J. Math. Anal. Appl.

409 (2014) 851867

Contents lists available at ScienceDirect
Journal of Mathematical Analysis and
Applications
journal homepage: www.elsevier.com/locate/jmaa
Exponential stability of uniform EulerBernoulli beams with
non-collocated boundary controllers

Yun Lan Chen

, Gen Qi Xu
Department of Mathematics, Tianjin University, Tianjin, 300072, China
a r t i c l e i n f o
Article history:
Received 7 June 2012
Available online 27 July 2013
Submitted by David Russell
Keywords:
EulerBernoulli beam
Non-collocated controllers
Asymptotic analysis
Riesz basis
Exponential stability
a b s t r a c t
We study the stability of a robot systemcomposed of two EulerBernoulli beams with non-
collocated controllers. By the detailed spectral analysis, we prove that the asymptotical
spectra of the systemare distributed in the complex left-half plane and there is a sequence
of the generalized eigenfunctions that forms a Riesz basis in the energy space. Since there
exist at most finitely many spectral points of the system in the right half-plane, to obtain
the exponential stability, we showthat one can choose suitable feedback gains such that all
eigenvalues of the system are located in the left half-plane. Hence the Riesz basis property
ensures that the systemis exponentially stable. Finally we give some simulation for spectra
of the system.
2013 Elsevier Inc. All rights reserved.
1. Introduction
Inrecent years there has beenanincreasedinterest insystems composedof multiple robots since they exhibit cooperative
behaviors. Such systems are of interest for several reasons: on one hand, the tasks may be inherently too complex for a single
robot to accomplish, or performance benefits canbe gainedfromusing multiple robots; onthe other hand, building andusing
several simple robots can be easier, cheaper, more flexible and more fault-tolerant than having a single powerful robot for
each separate task. Therefore, numerous cooperation problems emerge in the engineering. Thus the research on single robot
systems was naturally extended to that on multiple-robot systems since ultimately a single robot, no matter how capable,
is spatially limited. The main questions are how to cooperate to achieve the predictable aim.
The multiple-robot systems are different from other distributed systems because of their implicit real-world
environment, they have presumably more difficulty in modeling than traditional components of distributed system
environments (i.e., computers, databases, networks). In the present paper, we consider how two robot-arms should
cooperate to be qualified in one stable movement. Our system shown as in Fig. 1 consists of two robot arms, their common
task is to grab stably the huge mass M. To complete this task, we make the exterior forces F
i
, i = 1, 2 act on two root ends
such that the other ends can clamp the mass, and we observe the velocity of the mass end so that the system completes the
performance. To model this system, we regard the mass as a tip mass and suppose that the arms are the same and uniform
with length one, whose positions are denoted by w
i
(x, t) with the root ends x = 0 and the arm motions described by the
EulerBernoulli beams. To complete the performance, we observe the rotation angles w
ix
(0, t), rotation angle velocities
w
ixt
(0, t) and the velocity of other ends w
it
(1, t). As feedback controls we assume the control forces of the form
F
i
= F
i
(w
ixt
(0, t), w
ix
(0, t), w
it
(1, t)), i = 1, 2.
The performance requirement is to make the displacement continuous at the other end x = 1, that is,
w
1
(1, t) = w
2
(1, t), t > 0,

This research was supported by the Natural Science Foundation of China (Grant NSFC-61174080, 61104130).

Corresponding author.
E-mail addresses: yunlanchen@tju.edu.cn (Y.L. Chen), gqxu@tju.edu.cn (G.Q. Xu).
0022-247X/$ see front matter 2013 Elsevier Inc. All rights reserved.
http://dx.doi.org/10.1016/j.jmaa.2013.07.048
852 Y.L. Chen, G.Q. Xu / J. Math. Anal. Appl. 409 (2014) 851867
Fig. 1. Two robot arms.
here we neglect the size of mass. Thus, the motion of the system is governed by the partial differential equations:
_

2
w
i
(x, t)
t
2
+

4
w
i
(x, t)
x
4
= 0, i = 1, 2, x (0, 1), t > 0,
M

2
w(1, t)
t
2
=
2

i=1

3
w
i
(1, t)
x
3
, t > 0,
w
1
(0, t) = w
2
(0, t) = 0, t > 0,

2
w
i
(0, t)
x
2
= F
i
(w
ixt
(0, t), w
ix
(0, t), w
it
(1, t)), i = 1, 2, t > 0,
w
1
(1, t) = w
2
(1, t) = w(1, t),
w
1
(1, t)
x
=
w
2
(1, t)
x
,
2

i=1

2
w
i
(1, t)
x
2
= 0, t > 0,
w
i
(x, 0) = w
i0
, w
it
(x, 0) = w
i1
, i = 1, 2,
where x stands for the position and t the time and M is the tip mass at x = 1.
First, we consider the steady state of the system, w
i
, namely w
i
satisfy differential equation
_

_
d
4
w
i
(x)
dx
4
= 0, x (0, 1), i = 1, 2,
w
1
(0) = w
2
(0) = 0,
d
2
w
i
(0)
dx
2
= F
i
(0, w
ix
(0), 0), i = 1, 2,
w
1
(1) = w
2
(1),
d w
1
(1)
dx
=
d w
2
(1)
dx
,
2

i=1
d
2
w
i
(1)
dx
2
= 0,
2

i=1
d
3
w
i
(1)
dx
3
= 0.
From this we can determine the control forces

F
i
= F(0, w
ix
, 0), i = 1, 2, that are constant forces.
Next we determine the forces F
i
, i = 1, 2. Let F
i
(z, u, v), i = 1, 2 be twice continuously differentiable functions.
Linearizing F
i
at fixed point

W
i
= (0, w
ix
(0), 0) yields
F
i
(w
ixt
(0, t), w
ix
(0, t), w
it
(1, t)) = F
i
(0, w
ix
(0), 0) +
F
i
z

W
i
w
ixt
(0, t)
+
F
i
u

W
i
(w
ix
(0, t) w
ix
(0)) +
F
i
v

W
i
w
it
(1, t) +o(w
i
w
i
).
Let y
i
(x, t) = w
i
(x, t) w
i
(x) and denote by

i
=
F
i
z

W
i
,
i
=
F
i
u

W
i
,
i
=
F
i
v

W
i
, i = 1, 2.
Y.L. Chen, G.Q. Xu / J. Math. Anal. Appl. 409 (2014) 851867 853
Then the behavior of the error system is determined by the following linearized equations
_

