and \(1\leq r\leq n\). The rth compound matrix \(C_{r}(A)\) of A is the matrix with the \((j_{\alpha },j_{\beta })\)th entry \(\operatorname{det}A[\alpha ,\beta ]\), where \(\alpha \in Q_{r,m}\) and \(\beta \in Q_{r,n}\) are ordered lexicographically; \(C_{r}(A)\) is a \(C^{r}_{n}\)th matrix (\(C^{r}_{n}=\frac{n!}{r!(n-r)!}\)).
Definition 2.2
([13, p. 10])
Let x, y be two real n-element arrays. If they satisfy
then we say that x is weakly majorized by y and denote this by \(x \prec _{w} y\).
Definition 2.3
([13, p. 10])
If they satisfy
then we say that x is weakly submajorized by y and denote this by \(x \prec ^{w} y\).
Definition 2.4
([13, p. 11])
If \(x\prec _{w} y\) and
then we say that x is majorized by y and denote this by \(x \prec y\).
Lemma 2.1
([14, p. 69])
If \(X=X^{H} \in \mathbb{C}^{n\times n}\), then there exists a unitary matrix \(U \in \mathbb{C}^{n\times n}\) such that
Lemma 2.2
([15, p. 490])
If \(X(t),Y(t) \in \mathbb{C}^{n \times n}\), then
Lemma 2.3
([16])
If \((x_{1},x_{2},\ldots ,x_{k})>0\), \((y_{1},y _{2},\ldots ,y_{k})>0\), then for any \(k=1,\ldots , n\),
Lemma 2.4
([17])
If \(f_{1},f_{2},\ldots ,f_{n}\) are real positive continuous functions on \([a,b]\) and \(r_{1},r_{2},\ldots ,r _{n} \) are real positive numbers with \(\sum_{i=1}^{n}\frac{1}{r _{i}}=1\), then we get the following Hölder integral inequality:
where the equality holds if and only if \(f_{1}^{r_{1}},f_{2}^{r_{2}}, \ldots ,f_{n}^{r_{n}}\) are effectively proportional.
Lemma 2.5
([13, p. 300, B.1])
Let \(X=X^{H} \in \mathbb{C} ^{n\times n}\). Then
that is,
Lemma 2.6
([13, p. 92, C.1])
If \(x \prec y\), then
Lemma 2.7
([13, p. 166, A.1.d])
If \(x_{[1]}\geq x_{[2]} \geq \cdots \geq x_{[n]}>0\), \(y_{[1]}\geq y_{[2]}\geq \cdots \geq y _{[n]}>0\), and \(x \prec y\), then
Lemma 2.8
([13, p. 208, A.4])
If \(x_{[1]}\geq x_{[2]} \geq \cdots \geq x_{[n]}>0\) and \(y_{[1]}\geq y_{[2]}\geq \cdots \geq y_{[n]}>0\), then for any \(k=1,\ldots , n\),
Lemma 2.9
([13, p. 655])
If \(a_{i}>0\) (\(i=1,\ldots ,n\)), then for \(k=1,\ldots , n\),
Lemma 2.10
([13, p. 136, H.3.b])
If \(x_{[1]}\geq \cdots \geq x_{[n]}\), \(y_{[1]}\geq \cdots \geq y_{[n]}\), and \(x \prec _{w} y\), then for any real array \(u_{[1]}\geq \cdots \geq u_{[n]}\geq 0\),
Lemma 2.11
([13, p. 503, F.2.c])
If \(X\in \mathbb{C}^{n \times n}\) with eigenvalues \(\lambda _{1},\ldots ,\lambda _{n}\), then the eigenvalues of \(C_{r}(X)\) are \(\lambda _{i_{1}}\cdots \lambda _{i_{r}}\), \(1\leq i_{1}<\cdots <i_{r}\leq n\).
3 Lower bounds on eigenvalue product
In this section, we offer some new lower bounds on the eigenvalue product of the solution of the Lyapunov matrix differential equation (1).
Theorem 3.1
Let \(P(t)\) be a Hermitian positive semidefinite solution of (1), then for any \(k=1,\ldots , n\) and \(t\geq t _{0}\),
Proof
By Lemma 2.1 there exists a unitary matrix \(U(t) \in \mathbb{C}^{n\times n}\) such that
Using Lemma 2.2 and substituting (8) into (1) yield
Multiplying \(U(t)\) and \(U^{H}(t)\) on the left and right of (9), respectively, we have
where
Since \(U(t)U^{H}(t)=I\), from (10) it it easy to see that
Since \(D_{P(t)}=U(t)U^{H}(t)D_{P(t)}\), we have
Substituting (13) into (12) implies
Obviously,
and thus by (14) we obtain
By computation we easily see that the diagonal elements of \(D_{P(t)} \dot{U}(t)U^{H}(t)-\dot{U}(t)U^{H}(t) D_{P(t)}\) are all 0. Thus, by (14),
where \(\widetilde{a}_{ii}(t)\), \(\widetilde{q}_{ii}(t)\) (\(i=1,2,\ldots ,n\)) are the diagonal elements of \(\widetilde{A}(t)\) and \(\widetilde{Q}(t)\), respectively.
