Applied Mathematics and Computation 218 (2011) 574–582
Contents lists available at ScienceDirect
Applied Mathematics and Computation journal homepage: www.elsevier.com/locate/amc
Lyapunov inequalities and stability for discrete linear Hamiltonian systems q Qi-ming Zhang a,b, X.H. Tang a,⇑ a b
School of Mathematical Sciences and Computing Technology, Central South University, Changsha, Hunan 410083, PR China College of Science, Hunan University of Technology, Zhuzhou 412007, Hunan, PR China
a r t i c l e
i n f o
a b s t r a c t In this paper, we establish several new Lyapunov type inequalities for discrete linear Hamiltonian systems when the end-points are not necessarily usual zeros, but rather, generalized zeros, which generalize and improve almost all related existing ones. Applying these inequalities, an optimal stability criterion is obtained. Ó 2011 Elsevier Inc. All rights reserved.
Keywords: Discrete Hamiltonian system Lyapunov inequality Generalized zero Stability
1. Introduction Consider a discrete linear Hamiltonian system
DxðnÞ ¼ aðnÞxðn þ 1Þ þ bðnÞyðnÞ;
DyðnÞ ¼ cðnÞxðn þ 1Þ aðnÞyðnÞ;
ð1:1Þ
where a(n), b(n) and c(n) are real-valued functions defined on Z, D denotes the forward difference operator defined by Dx(n) = x(n + 1) x(n). Throughout this paper, we always assume that
bðnÞ P 0;
8n 2 Z: >
ð1:2Þ
r
>
Let u(n) = (x(n), y(n)) , u (n) = (x(n + 1), y(n)) and
J¼
0
1
1 0
;
HðnÞ ¼
cðnÞ aðnÞ : aðnÞ bðnÞ
Then we can rewrite (1.1) as a standard discrete linear Hamiltonian system
DuðnÞ ¼ JHðnÞur ðnÞ:
ð1:3Þ
In recent papers [1–8], some dynamical behaviors of solutions of system (1.3) have been discussed, such as existence of periodic solutions, strong limit points, Weyl–Titchmarsh theory, spectral theory, eigenvalue problems and disconjugacy. For the second-order difference equation
D½pðnÞDxðnÞ þ qðnÞxðn þ 1Þ ¼ 0;
ð1:4Þ
where p(n) > 0, which has been investigated extensively and applied in literatures. If we let y(n) = p(n)Dx(n), then (1.4) can be written as an equivalent Hamiltonian system of type (1.1):
q This work is partially supported by the NNSF of China (No. 10771215) and by Scientific Research Fund of Hunan Provincial Education Department (No. 10C0655). ⇑ Corresponding author. E-mail address:
[email protected] (X.H. Tang).
0096-3003/$ - see front matter Ó 2011 Elsevier Inc. All rights reserved. doi:10.1016/j.amc.2011.05.101
Q.-m. Zhang, X.H. Tang / Applied Mathematics and Computation 218 (2011) 574–582
DxðnÞ ¼
1 yðnÞ; pðnÞ
DyðnÞ ¼ qðnÞxðn þ 1Þ;
575
ð1:5Þ
where
aðnÞ ¼ 0; bðnÞ ¼
1 ; pðnÞ
cðnÞ ¼ qðnÞ:
In the meantime, system (1.1) is also a discrete analogy of the following first-order polar Hamiltonian system
x0 ðtÞ ¼ aðtÞxðtÞ þ bðtÞyðtÞ;
y0 ðtÞ ¼ cðtÞxðtÞ aðtÞyðtÞ:
ð1:6Þ
In the discrete case, instead of usual zero, we adopt the following concept of generalized zero, which is due to Hartman [9]. Definition 1.1 [9]. A function f : Z ! R is said to have a generalized zero at n0 2 Z provided either f(n0) = 0 or f(n0)f(n0 + 1) < 0. It is a classical topic for us to study Lyapunov type inequalities which have proved to be very useful in oscillation theory, disconjugacy, eigenvalue problems and numerous other applications in the theory of differential and difference equations. In recent years, there are many literatures which improved and extended the classical Lyapunov inequality for Hamiltonian systems including continuous and discrete cases. Here we only mention some references [9–22]. In the recent paper [18], Guseinov and Kaymakcalan researched system (1.1) and obtained some interesting Lyapunov type inequalities and stability criteria. Theorem 1.2 [18]. Let a; b 2 Z with a 6 b 2. Assume (1.1) has a real solution (x(n), y(n)) such that x(a) = x(b) = 0 and x(n) is not identically zero on [a, b]. Then one has the following inequality b2 X
" jaðnÞj þ
b1 X
n¼a
bðnÞ
n¼a
b2 X
#1=2
cþ ðnÞ
P 2;
ð1:7Þ
n¼a
where and in the sequel
cþ ðnÞ ¼ maxfcðnÞ; 0g:
ð1:8Þ
Theorem 1.3 [18]. Suppose that
1 aðnÞ > 0;
bðnÞ > 0;
cðnÞ > 0; 8 n 2 Z
ð1:9Þ
and let a; b 2 Z with a 6 b 2. Assume (1.1) has a real solution (x(n), y(n)) such that x(n) has generalized zeros at a and b, and x(n) is not identically zero on [a, b], i.e.