2
y
i
(x, t)
t
2
+

4
y
i
(x, t)
x
4
= 0, i = 1, 2, x (0, 1),
M
d
2
y(1, t)
dt
2
=
2

i=1

3
y
i
(1, t)
x
3
,
y
1
(0, t) = y
2
(0, t) = 0,

2
y
i
(0, t)
x
2
= u
i
(t), i = 1, 2,
y
1
(1, t) = y
2
(1, t) = y(1, t),
y
1
(1, t)
x
=
y
2
(1, t)
x
,
2

i=1

2
y
i
(1, t)
x
2
= 0,
y
i
(x, 0) = w
i0
w
i
, y
it
(x, 0) = w
i1
w
i
, i = 1, 2,
(1.1)
where
u
i
(t) =
i
y
i
(0, t)
x
+
i

2
y
i
(0, t)
xt
+
i
y
i
(1, t)
t
, i = 1, 2.
For simplification, we can take F
i
=

F
i
+ u
i
(t). In control input u
i
(t), the terms
y
i
(0,t)
x
and

2
y
i
(0,t)
xt
are collocated to the
force

F
i
, but the term
y
i
(1,t)
t
is from the tip mass end. In the performance, we ensure the position term y
1
(1, t) = y
2
(1, t)
mainly based on the term
y
i
(1,t)
t
, so
j
= 0 is necessary. Therefore the system (1.1) can be regarded as a linear system
with the non-collocated feedback control law (namely, the actuators and the sensor are located at different positions) and

i
,
i
,
i
can be regarded as the feedback gain constants.
Finally, under guaranteeing the performance, we seek for the conditions on the coefficients
i
,
i
and
i
that ensure the
system (1.1) is exponentially stable. From the description above we see that guaranteeing the performance of the system
requests the coefficients
j
= 0, j = 1, 2. Under these restrictions, the stability of the original nonlinear systemis equivalent
to that of (1.1) by determining the parameters
i
,
i
and
i
. Therefore, in the present paper, we mainly pay our attention to
determine the parameters
i
,
i
and
i
.
Non-collocated control has been widely used in engineering practice due to its more feasiblity (e.g. see, [1,4,18,11,
14,5,20,22,27,29]); however, there is few theoretical study including stability analysis on such systems from the view of
mathematical control. The first difficulty for the non-collocated control comes from the non-minimum-phase of the open-
loop form, a small increment of the feedback gain will lead to an unstable closed-loop system. The second difficulty arises
from the non-dissipativity for closed-loop form, which gives rise to difficulty in applying the traditional Lyapunov methods
or the energy multiplier methods to analyze the stability. Comparing with the huge works on the stabilization of collocated
PDEs in existed literature, the study for non-collocated PDEs is fairly scarce. To analyze the stability of the systemwith non-
collocated feedback, some authors used the finite-dimensional approximations of the observers for infinite-dimensional
systems (for analytic systems), we refer the reader to the literature [2,9,10,12,13]. There were some authors using non-
collocated feedback controllers to stabilize the system (e.g. see [17,8,7]). To remove the negative effect of non-collocated
term, however, these authors must use more complex feedback control signal. The method used by them, however, does not
fit our model. Recently our paper [3] adopt direct feedback manner to stabilize exponentially a wave system with variable
coefficients, in which the key point is to choose the suitable feedback gains. In the present paper, we shall use the idea in [3]
to determine the parameters
i
,
i
and
i
, i = 1, 2.
In what follows, we describe the method used in this paper. Consider the energy functional of the system (1.1), which is
defined as
E(t) =
1
2
2

j=1
_
1
0
[|y
jxx
(x, t)|
2
+|y
2
jt
(x, t)|
2
]dx +
1
2
My
2
t
(1, t) +
2

j=1
1
2

j
|y
jx
(0, t)|
2
.
Formally differentiating with respect to t yields
dE(t)
dt
=
2

j=1
_
1
0
[y
jxx
y
jxxt
+y
jt
y
jtt
]dx +My
t
(1, t)y
tt
(1, t) +
2

j=1

j
y
jx
(0, t)y
jxt
(0, t)
=
2

j=1
[
j
y
2
jxt
(0, t)
j
y
jt
(1, t)y
jxt
(0, t)].
854 Y.L. Chen, G.Q. Xu / J. Math. Anal. Appl. 409 (2014) 851867
Clearly, the energy of the system is not dissipative for all
j
,
j
, which means a non-minimum-phase system. Our aim is to
choose suitable
j
,
j
such that the energy of the system decays exponentially, at least decays, i.e.,
dE(t)
dt
< 0. In the present
paper, the main task is to prove that this can be done.
To analyze the stability of (1.1), we employ the asymptotic analysis technique to obtain the asymptotic frequency of the
system(the asymptotic spectrumof the systemoperator). Furthermore we prove the Riesz basis property of the eigenvectors
of this system. Note that the Riesz basis property implies the spectrum determined growth condition. Hence we can assert
the stability of the systemcan be determined via spectrumof the systemoperator. This approach has been extensively used
in the system analysis, for instance, [21,6,19,24,23] for single beam or string system and [25] for the serially connected
Timoshenko beams. Our model is different from the literature mentioned above, however, there may exist finitely many
eigenvalues of the system in the right half-plane, although its asymptotic spectra are in the left half-plane. So we shall
choose suitable parameters
i
and
i
such that all spectra are located in the left half-plane.
The rest is organized as follows. In Section 2, we discuss the well-posedness of the system (1.1). In Section 3, we carry
out a complete asymptotic analysis for the spectrum of the system operator. In Section 4, we prove the Riesz basis property
of the eigenvectors and generalized eigenvectors of the system operator. In Section 5, we discuss the exponential stability
of the system under some conditions and give some simulation for the spectrum of the system. Section 6 concludes.
2. Well-posedness of the system
In this section, we shall study the well-posedness of system (1.1). We begin with formulating the system into an
appropriate Hilbert state space.
Let H
k
(0, 1) be the usual Sobolev space. Set
Y := {y = (y
1
, y
2
) H
2
(0, 1) H
2
(0, 1)|y
1
(0) = y
2
(0) = 0, y
1
(1) = y
2
(1), y