According to (15), for all \(t\geq t_{0}\), it is evident that
which is equivalent to
By solving this differential equation we get
Hence
which implies that
Applying Lemma 2.3 to (16) yields
By Lemma 2.4 (let \(r_{i}=k\)) and (17) we have
From Lemma 2.5 we have
Then by Lemma 2.6
that is,
By Lemma 2.5 we have
Using Lemma 2.7, we obtain
By (11), substituting (19) and (20) into (18), we have
which completes the proof. □
By Theorem 3.1 we easily get the following result.
Theorem 3.2
Let \(P(t)\) be a Hermitian positive semidefinite solution of (1). Then for any \(k=1,\ldots , n\) and \(t\geq t _{0}\),
By Theorems 3.1 and 3.2 we immediately get the following corollaries.
Corollary 3.1
Let \(P(t)\) be a Hermitian positive semidefinite solution of (1). Then for any \(k=1,\ldots , n\) and \(t\geq t _{0}\),
Corollary 3.2
Let \(P(t)\) be a Hermitian positive semidefinite solution of (1). Then for any \(k=1,\ldots , n\) and \(t\geq t _{0}\),
4 Upper bounds on eigenvalue product
In this section, we offer some new upper bounds on the eigenvalue product of the solution for the Lyapunov matrix differential equation (1).
Theorem 4.1
Let \(P(t)\) be a Hermitian positive definite solution of (1). Then for any \(k=1,\ldots , n\) and \(t\geq t _{0}\),
Proof
By (16) we get
By Lemma 2.5 we get
Then
Substituting (26) into (25) gives
From Lemma 2.5 we have
Then by Lemma 2.6 we have
that is,
By Lemma 2.5 we have
Using Lemma 2.7, we obtain
Substituting (28) and (29) into (27), by Lemma 2.8 we get
Applying Lemma 2.9 to (30), we get
By (31) let
in Lemma 2.10. Then
In terms of (11) and (32), we get
which completes the proof. □
By Theorem 4.1 we immediately have the following corollary.
Corollary 4.1
Let \(P(t)\) be a Hermitian positive definite solution of (1). Then for any \(k=1,\ldots , n\) and \(t\geq t _{0}\),
5 The stability analysis of a class of time-varying nonlinear systems
In this section, we give an application of our eigenvalue bounds in a class of time-varying nonlinear systems. The time-varying nonlinear systems have wide applications in control and optimization, such as quantum mechanics, solid mechanics, parameter identification, electrical systems, and automatic control systems and can be used to discuss the stability of the time-varying nonlinear system (2). Next, we show that our eigenvalue bounds can be used to discuss the stability of a class of time-varying nonlinear systems.
For system (2), take
where \(C_{k}(P(t))\) denotes the kth compound matrix of \(P(t)\), \(P(t)\) is the positive definite solution of the Lyapunov matrix differential equation (1)
and \(E=(I_{n}\ 0)\) is an \(n\times C^{k}_{n}\)th matrix. For E, we have the following three special conditions:
-
(i)
When \(k=1\), \(C_{1}(P(t))=P(t)\), and E is the identity matrix.
-
(ii)
When \(k=n-1\), E is the identity matrix.
-
(iii)
When \(k=n\), \(C_{n}(P(t))=\det (P(t))\), and \(E=1\).
In the following theorem, we give a condition under which the time-varying nonlinear system (2) is uniformly asymptotic stable.
Theorem 5.1
For the time-varying nonlinear system (2), if
then the large range for system (2) is uniformly asymptotically stable, where \(l_{k}(t)\) (\(k=1,\ldots ,n\)) are defined by Theorem 4.1.
Proof
Choose the Lyapunov function
Then
As \(P(t)\) is positive definite, we have \(P(t)P^{-1}(t)=I\). Hence
which means that
Substituting (1), (2), and (36) into (35) yields
Applying Lyapunov stability theory, a sufficient condition of uniform asymptotic stability in large range for system (2) is that, for any \(x(t)\neq 0\),
Using Theorem 4.1, we get
and
In terms of (37), (38), and (39), since \(Q(t)=I\), using Lemma 2.11, we obtain
If condition (34) is satisfied, then substitution of (34) into (40) yields
Hence, using Lyapunov stability theory, the large range for system (2) is uniformly asymptotically stable. □
Remark 5.1
For Theorem 5.1, we have the following three special conditions:
-
(i)
When \(k=1\), \(C_{1}(P(t))=P(t)\), and E is the identity matrix. For system (2), take
$$ \Delta f\bigl(x(t),\varPhi (t)\bigr)= \bigl\Vert x(t) \bigr\Vert ^{2}P^{2}(t)x(t)+B(t)\varPhi (t)r(t). $$If
$$ \frac{-1}{l^{2}_{1}(t)}+2 \bigl\Vert x(t) \bigr\Vert ^{2}l_{1}(t)< 0, $$(41)then system (2) is uniformly asymptotically stable, where \(l_{1}(t)\) is defined in Theorem 4.1.