xðaÞ ¼ 0 or xðaÞxða þ 1Þ < 0;
xðbÞ ¼ 0 or xðbÞxðb þ 1Þ < 0;
max jxðnÞj > 0:
a6n6b
ð1:10Þ
Then one has the following inequality b1 X
" jaðnÞj þ
n¼a
b X
bðnÞ
n¼a
b1 X
#1=2
cðnÞ
> 1:
ð1:11Þ
n¼a
When a(n), b(n) and c(n) are N-periodic functions, by applying Floquet theory and the above theorems, Guseinov and Kaymakcalan [18] also obtained the following stability criterion. Theorem 1.4 [18]. Assume that
aðn þ NÞ ¼ aðnÞ; bðn þ NÞ ¼ bðnÞ; cðn þ NÞ ¼ cðnÞ; 8 n 2 Z; bðnÞ > 0; cðnÞ > 0; bðnÞcðnÞ a2 ðnÞ P 0ð – 0Þ; 8 n 2 Z
ð1:12Þ ð1:13Þ
and N X n¼1
(" jaðnÞj þ
b0 þ
N X n¼1
# bðnÞ
N X
)1=2
cðnÞ
6 1;
ð1:14Þ
n¼1
where b0 = max16n6Nb(n). Then system (1.1) is stable. In the case a and b are usual zeros of x(n), i.e. x(a) = x(b) = 0, we remark that the bound 2 in the right side of (1.7) of Theorem 1.2 is the best possible constant in some sense, see [13]. However, when a or b is the generalized zero of x(n), the corresponding bound reduces to constant 1, see (1.11) in Theorem 1.3. In this paper, by using some simpler methods different from [18], we obtain a better Lyapunov type inequality than (1.11)
576
Q.-m. Zhang, X.H. Tang / Applied Mathematics and Computation 218 (2011) 574–582 b1 X
jaðnÞj þ
" b X
n¼a
b1 X
bðnÞ
n¼a
#1=2 þ
c ðnÞ
P 2;
ð1:15Þ
n¼a
only under the assumption
1 aðnÞ > 0;
8n 2 Z:
ð1:16Þ
Furthermore, applying the theorems obtained in this paper, the stability conditions (1.13) and (1.14) are greatly improved by the following conditions: there exists a non-negative function h(n) such that
jaðnÞj 6 hðnÞbðnÞ; n 2 ½1; N; N X
ð1:17Þ
cðnÞ h2 ðnÞbðnÞ > 0
ð1:18Þ
n¼1
and N X
jaðnÞj þ
" N X
n¼1
bðnÞ
n¼1
N X
#1=2 þ
c ðnÞ
< 2:
ð1:19Þ
n¼1
2. Lyapunov type inequalities In this section, we establish some new Lyapunov type inequalities. Theorem 2.1. Suppose that (1.16) holds and let a; b 2 Z with a 6 b 1. Assume (1.1) has a real solution (x(n), y(n)) such that (1.10) holds. Then one has the following inequality b1 X
jaðnÞj þ
" b X
n¼a
bðnÞ
n¼a
b1 X
#1=2
cþ ðnÞ
P 2:
ð2:1Þ
n¼a
Proof. It follows from (1.10) that there exist n, g 2 [0, 1) such that
ð1 nÞxðaÞ þ nxða þ 1Þ ¼ 0
ð2:2Þ
ð1 gÞxðbÞ þ gxðb þ 1Þ ¼ 0:
ð2:3Þ