1
(1) = y

2
(1)}
and Z = {z = (z
1
, z
2
) L
2
[0, 1] L
2
[0, 1]}. Let state space
H = Y Z C, (2.1)
equipped the inner product, for any (y, z, p), (f , g, h) H, via
(y, z, p), (f , g, h)
H
=
2

j=1
_
1
0
_
y

j
(x)f

j
(x) +z
j
(x)g
j
(x)
_
dx +
2

j=1

j
y

j
(0)f

j
(0) +Mph,
by which the induced norm is
(y, v, p)
2
=
2

i=1
_
1
0
[|y

i
(x)|
2
+|v
i
(x)|
2
]dx +
2

i=1

i
|y

i
(0)|
2
+M|p|
2
. (2.2)
Obviously, (H, ) is a Hilbert space.
Define the operator Ain H by
A(y, v, p) =
_
v, y
(4)
,
1
M
2

j=1
y

j
(1)
_
(2.3)
with domain
D(A) =
_

_
(y, v, p) Y [H
4
(0, 1)]
2
Y C,
y

j
(0) =
j
y

j
(0) +
j
v

j
(0) +
j
v
j
(1), j = 1, 2,
p = v
1
(1) = v
2
(1),
2

j=1
y

j
(1) = 0,
_

_
. (2.4)
With the operator Aat hand, we can write system (1.1) into an evolutionary equation in H
_
dY(t)
dt
= AY(t), t > 0,
Y(0) = Y
0
(2.5)
where Y(t) = (y(, t),
y
t
(, t),
y
t
(1, t)) and Y
0
= (U
0
, U
1
, U
2
) H is the initial data given.
To discuss the well-posedness of system (1.1), we need the following definition of dissipative operator.
Let X be a Banach space and X

be its dual space, and denote by x

, x the value of x

at x X. For every x X, the

duality set of x, F(x) X

, is defined as
F(x) = {x

|x

, x = x
2
= x

2
}.
Definition 2.1 ([15]). A linear operator Ais dissipative if for every x D(A), there is a x

F(x) such that Ax, x

0.
Y.L. Chen, G.Q. Xu / J. Math. Anal. Appl. 409 (2014) 851867 855
Lemma 2.1. Let H and Abe defined by (2.1), (2.3) and (2.4) respectively. Then Ais a closed and densely defined linear operator
in H. If
j

j

j
, j = 1, 2, then A(
1
M

2
j=1

j

j
)I is a dissipative operator in H.
Proof. It is easy to check that Ais a densely defined and closed linear operator. Here we only prove A(
1
M

2
j=1

j

j
)I is
dissipative.
For any F = (y, v, p) D(A), a directly calculation gives
AF, F
H
=
2

i=1
_
1
0
_
v

i
(x)y

i
(x) y
(4)
i
(x)v
i
(x)
_
dx +
2

i=1
y

i
(1)p +
2

i=1

i
v

i
(0)y

i
(0)
=
2

i=1
_
v

i
(0)y

i
(0) +
i
v

i
(0)y

i
(0)
_
+
2

i=1
__
1
0
v

i
(x)y

i
(x)dx
_
1
0
y

i
(x)v

i
(x)dx
_
.
So we have
2AF, F
H
=
2

i=1

_
v

i
(0)y

i
(0) +v

i
(0)y

i
(0)
_
+
2

i=1

i
_
v

i
y

i
(0) +y

i
(0)v

i
(0)
_
.
Using the connective and boundary conditions, we get that
AF, F
H
=
2

i=1
_

i
|v

i
(0)|
2
+
i
(v
i
(1)v

i
(0))
_

2

i=1

i
|v

i
(0)|
2
+
2

i=1

i
_

i
|v
i
(1)|
2
+
1

i
|v

i
(0)|
2
_

i=1
__

i
+

i

i
_
|v

i
(0)|
2
+

i

i
M
M|p|
2
_

i=1
__

i
+

i

i
_
|v

i
(0)|
2
+

i

i
M
F, F
H
_
where
i
> 0. So it holds that

__
A
1
M
2

i=1

i
I
_
F, F
_
H

i=1
_

i
+

i

i
_
|v

i
(0)|
2
0
if
i

i

i
, i = 1, 2. The desired result follows.
Lemma 2.2. Let H and Abe defined as before. Then 0 (A) and A
1
is compact on H, and hence the spectrum (A) of A
consists of all isolated eigenvalues of finite algebraic multiplicities. In particular, all eigenvalues appear in conjugate pairs on the
complex plane.
Proof. For any F = (f , g, h) H, we consider the solvability of equation
AY = F, Y = (y, v, p) D(A),
that is, y, v and p satisfy the equations
v(x) = f (x), (2.6)
y
(4)
i
(x) = g
i
(x), i = 1, 2, (2.7)
1
M
2

i=1
y

i
(1) = h, (2.8)
and the boundary conditions
_

_
y
1
(0) = y
2
(0) = 0, y
1
(1) = y
2
(1),
y

i
(0) =
i
y

i
(0) +
i
v

i
(0) +
i
v
i
(1), i = 1, 2,
p = v
1
(1) = v
2
(1), y

1
(1) = y

2
(1),
2

i=1
y

i
(1) = 0.
(2.9)
856 Y.L. Chen, G.Q. Xu / J. Math. Anal. Appl. 409 (2014) 851867
Solving the differential equation (2.7) yields
y
i
(x) = y
i
(0) +xy

i
(0) +
x
2
2
y

i
(0) +
x
3
3!
y

i
(0)
_
x
0
(x r)
3
3!
g
i
(r)dr. (2.10)
For simplification, we denote
G
i
(x) =
_
x
0
(x r)
3
3!
g
i
(r)dr, i = 1, 2,
and
k
1
= G
1
(x) G
2
(x)|
x=1
, k
2
= G
1x
(x) G
2x
(x)|
x=1
,
k
3
=
1
G