-
(ii)
When \(k=n-1\), E is the identity matrix. For system (2), take
$$ \Delta f\bigl(x(t),\varPhi (t)\bigr)= \bigl\Vert x(t) \bigr\Vert ^{2}P(t)C_{n-1}\bigl(P(t)\bigr)x(t)+B(t)\varPhi (t)r(t). $$If
$$ \frac{-1}{l^{2}_{1}(t)}+2 \bigl\Vert x(t) \bigr\Vert ^{2}l_{n-1}(t)< 0, $$(42)then system (2) is uniformly asymptotically stable, where \(l_{1}(t)\) and \(l_{n-1}(t)\) are defined in Theorem 4.1.
-
(iii)
When \(k=n\), \(C_{n}(P(t))=\det (P(t))\) and \(E=1\). For system (2), take
$$ \Delta f\bigl(x(t),\varPhi (t)\bigr)= \bigl\Vert x(t) \bigr\Vert ^{2}\det \bigl(P(t)\bigr)P(t)x(t)+B(t)\varPhi (t)r(t). $$If
$$ \frac{-1}{l^{2}_{1}(t)}+2 \bigl\Vert x(t) \bigr\Vert ^{2}l_{n}(t)< 0, $$(43)then system (2) is uniformly asymptotically stable, where \(l_{1}(t)\) and \(l_{n}(t)\) are defined in Theorem 4.1.
6 A numerical example
In this section, we demonstrate the effectiveness of our results and compare our eigenvalue bounds with those of the previous results by a real application example.
Example 6.1
Consider the time-varying nonlinear system (2) with
where \(x(t), r(t)\in \mathbb{C}^{3}\) and
Choosing \(t_{0}=\frac{1}{16}\), we have
The solution of (1) can be expressed as
Using Mathematical tool and (44), we get that
where
Obviously, when \(t>0\), \(P(t)\) is positive definite.
(i) First, we show that the eigenvalue bounds for \(P(t)\) have wide applications in the time-varying nonlinear systems. Using Theorem 4.1, we get
-
(1)
When \(k=1\), take
$$ \Delta f\bigl(x(t),\varPhi (t)\bigr)= \bigl\Vert x(t) \bigr\Vert ^{2}P^{2}(t)x(t)+B(t)\varPhi (t)r(t). $$If
$$ \frac{-1}{l^{2}_{1}(t)}+2 \bigl\Vert x(t) \bigr\Vert ^{2}l_{1}(t)= -\frac{16}{(1+4t)^{2}t ^{\frac{3}{2}}}+ \bigl\Vert x(t) \bigr\Vert ^{2} \frac{1+4t}{2}t^{\frac{3}{4}}< 0, $$then using Theorem 5.1, we get that system (2) is uniformly asymptotically stable.
-
(2)
When \(k=2\), take
$$ \Delta f\bigl(x(t),\varPhi (t)\bigr)= \bigl\Vert x(t) \bigr\Vert ^{2}P(t)C_{2}\bigl(P(t)\bigr)x(t)+B(t)\varPhi (t)r(t). $$If
$$ \frac{-1}{l^{2}_{1}(t)}+2 \bigl\Vert x(t) \bigr\Vert ^{2}l_{2}(t)=- \frac{16}{(1+4t)^{2}t ^{\frac{3}{2}}}+2 \bigl\Vert x(t) \bigr\Vert ^{2}2^{\frac{2}{3}}t^{\frac{17}{12}} \cdot \biggl( \frac{9}{7}\cdot 2^{-\frac{8}{3}}t^{\frac{7}{6}}+ \frac{19}{28}\biggr)^{2}< 0, $$then using Theorem 5.1, we get that system (2) is uniformly asymptotically stable.
-
(3)
When \(k=3\), take
$$ \Delta f\bigl(x(t),\varPhi (t)\bigr)= \bigl\Vert x(t) \bigr\Vert ^{2}\det \bigl(P(t)\bigr)P(t)x(t)+B(t)\varPhi (t)r(t). $$If
$$ \frac{-1}{l^{2}_{1}(t)}+2 \bigl\Vert x(t) \bigr\Vert ^{2}l_{3}(t)=- \frac{16}{(1+4t)^{2}t ^{\frac{3}{2}}}+ \bigl\Vert x(t) \bigr\Vert ^{2} \frac{1}{8}t^{\frac{5}{4}}\biggl(2^{ \frac{11}{3}}t^{\frac{7}{6}}+ \frac{1}{2}\biggr)^{3}< 0, $$then using Theorem 5.1, we get that system (2) is uniformly asymptotically stable.