and
Multiplying the first equation of (1.1) by y(n) and the second one by x(n + 1), and then adding, we get
D½xðnÞyðnÞ ¼ bðnÞy2 ðnÞ cðnÞx2 ðn þ 1Þ:
ð2:4Þ
Summing the Eq. (2.4) from a to b 1, we can obtain
xðbÞyðbÞ xðaÞyðaÞ ¼
b1 X
bðnÞy2 ðnÞ
n¼a
b1 X
cðnÞx2 ðn þ 1Þ:
ð2:5Þ
n¼a
From the first equation of (1.1), we have
½1 aðnÞxðn þ 1Þ ¼ xðnÞ þ bðnÞyðnÞ:
ð2:6Þ
Combining (2.6) with (2.2), we have
xðaÞ ¼
nbðaÞ yðaÞ: 1 ð1 nÞaðaÞ
ð2:7Þ
Similarly, it follows from (2.6) and (2.3) that
xðbÞ ¼
gbðbÞ yðbÞ: 1 ð1 gÞaðbÞ
ð2:8Þ
Substituting (2.7) and (2.8) into (2.5), we have b1 X
bðnÞy2 ðnÞ
n¼a
which implies that
b1 X n¼a
cðnÞx2 ðn þ 1Þ ¼
gbðbÞ nbðaÞ y2 ðbÞ þ y2 ðaÞ; 1 ð1 gÞaðbÞ 1 ð1 nÞaðaÞ
577
Q.-m. Zhang, X.H. Tang / Applied Mathematics and Computation 218 (2011) 574–582 b1 b1 X X ð1 nÞ½1 aðaÞ gbðbÞ bðnÞy2 ðnÞ þ cðnÞx2 ðn þ 1Þ: bðaÞy2 ðaÞ þ y2 ðbÞ ¼ 1 ð1 nÞaðaÞ 1 ð1 gÞaðbÞ n¼a n¼aþ1
ð2:9Þ
Denote that
ð1 nÞ½1 aðaÞ ~ bðaÞ ¼ bðaÞ; 1 ð1 nÞaðaÞ
~ bðbÞ ¼
g 1 ð1 gÞaðbÞ
ð2:10Þ
bðbÞ
and
~ bðnÞ ¼ bðnÞ;
a þ 1 6 n 6 b 1:
ð2:11Þ
Then we can rewrite (2.9) as b X
2 ~ bðnÞy ðnÞ ¼
b1 X
n¼a
cðnÞx2 ðn þ 1Þ:
ð2:12Þ
n¼a
On the other hand, summing the first equation of (1.1) from a to s 1 and using (2.7), we obtain s1 X
xðsÞ ¼ xðaÞ þ
aðnÞxðn þ 1Þ þ
s1 X
n¼a
¼
s1 X
bðnÞyðnÞ ¼
n¼a
aðnÞxðn þ 1Þ þ
n¼a
s1 X
~ bðnÞyðnÞ;
s1 s1 X X nbðaÞ aðnÞxðn þ 1Þ þ bðnÞyðnÞ yðaÞ þ 1 ð1 nÞaðaÞ n¼a n¼a
a þ 1 6 s 6 b:
ð2:13Þ
n¼a
Similarly, summing the first equation of (1.1) from s to b 1 and using (2.8), we have b1 X
xðsÞ ¼ xðbÞ
aðnÞxðn þ 1Þ
b1 X
n¼s
¼
b1 X
bðnÞyðnÞ ¼
n¼s
aðnÞxðn þ 1Þ
n¼s
b X
~ bðnÞyðnÞ;
b1 b1 X X gbðbÞ yðbÞ aðnÞxðn þ 1Þ bðnÞyðnÞ 1 ð1 gÞaðbÞ n¼s n¼s
a þ 1 6 s 6 b:
ð2:14Þ
n¼s
It follows from (2.13) and (2.14) that
jxðsÞj 6
s1 X
jaðnÞjjxðn þ 1Þj þ
n¼a
s1 X
~ bðnÞjyðnÞj;
aþ16s6b
~ bðnÞjyðnÞj;
a þ 1 6 s 6 b:
n¼a
and
jxðsÞj 6
b1 X
jaðnÞjjxðn þ 1Þj þ
n¼s
b X n¼s
Adding the above two inequalities, we have
2jxðsÞj 6
b1 X
b X
jaðnÞjjxðn þ 1Þj þ
n¼a
~ bðnÞjyðnÞj;
a þ 1 6 s 6 b:
ð2:15Þ
n¼a
Let jx(s⁄)j = maxa+16n6bjx(n)j. Applying the Cauchy inequality and using (2.12), we have
2jxðs Þj 6
b1 X
jaðnÞjjxðn þ 1Þj þ
n¼a
b X
~ bðnÞjyðnÞj 6 jxðs Þj