1
(x) G

1
(x)|
x=0

1
f
1x
(0)
1
f
1
(1),
k
4
=
2
G

2
(x) G

2
(x)|
x=0

2
f
2x
(0)
2
f
2
(1),
k
5
= G

1
(x) +G

2
(x)|
x=1
, k
6
= G

1
(x) +G

2
(x)|
x=1
+Mh.
Set
_
X = [y

1
(0), y

1
(0), y

1
(0), y

2
(0), y

2
(0), y

2
(0)]
T
,
K = [k
1
, . . . , k
6
]
T
.
(2.11)
Substituting (2.10) into the boundary conditions (2.9) leads to the following algebraic equations
CX = K (2.12)
where the coefficient matrix is
C =
_
_
_
_
_
_
_
_
_
_
_
_
1
1
2
1
3!
1
1
2

1
3!
1 1
1
2
1 1
1
2

1
1 0 0 0 0
0 0 0
2
1 0
0 1 1 0 1 1
0 0 1 0 0 1
_

_
.
A direct calculation shows det[C] =
1
2

1
+ (1 +

1

2
) = 0. So Eqs. (2.12) have a unique solution X and hence we can
determine uniquely functions y
i
(x) H
4
(0, 1), i = 1, 2. Clearly, (y, v, p) = (y, f , f
1
(1)) D(A) and A(y, v, p) = (f , g, h).
The inverse operator theorem asserts that A
1
is bounded on H, which indicates that 0 (A). In addition, the fact that
D(A) Y [H
4
(0, 1)]
2
Y C H shows that A
1
is compact due to Sobolevs Embedding theorem. The spectral theory
of linear compact operator shows that (A) consists of all isolated eigenvalues of finite algebraic multiplicities. Note that
Ais a real linear operator on H, so its spectrum distributes symmetrically with respect to the real axis. This completes the
proof.
Thanks to the above lemmas and LumerPhillips theorem (see, [15]), we have the following result.
Theorem 2.1. Let A and H be defined as before. Then A generates a C
0
semigroup T(t) on H. Hence, the system (1.1) is well-
posed, that is, for any Y
0
H, the system (1.1) has a unique solution Y(t) = T(t)Y
0
.
3. Asymptotic eigenvalue problem
To investigate the properties of the semigroup T(t) generated by A, we need to find out some spectral properties of
A. We know from Lemma 2.2 that the spectrum of A consists of all isolated eigenvalues, so we need only to discuss the
eigenvalue of A and its asymptotical distribution. In this section, we shall calculate the asymptotic values of eigenvalues
of A.
Let C, we consider the existence of a nonzero solution of the equation
A(y, v, p) = (y, v, p), (y, v, p) D(A).
Y.L. Chen, G.Q. Xu / J. Math. Anal. Appl. 409 (2014) 851867 857
This is equivalent to v = y, p = v
1
(1) = v
2
(1) = y(1) and y
i
satisfies the following boundary eigenvalue problem
_

2
y
i
(x) +y
(4)
i
(x) = 0,
y
1
(0) = y
2
(0) = 0,
y

i
(0) (
i
+
i
)y

i
(0)
i
y
i
(1) = 0,
y
1
(1) = y
2
(1) = y(1), y

1
(1) = y

2
(1),
2

i=1
y

i
(1) = 0,
2

i=1
y

i
(1) M
2
y(1) = 0.
(3.1)
To solve the above equations, let =
2
, we rewrite above equations into the equations about parameter as follows:
_

4
y
i
(x) +y
(4)
i
(x) = 0,
y
1
(0) = y
2
(0) = 0,
y

i
(0) (
i
+
2

i
)y

i
(0)
2

i
y
i
(1) = 0, i = 1, 2,
y
1
(1) = y
2
(1) = y(1), y

1
(1) = y

2
(1),
2

i=1
y

i
(1) = 0,
2

i=1
y

i
(1)
1
2
M
4
2

i=1
y
i
(1) = 0.
(3.2)
Here we use the equality y(1) = y
1
(1) = y
2
(1) =
1
2
(y
1
(1)+y
2
(1)). Obviously, the differential equation has general solution
y
i
(x) =
4

j=1
a
i,j
e

j
x
where all
j
are the distinct root of equation
4
= 1. Substituting these expressions into the boundary conditions in (3.2)
leads to
_

_
a
11
+a
12
+a
13
+a
14
= 0,
a
21
+a
22
+a
23
+a
24
= 0,
c
11
()a
11
+c
12
()a
12
+c
13
()a
13
+c
14
()a
14
= 0,
c
21
()a
21
+c
22
()a
22
+c
23
()a
23
+c
24
()a
24
= 0,
e

(a
11
a
21
) +e

(a
12
a
22
) +e

(a
13
a
23
) +e

(a
14
a
24
) = 0,

1
e

(a
11
a
21
) +
2
e

(a
12
a
22
) +
3
e

(a
13
a
23
) +
4
e

(a
14
a
24
) = 0,

2
1
e

(a
11
+a
21
) +
2
2
e

(a
12
+a
22
) +
2
3
e

(a
13
+a
23
) +
2
4
e

(a
14
+a
24
) = 0,
_

3
1

M
2

_
e

(a
11
+a
21
) +
_

3
2

M
2

_
e

(a
12
+a
22
) +
_

3
3

M
2

_
e

(a
13
+a
23
)
+
_

3
4

M
2

_
e

(a
14
+a
24
) = 0
(3.3)
where
c
ij
() = (
j
)
2

j

i
e

.
Denote by d
j
() =
3
j

M
2
and
D() = det
_
_
_
_
_
_
_
_
_
1 1 1 1 0 0 0 0
0 0 0 0 1 1 1 1
c
11
() c
12
() c
13
() c
14
() 0 0 0 0
0 0 0 0 c
21
() c
22
() c
23
() c
24
()
e

1
e

2
e

3
e

4
e

1
e

2
e

3
e

4
e

2
1
e

2
2
e

2
3
e

2
4
e

2
1
e

2
2
e

2
3
e

2
4
e

d
1
()e

d
2
()e

d
3
()e

d
4
()e

d
1
()e

d
2
()e

d
3
()e

d
4
()e

_
. (3.4)
Then we have the following result.
858 Y.L. Chen, G.Q. Xu / J. Math. Anal. Appl. 409 (2014) 851867
Lemma 3.1. Let A be defined as before and D() be defined as (3.4), then =
2
C is an eigenvalue of A if and only if
D() = 0, i.e.,
(A) = { =
2
|D() = 0, C}.
We are now in a position to determine the zeros of D(). Due to the symmetry of the spectrum of Awith respect to the
real axis, we only need to discuss the case that arg (0, ) (complex up-half-plane), or equivalently arg (0,