(ii) Next, we give our upper and lower eigenvalue bounds.
If \(k=2\), then by Theorems 3.2 and 4.1 we obtain
If \(k=3\), then by Theorems 3.2 and 4.1 we have
and
In addition, by solving equation (1) we obtain
We depict the plot (see Fig. 1) based on all these results. In the plot, the dashed line presents the result of (45). The dot line denotes the result of (46). The real line shows the result of (47). From the plot we can get the bounds of the solution of (1).
References
Jodar, L., Ponsoda, E.: Computing the continuous numerical solutions of matrix differential equations. Comput. Math. Appl. 29, 73–84 (1995)
Kucera, V.: A review of the matrix Riccati equation. Kybernetika 9, 42–61 (1973)
Mori, T., Fukuma, N., Kumahara, M.: Bounds in the Lyapunov matrix differential equation. IEEE Trans. Autom. Control 32, 55–57 (1987)
Hmamed, A.: Differential and difference Lyapunov equations: simultaneous eigenvalue bounds. Int. J. Syst. Sci. 21, 1335–1344 (1990)
Dai, H., Bai, Z.Z.: On eigenvalue bounds and iteration methods for discrete algebraic Riccati equations. J. Comput. Math. 29, 341–366 (2011)
Zhang, J., Liu, J.Z.: New estimates for the solution of the Lyapunov matrix differential equation. Electron. J. Linear Algebra 20, 6–19 (2010)
Liu, J.Z., Zhang, J., Liu, Y.: A new upper bound for the eigenvalues of the continuous algebraic Riccati equation. Electron. J. Linear Algebra 20, 314–321 (2010)
Liu, J.Z., He, L.L.: A new trace bound for a general square matrix product. IEEE Trans. Autom. Control 52, 349–352 (2007)
Zhu, Y.L., Pagilla, P.R.: Bounds on the solution of the time-varying linear matrix differential equation \(\dot{P}(t)=A^{H}(t)P(t)+P(t)A(t)+Q(t)\). IMA J. Math. Control Inf. 23, 269–277 (2006)
Patel, R.V., Toda, M.: On norm bounds on the solution of the algebraic Riccati equation. IEEE Trans. Autom. Control 23, 87–88 (1978)
Lee, C.H.: On the robust stability of interval time-delay systems—an application of the upper solution bounds of the Lyapunov equation. J. Franklin Inst. 350, 258–274 (2013)
Chen, Y.H., Yi, C.F., Qiao, D.Y.: Improved neural solution for the Lyapunov matrix equation based on gradient search. Inf. Process. Lett. 113, 876–881 (2013)
Marshall, A.W., Olkin, I., Arnold, B.C.: Inequalities: Theory of Majorization and Its Applications, 2nd edn. Springer, New York (2009)
Marcus, M., Minc, H.: A Survey of Matrix Theory and Matrix Inequalities. Allyn & Bacon, Boston (1964)
Horn, R.A., Johnson, C.R.: Topics in Matrix Analysis. Cambridge University Press, Cambridge (1991)
Zhang, H.F.: The extension of Hölder inequality and isolation of Yang–Zhang inequality. Acta Math. Appl. Sin. 21, 423–427 (1998)
Liu, Z.: A refinement of Hölder’s integral inequality. Tamkang J. Math. 34, 383–385 (2003)
Funding
The work was supported in part by National Natural Science Foundation of China (11771370, 11571292, 11471279), Natural Science Foundation of Hunan Province (2018JJ2376, 2017JJ3305), and Research Foundation of Education Bureau of Hunan Province (16B257, 18B057).
Author information
Authors and Affiliations
Contributions
All the authors participated in every phase of the research conducted for this paper. All authors read and approved the final manuscript.
Corresponding authors
Ethics declarations
Competing interests
The authors have no any competing interests in the subject of this research.
Additional information
Publisher’s Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Rights and permissions
Open Access This article is distributed under the terms of the Creative Commons Attribution 4.0 International License (http://creativecommons.org/licenses/by/4.0/), which permits unrestricted use, distribution, and reproduction in any medium, provided you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made.
About this article
Cite this article
Liu, J., Zhang, J. & Huang, H. The eigenvalue product bounds of the Lyapunov matrix differential equation and the stability of a class of time-varying nonlinear system. J Inequal Appl 2019, 172 (2019). https://doi.org/10.1186/s13660-019-2119-2
Received:
Accepted:
Published:
DOI: https://doi.org/10.1186/s13660-019-2119-2