n¼a
¼ jxðs Þj
b1 X
jaðnÞj þ
n¼a
" b X
jaðnÞj þ
" b X
n¼a
~ bðnÞ
n¼a
b1 X
b1 X
n¼a
#1=2 2
cðnÞx ðn þ 1Þ
n¼a
~ bðnÞ
6 jxðs Þj
8 b1
b X
#1=2 2 ~ bðnÞy ðnÞ
n¼a
" jaðnÞj þ
b X n¼a
~ bðnÞ
#1=2 9 = c ðnÞ : ; n¼a
b1 X
þ
ð2:16Þ
Dividing the latter inequality of (2.16) by jx(s⁄)j, we obtain b1 X
" jaðnÞj þ
n¼a
b X n¼a
~ bðnÞ
b1 X
#1=2 þ
c ðnÞ
P 2:
n¼a
Since
~ bðnÞ 6 bðnÞ;
a 6 n 6 b;
then it follows from (2.17) that (2.1). h In the case x(b) = 0, i.e. g = 0, we have the following equation
ð2:17Þ
578
Q.-m. Zhang, X.H. Tang / Applied Mathematics and Computation 218 (2011) 574–582 b1 X
2 ~ bðnÞy ðnÞ ¼
b2 X
n¼a
cðnÞx2 ðn þ 1Þ
ð2:18Þ
n¼a
and inequality
2jxðsÞj 6
b2 X
jaðnÞjjxðn þ 1Þj þ
n¼a
b1 X
~ bðnÞjyðnÞj;
aþ16s6b1
ð2:19Þ
n¼a
instead of (2.12) and (2.15), respectively. Similar to the proof of (2.17), we have b2 X
jaðnÞj þ
" b1 X
n¼a
~ bðnÞ
n¼a
b2 X
#1=2 þ
c ðnÞ
P 2:
ð2:20Þ
n¼a
~ Since bðnÞ 6 bðnÞ for a 6 n 6 b, it follows that b2 X
jaðnÞj þ
" b1 X
n¼a
bðnÞ
n¼a
b2 X
#1=2
cþ ðnÞ
P 2:
ð2:21Þ
n¼a
Therefore, we have the following theorem. Theorem 2.2. Suppose that (1.16) holds and let a; b 2 Z with a 6 b 2. Assume (1.1) has a real solution (x(n), y(n)) such that x(a) = 0 or x(a)x(a + 1) < 0 and x(b) = 0 and x(n) is not identically zero on [a, b]. Then inequality (2.21) holds. h
Remark 2.3. We obtain the same Lyapunov type inequality (2.21) as (1.7) under weaker assumptions than the ones of Theorem 1.2, which improves greatly the following inequality b2 X
jaðnÞj þ
" b1 X
n¼a
bðnÞ
n¼a
b2 X
#1=2
cþ ðnÞ
>1
n¼a
obtained in [18, Theorem 1.4]. Theorem 2.4. Suppose that (1.16) holds and let a; b 2 Z with a 6 b 1. Assume (1.1) has a real solution (x(n), y(n)) such that x(a) = 0 or x(a)x(a + 1) < 0 and (x(b), y(b)) = (kx(a), ly(a)) with 0 < kl 6 l2 6 1 and x(n) is not identically zero on [a, b]. Then one has the following inequality b1 X
jaðnÞj þ
" b1 X
n¼a
bðnÞ
n¼a
b1 X
#1=2 þ
c ðnÞ
P 2:
ð2:22Þ
n¼a
Proof. It follows from the assumption x(a) = 0 or x(a)x(a + 1) < 0 that there exists n 2 [0, 1) such that (2.2) holds. Further, by the proof of Theorem 2.1, (2.4)–(2.7) hold. Since (x(b), y(b)) = (kx(a), ly(a)), then by (2.5), we have