2
).
First, we consider the cases that arg (0,

4
) and arg (
3
4
, ).
When arg (0,

4
), we have arg (0,

8
). We order
j
as follows

1
= e
3
4
i
,
2
= e
5
4
i
,
3
= e

4
i
,
4
= e
7
4
i
so that
(
1
) (
2
) < 0 < (
3
) (
4
), arg
_
0,

8
_
.
When arg (
3
4
, ), we have arg (
3
8
,

2
). Taking

1
= e
3
4
i
,
2
= e

4
i
,
3
= e
5
4
i
,
4
= e
7
4
i
,
we have
(
1
) (
2
) < 0 < (
3
) (
4
), arg
_
3
8
,

2
_
.
In both cases, we always have max

(
2
) 0 min

(
3
). Therefore when || with arg (0,

8
)
(
3
8
,

2
), we find out
lim
||
D()

3
e
2(
3
+
4
)
= det

1 1 0 0 0 0 0 0
0 0 0 0 1 1 0 0

1

2

1
0 0 0 0 0 0
0 0 0 0
1

2

2

2
0 0
0 0 1 1 0 0 1 1
0 0
3

4
0 0
3

4
0 0
2
3

2
4
0 0
2
3

2
4
0 0
M
2

M
2
0 0
M
2

M
2

= 0.
So there is at most finitely many eigenvalues of Ain the region { C| arg (0,

4
) (
3
4
, )}.
For region S = { C| arg (

4
,
3
4
)}, corresponding -plane domain is arg (

8
,
3
8
), we range
j
as follows

1
= e
3
4
i
,
2
= e
5
4
i
,
3
= e

4
i
,
4
= e
7
4
i
, for arg
_

8
,

4
_
and

1
= e
3
4
i
,
2
= e

4
i
,
3
= e
5
4
i
,
4
= e
7
4
i
, for arg
_

4
,
3
8
_
,
so that
(
1
) (
2
) < 0 < (
3
) (
4
), arg
_

8
,

8
_
.
Note that, in that case, it always holds that
2
=
3
and
1
=
4
. Thus,
D() = det

1 1 1 1 0 0 0 0
0 0 0 0 1 1 1 1
c
11
() c
12
() c
13
() c
14
() 0 0 0 0
0 0 0 0 c
21
() c
22
() c
23
() c
24
()
e

1
e

2
e

3
e

4
e

1
e

2
e

3
e

4
e

2
1
e

2
2
e

2
3
e

2
4
e

2
1
e

2
2
e

2
3
e

2
4
e

d
1
()e

d
2
()e

d
3
()e

d
4
()e

d
1
()e

d
2
()e

d
3
()e

d
4
()e

= det

1 1 1 1 0 0 0 0
0 0 0 0 1 1 1 1
c
11
() c
12
() c
13
() c
14
() 0 0 0 0
0 0 0 0 c
21
() c
22
() c
23
() c
24
()
e

1
e

2
e

2
e

1
e

1
e

2
e

2
e

1
e

2
1
e

2
2
e

2
2
e

2
1
e

2
1
e

2
2
e

2
2
e

2
1
e

d
1
()e

d
2
()e

d
3
()e

d
4
()e

d
1
()e

d
2
()e

d
3
()e

d
4
()e

.
Y.L. Chen, G.Q. Xu / J. Math. Anal. Appl. 409 (2014) 851867 859
A direct but complicated calculation gives
D() = 2i
3
e
2
4
_
_
4

2
1

2
M
_
[ie
2
2
+e
2
2
+i +1] +
8i
1

2
8i
1
M 8i
2
M

e
2
2
+
8i
1

2
+8i
1
M +8i
2
M

e
2
2

16i
1

2
+8M(
1
+
2
)

+O(
2
) +O
_
e
c|
1
|
_
_
, (3.5)
where c is some positive constant. Therefore, the asymptotic zeros of D() in the region arg (

8
,
3
8
) are determined by
_
4

2
1

2
M
_
[ie
2
2
+e
2
2
+i +1] +
8i
1

2
8i
1
M 8i
2
M

e
2
2
+
8i
1

2
+8i
1
M +8i
2
M

e
2
2

16i
1

2
+8M(
1
+
2
)

= 0. (3.6)
In the next theorem, we shall give the asymptotic expression of eigenvalues for A.
Theorem 3.1. Let D() be defined as (3.4) in the sector S with =
2
. Then the zeros of D() in S have two branches
1,n
and

2,n
, they have asymptotic expansion

1,n
=
_
n +
1
2
_
i

2
+O
_
1
n
_
=
m
1
i

2
+O
_
1
n
_
and

2,n
=
_
n +
1
4
_
i

2
+O
_
1
n
_
=
m
2
i

2
+O
_
1
n
_
,
where
2
= e
i
4
. Hence all eigenvalues of Acan be written as {
j,n
,
j,n
|
j,n
=
2
j,n
, j = 1, 2}, they have asymptotic expressions

1,n
=
2
1,n
=

1
+
2

2
+m
2
1

2
i +O
_
1
n
_
, m
1
= n +
1
2
,

2,n
=
2
2,n
=

1
+
2

2
+
_
m
2
2

2
+
2
M
_
i +O
_
1
n
_
, m
2
= n +
1
4
.
Proof. Let D() = 0, =
2
S. According to (3.5) its asymptotic zeros are determined by (3.6), or equivalently,
_
4

2
1

2
M
_
[ie
2
2
+e
2
2
+i +1] +O
_
1

_
= 0, arg
_

8
,
3
8
_
.
Solving function equation ie
2
2
+e
2
2
+i +1 = 0 yields

1,n
=
_
n +
1
2
_
i

2
=
m
1
i

2
,
2,n
=
_
n +
1
4
_
i

2
=
m
2
i

2
.
Let
j,n
=
j,n
+
j,n
, j = 1, 2 be the zeros of (3.6). For j = 1, substituting it into (3.6) yields
4

2M
1

2
[ie
2
2

1,n
e
2
2

1,n
+(i +1)]