ðkl 1ÞxðaÞyðaÞ ¼
b1 X
bðnÞy2 ðnÞ
n¼a
b1 X
cðnÞx2 ðn þ 1Þ:
ð2:23Þ
n¼a
Substituting (2.7) into (2.23), we have b1 X
bðnÞy2 ðnÞ
n¼a
b1 X
cðnÞx2 ðn þ 1Þ ¼
n¼a
ð1 klÞnbðaÞ 2 y ðaÞ; 1 ð1 nÞaðaÞ
which implies that
j1 bðaÞy2 ðaÞ þ
b1 X n¼aþ1
bðnÞy2 ðnÞ ¼
b1 X
cðnÞx2 ðn þ 1Þ;
ð2:24Þ
n¼a
where
j1 ¼
1 ð1 klÞn ð1 nÞaðaÞ : 1 ð1 nÞaðaÞ
On the other hand, summing the first equation of (1.1) from a to s 1 and using (2.7), we obtain
ð2:25Þ
Q.-m. Zhang, X.H. Tang / Applied Mathematics and Computation 218 (2011) 574–582
xðsÞ ¼ xðaÞ þ
s1 X
aðnÞxðn þ 1Þ þ
n¼a
s1 X
bðnÞyðnÞ ¼
n¼a
579
s1 s1 X X n aðnÞxðn þ 1Þ þ bðnÞyðnÞ bðaÞyðaÞ þ 1 ð1 nÞaðaÞ n¼a n¼a
s1 s1 X X ð1 nÞ½1 aðaÞ aðnÞxðn þ 1Þ þ bðnÞyðnÞ; bðaÞyðaÞ þ ¼ 1 ð1 nÞaðaÞ n¼a n¼aþ1
a þ 1 6 s 6 b:
ð2:26Þ
Similarly, summing the first equation of (1.1) from s to b 1 and using (2.7) and the fact that x(b) = kx(a), we have
xðsÞ ¼ xðbÞ
b1 X
aðnÞxðn þ 1Þ
n¼s
¼
b1 X
bðnÞyðnÞ ¼ kxðaÞ
n¼s
b1 X
aðnÞxðn þ 1Þ
n¼s
b1 X
bðnÞyðnÞ
n¼s
b1 b1 X X kn aðnÞxðn þ 1Þ bðnÞyðnÞ; bðaÞyðaÞ 1 ð1 nÞaðaÞ n¼s n¼s
a þ 1 6 s 6 b:
ð2:27Þ
It follows from (2.26) and (2.27) that
jxðsÞj 6
s1 s1 X X ð1 nÞ½1 aðaÞ jaðnÞjjxðn þ 1Þj þ bðnÞjyðnÞj; bðaÞjyðaÞj þ 1 ð1 nÞaðaÞ n¼a n¼aþ1
jxðsÞj 6
b1 b1 X X jkjn jaðnÞjjxðn þ 1Þj þ bðnÞjyðnÞj; bðaÞjyðaÞj þ 1 ð1 nÞaðaÞ n¼s n¼s
aþ16s6b
and
a þ 1 6 s 6 b:
Adding the above two inequalities, we have
2jxðsÞj 6 j2 bðaÞjyðaÞj þ
b1 X
b1 X
jaðnÞjjxðn þ 1Þj þ
n¼a
a þ 1 6 s 6 b;
bðnÞjyðnÞj;
ð2:28Þ
n¼aþ1
where
j2 ¼
1 ð1 jkjÞn ð1 nÞaðaÞ : 1 ð1 nÞaðaÞ
ð2:29Þ
Let jx(s⁄)j = maxa+16n6bjx(n)j. Applying (2.24), (2.28) and the Cauchy inequality, we have
2jxðs Þj 6 j2 bðaÞjyðaÞj þ
b1 X
jaðnÞjjxðn þ 1Þj þ
n¼a
6 jxðs Þj
b1 X
¼ jxðs Þj
" jaðnÞj þ " jaðnÞj þ
n¼a
6 jxðs Þj
b1 X
" jaðnÞj þ
n¼a
8 " b1
bðnÞjyðnÞj
n¼aþ1
n¼a b1 X
b1 X
!
b1 X j22 bðaÞ þ bðnÞ j1 n¼aþ1 b1 X
j bðaÞ þ bðnÞ j1 n¼aþ1 2 2
j1 bðaÞy2 ðaÞ þ
b1 X
!#1=2 bðnÞy2 ðnÞ
n¼aþ1
!
b1 X
#1=2 2
cðnÞx ðn þ 1Þ
n¼a
!
b1 b1 X X j22 bðaÞ þ bðnÞ cþ ðnÞx2 ðn þ 1Þ j1 n¼a n¼aþ1
!
b1 b1 X X j22 bðaÞ þ bðnÞ cþ ðnÞ j1 n¼a n¼aþ1
#1=2
#1=2 9 = : ;
ð2:30Þ
Dividing the latter inequality of (2.30) by jx(s⁄)j, we obtain b1 X
" jaðnÞj þ
n¼a
!