2
(8M
1
i 8M
2
i 8
1

2
i)
m
1
i +
2

1,n
e
2
2

1,n

2
(8M
1
i +8M
2
i 8
1

2
i)
m
1
i +
2

1,n
e
2
2

1,n
+

2
[16
1

2
i 8M(
1
+
2
)]
m
1
i +
2

1,n
= 0.
Expanding the exponential function leads to
8

2M
1

2
[(i 1)
2

1,n
+O(
1,n
)
2
]

2
(8M
1
i 8M
2
i 8
1

2
i)
m
1
+
2

1,n
(1 +2
2

1,n
+O(
2
1,n
))

2
(8M
1
i +8M
2
i 8
1

2
i)
m
1
i +
2

1,n
(1 2
2

1,n
+O(
2
1,n
)) +

2
[16
1

2
i 8M(
1
+
2
)]
m
1
i +
2

1,n
= 0
which implies

1,n
=

1
+
2

2(i 1)
1

2
m
1
i
+O
_
1
n
2
_
.
860 Y.L. Chen, G.Q. Xu / J. Math. Anal. Appl. 409 (2014) 851867
Hence

1,n
=
1,n
+
1,n
=
m
1
i

2
+

1
+
2

2(i 1)
1

2
m
1
i
+O
_
1
n
2
_
.
Similarly, for j = 2, substituting
2,n
into (3.6) yields
4

2M
1

2
_
ie
2
_
n+
1
4
_
i+2
2

2,n
+e

_
2
_
n+
1
4
_
i+2
2

2,n
_
+(i +1)
_
+

2
(8M
1
i 8M
2
i 8
1

2
i)
m
2
i +
2

2,n
e
2
_
n+
1
4
_
i+2
2

2,n
+

2
(8M
1
i +8M
2
i 8
1

2
i)
m
2
i +
2

2,n
e

_
2
_
n+
1
4
_
i+2
2

2,n
_
+

2
[16
1

2
i 8M(
1
+
2
)]
m
2
i +
2

2,n
= 0.
By simplification, we obtain
8

2M
1

2
(1 i)
2

2,n
+

2
i(8M
1
i 8M
2
i 8
1

2
i)
m
2
i +
2

2,n
+

2
(i)(8
1

2
i +8M
1
i +8M
2
i)
m
2
i +
2

2,n
+

2
[16
1

2
i 8M(
1
+
2
)]
m
2
i +
2

2,n
+O(
2
2,n
) = 0,
which implies

2,n
=
2
1

2
i M(
1
+
2
)

2(1 i)M
1

2
m
2
i
+O
_
1
n
2
_
.
So we have

2,n
=
2,n
+
2,n
=
m
2
i

2
+
2
1

2
i M(
1
+
2
)

2M(1 i)
1

2
m
2
i
+O
_
1
n
2
_
.
Therefore, we have

1,n
=
2
1,n
=

1
+
2

2
+m
2
1

2
i +O
_
1
n
_
and

2,n
=
2
2,n
=

1
+
2

2
+
_
m
2
2

2
+
2
M
_
i +O
_
1
n
_
.
This completes the proof.
4. Completeness of (generalized) eigenvectors of Aand Riesz basis property
In this section, we shall discuss the completeness of (generalized) eigenvectors of Aand its Riesz basis property. First we
establish the completeness of (generalized) eigenvectors of A. For this purpose, we need the following lemma.
Lemma 4.1. Suppose H is defined by (2.1). Let A
0
be a new operator in H defined as
A
0
(y, v, p) =
_
v, y
(4)
,
1
M
2

i=1
y

i
(1)
_
, (4.1)
with domain
D(A
0
) =
_

_
(y, v, p) Y [H
4
(0, 1)]
2
Y C :
y

i
(0)
i
y

i
(0) = 0, i = 1, 2,
p = v
1
(1) = v
2
(1),
2

i=1
y

i
(1) = 0
_

_
. (4.2)
Then A
0
is a skew self-adjoint operator in H, i.e., A

0
= A
0
.
The proof is a direct verification, we omit the details.
Y.L. Chen, G.Q. Xu / J. Math. Anal. Appl. 409 (2014) 851867 861
Theorem 4.1. The (generalized) eigenvectors of Ais complete in H, i.e.,
H = Sp(A) = span{E(
k
, A)H,
k
(A)}
where E(
k
, A) is the Riesz projector corresponding to
k
.
Proof. Let H and Abe defined as before and A
0
be defined as (4.1) and (4.2). To prove Sp(A) = H, let U
0
= (w
0
, z
0
, v
0
)
H and (w
0
, z
0
, v
0
)Sp(A), we shall show U
0
= 0.
Since U
0
Sp(A), R

(, A)U
0
can extend to an entire function on the complex plane C. Thus for any F = (f
1
, f
2
, f
3
) H,
the function defined by
G() = F, R

(, A)U
0

H
, C
also is an entire function. Since A is the generator of semigroup T(t), we have lim
+
|G()| = 0. In addition, for
(A),
G() = (R(, A)F, U
0
)
H
.
Now for (A) (A
0
), we write
Y
1
= R(, A)F, Y
2
= R(, A
0
)F, (4.3)
where Y
1
D(A), Y
2
= (z, u, q) D(A
0
). Let Y = Y
1
Y
2
= (y, v, p), then R(, A)F = R(, A
0
)F +Y and Y = (y, v, p)
satisfy the following equations
_

_
v(x) = y(x), p = v
1
(1) = v
2
(1),

2
y
i
(x) +y
(4)
i
(x) = 0, i = 1, 2,
y
1
(0) = y
2
(0) = 0,
y

i
(0) (
i
+
i
)y

i
(0)
i
y
i
(1) =
i
u

i
(0) +
i
u
i
(1), i = 1, 2,
y
1
(1) = y
2
(1), y

1
(1) = y

2
(1),
2

i=1
y

i
(1) = 0,
2

i=1
y

i
(1) M
2
y
1
(1) = 0.
(4.4)
In what follows, we shall estimate the norm of Y for real , i.e.,
Y
2
= (y, v, p)
2
=
2