#1=2
b1 b1 X X j22 bðaÞ þ bðnÞ cþ ðnÞ j1 n¼a n¼aþ1
P 2:
Set d = 1 (1 n)a(a). Then d > 0. Since (1 n)[1 a(a)] P 0, it follows that n 6 d, and so
½d ð1 jkjÞn2 6 d½d ð1 klÞn: This, together with (2.25) and (2.29), implies that
j22 ¼ j1
h
1ð1jkjÞnð1nÞaðaÞ 1ð1nÞaðaÞ
i2
1ð1klÞnð1nÞaðaÞ 1ð1nÞaðaÞ
2
¼
½d ð1 jkjÞn 6 1: d½d ð1 klÞn
ð2:31Þ
580
Q.-m. Zhang, X.H. Tang / Applied Mathematics and Computation 218 (2011) 574–582
Substituting this into (2.31), we obtain (2.22). h
3. Stability criteria In this section, we discuss the stability for solutions of system (1.1) by applying the Lyapunov type inequalities obtained in the last section. To this end, we assume that system (1.1) is N-periodic, i.e. the coefficients a(n), b(n) and c(n) satisfy the periodicity conditions
aðn þ NÞ ¼ aðnÞ; bðn þ NÞ ¼ bðnÞ; cðn þ NÞ ¼ cðnÞ; 8n 2 Z:
ð3:1Þ
Definition 3.1. System (1.1) is said to be stable if all solutions are bounded on Z, unstable if all non-zero solutions are unbounded on Z, and conditionally stable if there exists a non-zero solution bounded on Z. Let u(n) = (x(n), y(n))> and
0 AðnÞ ¼ @
1
bðnÞ 1aðnÞ
1 1aðnÞ
bðnÞcðnÞ 1cðnÞ aðnÞ 1 aðnÞ 1aðnÞ
A:
ð3:2Þ
Then detA(n) = 1 for n 2 Z and system (1.1) can be written as
uðn þ 1Þ ¼ AðnÞuðnÞ:
ð3:3Þ
Let U(n) with U(0) = I2 be a fundamental matrix solution of (3.3). Then
Uðn þ 1Þ ¼ AðnÞUðnÞ;
Uðn þ NÞ ¼ UðnÞUðNÞ;
8n 2 Z:
ð3:4Þ
It follows that detU(n) = detU(0) = 1 for n 2 Z. The Floquet multipliers (real or complex) of (3.3) are the roots of
detðkI2 UðNÞÞ ¼ 0; which is equivalent to
k2 qk þ 1 ¼ 0;
ð3:5Þ
where q is the trace of matrix U(N). Let k1 and k2 are the Floquet multipliers, then we have
k1 þ k2 ¼ q;
k1 k2 ¼ 1:
It follows from the Floquet theory [23] that corresponding to each (complex) root kk there is a solution uk(n) = (xk(n), yk(n))> of (3.3) (or (1.1)) with xk(n) – 0 such that
uk ðn þ NÞ ¼ kk uk ðnÞ;
k ¼ 1; 2;
8n 2 Z:
ð3:6Þ
These are the so-called Floquet solutions of (3.3) (or (1.1)). Lemma 3.2 [18]. System (1.1) (or (3.3)) is unstable if jqj > 2 and stable if jqj < 2. Lemma 3.3. Assume that (3.1) holds, and that there exists a non-negative function h(n) such that (1.17) and (1.18) hold. If
q2 P 4, then system (1.1) has a non-zero solution (x(n), y(n)) such that x(n) has a generalized zero in [1, N]. Proof. Suppose that jqj P 2. Then one has real Floquet multipliers kk and real Floquet solutions uk(n) = (xk(n), yk(n))>, k = 1, 2. Let us consider any Floquet solution, say u1(n) = (x1(n), y1(n))>. We assert that x1(n) must have at least one generalized zero in the segment [1, N]. Otherwise, one may assume that x1(n) > 0 for n 2 [1, N] and so x1(n) > 0 for n 2 Z. Define z(n) :¼ y1(n)/x1(n). Due to (3.6), one sees that z(n) is N-periodic. From (1.1), we have
x1 ðnÞDy1 ðnÞ y1 ðnÞDx1 ðnÞ cðnÞx1 ðnÞx1 ðn þ 1Þ aðnÞ½x1 ðnÞ þ x1 ðn þ 1Þy1 ðnÞ bðnÞy21 ðnÞ ¼ x1 ðnÞx1 ðn þ 1Þ x1 ðnÞx1 ðn þ 1Þ y ðnÞ y1 ðnÞ y ðnÞ y1 ðnÞ ¼ cðnÞ aðnÞ 1 þ bðnÞ 1 x1 ðnÞ x1 ðn þ 1Þ x1 ðnÞ x1 ðn þ 1Þ y1 ðnÞ y1 ðnÞ ¼ cðnÞ aðnÞ zðnÞ þ bðnÞzðnÞ : x1 ðn þ 1Þ x1 ðn þ 1Þ
DzðnÞ ¼
Since x1(n) > 0 for all n 2 Z, then it follows from (2.6) that
ð3:7Þ
581
Q.-m. Zhang, X.H. Tang / Applied Mathematics and Computation 218 (2011) 574–582
1 þ bðnÞzðnÞ ¼ 1 þ bðnÞ
y1 ðnÞ x1 ðn þ 1Þ ¼ ½1 aðnÞ > 0; x1 ðnÞ x1 ðnÞ
ð3:8Þ
which yields
y1 ðnÞ ½1 aðnÞzðnÞ : ¼ x1 ðn þ 1Þ 1 þ bðnÞzðnÞ
ð3:9Þ
Substituting (3.9) into (3.7), we obtain
DzðnÞ ¼ cðnÞ þ
2aðnÞ þ a2 ðnÞ zðnÞ bðnÞz2 ðnÞ : 1 þ bðnÞzðnÞ
ð3:10Þ
If b(n) > 0, then it is easy to verify that
2aðnÞ þ a2 ðnÞ zðnÞ bðnÞz2 ðnÞ a2 ðnÞ 6 h2 ðnÞbðnÞ: 6 bðnÞ 1 þ bðnÞzðnÞ
ð3:11Þ
If b(n) = 0, then it follows from (1.17) that a(n) = 0, hence
2aðnÞ þ a2 ðnÞ zðnÞ bðnÞz2 ðnÞ ¼ 0 ¼ h2 ðnÞbðnÞ: 1 þ bðnÞzðnÞ
ð3:12Þ
Combining (3.11) with (3.12), we have
2aðnÞ þ a2 ðnÞ zðnÞ bðnÞz2 ðnÞ 6 h2 ðnÞbðnÞ: 1 þ bðnÞzðnÞ
ð3:13Þ
Substituting (3.13) into (3.10), we obtain
DzðnÞ 6 cðnÞ þ h2 ðnÞbðnÞ:
ð3:14Þ
Summing it from 1 to N and noticing that z(n) is N-periodic, we obtain
06
N X
cðnÞ h2 ðnÞbðnÞ
n¼1
a contradiction with condition (1.18). h Theorem 3.4. Assume that 3.1, 1.17, 1.18 and 1.19 hold. Then system (1.1) is stable. Proof. Since (1.17) and (1.18), if jqj P 2, then one has real Floquet multipliers kk and real Floquet solutions uk(n) = (xk(n), yk(n))> such that (3.6) holds for k = 1, 2. Since k1k2 = 1, then 0 < minfk21 ; k22 g 6 1. Suppose k21 6 1. Then by Lemma 3.3, system (1.1) has a non-zero solution (x1(n), y1(n)) such that x1(n) has a generalized zero in [1, N], say n1. It follows from (3.6) that n1 + N is also a generalized zero of x1(n) and
ðx1 ðn1 þ NÞ; y1 ðn1 þ NÞÞ ¼ k1 ðx1 ðn1 Þ; y1 ðn1 ÞÞ: Applying Theorem 2.4 to the solution (x1(n), y1(n)) with a = n1, b = n1 + N and k = l = k1, we get n1X þN1
" jaðnÞj þ
n1X þN1
n¼n1
bðnÞ
n1X þN1
n¼n1
#1=2
cþ ðnÞ
P 2:
ð3:15Þ
n¼n1
Next, note that for any periodic function f(n) on Z with period N, the equation n0X þN1
f ðnÞ ¼
N X
n¼n0
f ðnÞ
n¼1
holds for all n0 2 Z. It follows from (3.15) that N X n¼1
" jaðnÞj þ
N X n¼1
bðnÞ
N X
#1=2 þ
c ðnÞ
P 2;
ð3:16Þ
n¼1
which contradicts condition (1.19). Thus jqj < 2 and hence system (1.1) is stable. h Applying Theorem 3.4 to the second-order difference Eq. (1.4), we can obtain the following classical stability result.