i=1
_
1
0
[|y

i
(x)|
2
+|v
i
(x)|
2
]dx +
2

i=1

i
|y

i
(0)|
2
+M|p|
2
=
2

i=1
_
1
0
[|y

i
(x)|
2
+|y
i
(x)|
2
]dx +
2

i=1

i
|y

i
(0)|
2
+M|y
1
(1)|
2
.
Since
_
1
0
|y

i
(x)|
2
dx =
_
1
0
y

i
(x)y

i
(x)dx = y

i
(x)y

i
(x) |
1
0
y

i
(x)y
i
(x) |
1
0
+
_
1
0
y
(4)
i
(x)y
i
(x)dx
= y

i
(1)y

i
(1) y

i
(0)y

i
(0) y

i
(1)y
i
(1)
2
_
1
0
|y
i
(x)|
2
dx,
for real R, we have
2

i=1
_
1
0
|y

i
(x)|
2
dx +
2
2

i=1
_
1
0
|y
i
(x)|
2
dx +
2

i=1

i
|y

i
(0)|
2
+M|y
1
(1)|
2
=
2

i=1
y

i
(1)y

i
(1)
2

i=1
(y

i
(0)
i
y

i
(0))y

i
(0)
2

i=1
y

i
(1)y
i
(1) +
2
My
2
1
(1)
=
2

i=1
(
i
y

i
(0) +
i
u

i
(0) +
i
y
i
(1) +
i
u
i
(1))y

i
(0)
=
2

i=1

i
|y

i
(0)|
2

i=1

i
u

i
(0)y

i
(0)
2

i=1

i
y
i
(1)y

i
(0)
2

i=1

i
u
i
(1)y

i
(0).
To estimate Y
2
, we split it into three steps:
862 Y.L. Chen, G.Q. Xu / J. Math. Anal. Appl. 409 (2014) 851867
Step 1. There exists positive constant C such that
2

i=1

i
|y

i
(0)|
2

C
||
2

i=1
(|u

i
(0)|
2
+|u
i
(1)|
2
)
and
2

i=1

i
|y
i
(1)|
2

C
||
2

i=1
[|u

i
(0)|
2
+|u
i
(1)|
2
],
where C = max{C
1
, C
2
, C
3
}.
According to the analysis in Section 3, we can set
y
i
(x) =
4

j=1
a
i,j
e

j
x
where
j
are the distinct root of equation
4
= 1. Eq. (4.4) becomes
HF = E (4.5)
where
F = [a
11
, a
12
, a
13
, a
14
, a
21
, a
22
, a
23
, a
24
]

,
y

i
(0)

= a
i1

1
+a
i2

2
+a
i3

3
+a
i4

4
, i = 1, 2,
E =
_
0, 0,

1
u

1
(0) +
1
u
1
(1)

3
,

2
u

2
(0) +
2
u
2
(1)

3
, 0, 0, 0, 0
_

where H = (h
ij
)
88
is the same as D()
ij
,
H
1
=
1
det H
_
_
_
_
_
_
_
_
_
_
A
11
A
12
A
18
A
21
A
22
A
28
A
31
A
32
A
38
A
41
A
42
A
48
A
51
A
52
A
58
A
61
A
62
A
68
A
71
A
72
A
78
A
81
A
82
A
88
_
_
_
_
_
_
_
_
_
_
,
where {A
ij
|i, j = 1, . . . , 8} is the algebraic complement of matrix H. Directly calculating we find out
det H = D()
O
_
2ie
2
4
_
4

2
1

2
M
_
[ie
2
2
+e
2
2
+i +1]
_
,
and
max
i,j=1,...,8
|A
ij
| O
_
e
2
4
e

2
[4
2
M(1 i)e

2
+
_

2 4
_
i
2
Me

2
]
_
.
Hence
H
1
max
i,j=1,...,8
|A
ij
|
| det H|
,
where is a positive constant. We have F H
1
E. Thus we obtain
2

i=1

i
|y

i
(0)|
2

C
1
||
2

i=1
(|u

i
(0)|
2
+|u
i
(1)|
2
).
Since

i
(0)

= |a
ij

2
j
| C
2
1

3
2

i=1
(|u

i
(0)|
2
+|u
i
(1)|
2
)
and
y

i
(0) (
i
+
i
)y

i
(0)
i
y
i
(1) =
i
u

i
(0) +
i
u
i
(1), i = 1, 2,
Y.L. Chen, G.Q. Xu / J. Math. Anal. Appl. 409 (2014) 851867 863
so we can obtain
2

i=1

i
|y
i
(1)|
2

C
3
||
2

i=1
[|u

i
(0)|
2
+|u
i
(1)|
2
].
Step 2. There exists positive constant C
4
such that
2

i=1

i
|u

i
(0)|
2

C
4
|
2
| ||
F
2
, |u
i
(1)|
C
4
||
F, i = 1, 2. (A
0
) R.
Since A
0
is a skew self-adjoint operator in H, we have
R(, A
0
)
1
||
, (A
0
),
Step 3. There exists positive constant C
5
such that
Y
C
5
||
F, R.
In fact, for R (A) (A
0
),
Y
2
=
2

i=1
_
1
0
|y

i
(x)|
2
dx +
2
2

i=1
_
1
0
|y
i
(x)|
2
dx +
2

i=1

i
|y

i
(0)|
2
+M|y
1
(1)|
2
=
2

i=1

i
|y

i
(0)|
2

i=1

i
u

i
(0)y

i
(0)
2

i=1

i
y
i
(1)y

i
(0)
2

i=1

i
u
i
(1)y

i
(0)
||
2

i=1

i
|y

i
(0)|
2
+
1
2
_
2

i=1

i
|u

i
(0)|
2
+
2

i=1

i
|y

i
(0)|
2
_
+
||
2
_
2

i=1

i
|y
i
(1)|
2
+
2

i=1

i
|y

i
(0)|
2
_
+
1
2
_
2

i=1

i
|u
i
(1)|
2
+
2

i=1

i
|y

i
(0)|
2
_

_
2C +
C
2||
+
1
2
+C +

i
||
_
2

i=1

i
|u

i
(0)|
2
+
i
|u
i
(1)|
2
by Step 1

_
2C +
C
2||
+
1
2
C +

i
||
_
_
C
4
||
3
+
C
2
4
||
2
2

i=1

i
_
F
2
by Step 2

C
2
5
||
2
F
2
where C
2
5
= sup
||>
_
2C +
C
2||
+
1
2
C +

i
||
_ _
C
4
||
+C
2
4

2
i=1

i
_
.
Since
Y
2
= R(, A
0
)F
1
||
F,
we find out
Y
1
= R(, A)F (R(, A
0
)F +Y)