582
Q.-m. Zhang, X.H. Tang / Applied Mathematics and Computation 218 (2011) 574–582
Corollary 3.5. Assume that
pðn þ NÞ ¼ pðnÞ > 0;
qðn þ NÞ ¼ qðnÞ;
8n 2 Z
ð3:17Þ
and
0<
N X
qðnÞ 6
n¼1
N X n¼1
qþ ðnÞ <
4 N P n¼1
:
ð3:18Þ
1 pðnÞ
Then Eq. (1.4) is stable. Acknowledgments The authors thank the referees for valuable comments and suggestions. References [1] S. Aubry, Discrete breathers: localization and transfer of energy in discrete Hamiltonian nonlinear systems, Review Article Physica D: Nonlinear Phenomena 216 (2006) 1–30. [2] M. Bohner, Linear Hamiltonian difference systems: disconjugacy and Jacobi-type conditions, J. Math. Anal. Appl 199 (1996) 804–826. [3] M. Bohner, Discrete linear Hamiltonian eigenvalue problems, Comput. Math. Appl. 36 (1998) 179–192. [4] L.H. Erbe, P.X. Yan, Disconjugacy for linear Hamiltonian difference systems, J. Math. Anal. Appl. 167 (1992) 355–367. [5] Y.M. Shi, Spectral theory of discrete linear Hamiltonian systems, J. Math. Anal. Appl 289 (2004) 554–570. [6] Y.M. Shi, Weyl–Titchmarsh theory for a class of discrete linear Hamiltonian systems, Linear Algebra Appl. 416 (2006) 452–519. [7] H.Q. Sun, Y.M. Shi, Strong limit point criteria for a class of singular discrete linear Hamiltonian systems, J. Math. Anal. Appl. 336 (2007) 224–242. [8] J.S. Yu, H.H. Bin, Z.M. Guo, Multiple periodic solutions for discrete Hamiltonian systems, Nonlinear Anal 66 (2007) 1498–1512. [9] P. Hartman, Difference equations: disconjugacy, principal solutions, Green’s functions, complete monotonicity, Trans. Amer. Math. Soc. 246 (1978) 1– 30. [10] R. Agarwal, C.D. Ahlbrandt, M. Bohner, A.C. Peterson, Discrete linear Hamiltonian systems: a survey, Dynam. Syst. Appl. 8 (1999) 307–333. [11] C.D. Ahlbrandt, A.C. Peterson, Discrete Hamiltonian Systems: Difference Equations, Continued Fractions, and Riccati Equations, Kluwer Academic, Boston, MA, 1996. [12] M. Bohner, S. Clark, J. Ridenhour, Lyapunov inequalities on time scales, J. Inequal. Appl. 7 (2002) 61–77. [13] S.S. Cheng, A discrete analogue of the inequality of Lyapunov, Hokkaido Math. J. 12 (1983) 105–112. [14] S.S. Cheng, Lyapunov inequalities for differential and difference equations, Fasc. Math. 23 (1991) 25–41. [15] S. Clark, D.B. Hinton, Discrete Lyapunov inequalities for linear Hamiltonian systems, Math. Inequal. Appl. 1 (1998) 201–209. [16] S. Clark, D.B. Hinton, Discrete Lyapunov inequalities, Dynam. Syst. Appl. 8 (1999) 369–380. [17] G. Sh. Guseinov, A. Zafer, Stability criteria for linear periodic impulsive Hamiltonian systems, J. Math. Anal. Appl. 335 (2007) 1195–1206. [18] G. Sh. Guseinov, B. Kaymakcalan, Lyapunov inequalities for discrete linear Hamiltonian systems, Comput. Math. Appl. 45 (2003) 1399–1416. [19] L.Q. Jiang, Z. Zhou, Lyapunov inequality for linear Hamiltonian systems on time scales, J. Math. Anal. Appl. 310 (2005) 579–593. [20] M.G. Krein, Foundations of the theory of k-zones of stability of canonical system of linear differential equations with periodic coefficients. In Memory of A.A. Andronov, Izdat. Acad. Nauk SSSR, Moscow, 1955, pp. 413–498 (Amer. Math. Soc. Transl. Ser. 2, 120 (1983) 1–70). [21] X. Wang, Stability criteria for linear periodic Hamiltonian systems, J. Math. Anal. Appl 367 (2010) 329–336. [22] V.A. Yakubovich, V.M. Starzhinsky, Linear Differential Equations with Periodic Coefficients, Parts I and II, Wiley, New York, 1975. [23] S.N. Elaydi, An Introduction to Difference Equations, third ed., Springer-Verlag, New York, 2004.