_
C
5
||
+
1
||
_
F, (A) (A
0
) R

.
Therefore lim

R(, A)F = 0.
Note that G() is an entire function of finite exponential type, G() is uniformly bounded along the line = > 0.
The Phragmn-Linderlf theorem (cf. [28]) asserts that
|G()| M, C.
The Liouville theorem says that G() 0. Observe that G() = (F, R

(, A)U
0
)
H
for any F H, so it must be
R

(, A)U
0
= 0 which implies U
0
= 0. Therefore Sp(A) = H.
864 Y.L. Chen, G.Q. Xu / J. Math. Anal. Appl. 409 (2014) 851867
In what follows, we shall study the Riesz basis generation of the (generalized) eigenvectors of A, we need the following
result that comes from [26].
Lemma 4.2. Let H be a separable Hilbert space, and Abe the generator of a C
0
semigroup T(t) on H. Suppose that the following
conditions are satisfied:
(1) (A) =
1
(A)
2
(A), where
2
(A) = {
k
}

k=1
consists of isolated eigenvalues of Awith finite multiplicity;
(2) sup
k1
m
a
(
k
) < , where m
a
(
k
) = dimE(
k
, A)H and E(
k
, A) is the Riesz projector associated with
k
;
(3) There is a constant such that
sup{|
1
(A)} inf{|
2
(A)}
and
inf
n=m
|
n

m
| > 0.
Then the following assertions are true.
(i) There exist two T(t)-invariant closed subspaces H
1
, H
2
with the property that (A|
H
1
) =
1
(A), (A|
H
2
) =

2
(A), E(
k
, A)H
2
forms a subspace Riesz basis for H
2
and
H = H
1

H
2
.
(ii) If sup
k1
E(
k
, A) < , then
D(A) H
1
H
2
H. (4.6)
(iii) H has the decomposition
H = H
1
H
2
(topological direct sum),
if and only if
sup
n1
_
_
_
_
_
n

k=1
E(
k
, A)
_
_
_
_
_
< . (4.7)
Applying Theorem 4.1 and Lemma 4.2 to our problem, we have the following result.
Theorem 4.2. Let H and Abe defined as (2.1) and (2.3), respectively. Then there is a sequence of eigenvectors and generalized
eigenvectors of A that forms a Riesz basis for H. In particular, the system associated with A satisfies the spectrum determined
growth condition.
Proof. Set
1
(A) = {},
2
(A) = (A). Lemma 2.2 shows that A generates a C
0
semigroup and the condition (1) in
Lemma 4.2 holds. Theorem 3.1 shows that, for sufficient large n,
j,n
are separable and m
a
(
j,n
) = 1, which means that the
conditions (2) and (3) in Lemma 4.2 are fulfilled. Therefore, according to Lemma 4.2, there is a sequence of eigenvectors and
generalized eigenvectors of Athat forms a Riesz basis for H
2
due to m
a
(
j,n
) = 1 for sufficient large n. Theorem 4.1 shows
that the eigenvectors and generalized eigenvectors are complete in H, that is, H
2
= H. Therefore the sequence is also a
Riesz basis for H.
Note that when n is sufficient large, all
j,n
are simple eigenvalues of A. There are probably finitely many eigenvalues
of multiplicity two. The Riesz basis property of the eigenvectors and generalized eigenvectors of Aensures that the system
associated with Asatisfies the spectrum determined growth condition. The desired result follows.
5. Stability analysis of the cooperation system
5.1. Stability analysis
In this section, we shall discuss the stability of the system (1.1). It is well known that if (
1
,
2
) = 0, then the system is
exponentially stable. However, the performance of the nonlinear system requires
j
= 0. So the requirement is inherited to
the system (1.1). So we only need to consider the cases of (
1
,
2
) > 0.
We know from Theorem 4.2 that the growth rate of the system is determined via the maximal real part of spectrum of
A. We denote by
s(, ) = sup
(A)
, (5.1)
where = (
1
,
2
), = (
1
,
2
). Then we have the following result.
Theorem 5.1. Let H and A be defined as (2.1) and (2.3), respectively. If we can choose the gain coefficients and such that
s(, ) < 0, then the closed loop system (2.1) is exponentially stable.
Y.L. Chen, G.Q. Xu / J. Math. Anal. Appl. 409 (2014) 851867 865
Fig. 2. The spectrum graph.
For sufficient small > 0, denote by the negative real
=

1
+
2

2
+.
According to the asymptotic expression of the eigenvalues of A in Theorem 3.1, we see that there are at most finitely
many number of eigenvalues of A in the half plane > . Since the asymptotic values of eigenvalues are not obviously
dependent on , we can assert that only the finite eigenvalues of Adepend upon the value of strongly. Therefore, we have
to check whether there exist > 0 such that s(, ) < 0. Obviously, if = 0, Fromthe proof of Lemma 2.1 we can see that
Ais a dissipative operator in H. In this case, we have s(, 0) < 0.
To obtain the stability of (1.1) for > 0, we need the following result, which comes from [16, Theorem 2.1].
Lemma 5.1. Suppose that B R
n
is an open and connected set, h(,

r ) is continuous in (,

r ) C B and analytic in
C, and zeros of h(,

r ) in the right half plane { C| 0} are uniformly bounded. If for any

r B
1
B, where
B
1
is a bounded, closed and connected set, h(,

r ) has no zero on the imaginary axis, then the sum of the orders of the zero of
h(,

r ) in the open right half plane ( > 0) is a fixed number for B

1
, that is, it is independent of the parameter

r B
1
.
To apply Lemma 5.1 to our case, we take
h(,

r ) = D(, , )
where D(, , ) = D() is defined as in (3.4) and =
2
. Here we only consider the case that > 0 is fixed and varies
in neighbor of the original point, B
1
O(r) R
2
, that is,

r = . Obviously, h(,

r ) is analytic in and continuous in

r . Theorem 3.1 ensures that zeros of h(,

r ) in the right half-plane are uniformly bounded. By Lemma 5.1, there exists a
small neighbor B
1
O(r) in which the zeros of h(,

r ) are constants in the right half-plane. Since

r = = 0, there are
no zeros of h(,

r ) in the right half-plane. Therefore, there is no zero of h(,

r ) in the right half-plane for all

r = B
1
.
Applying Lemma 5.1 to our model, using the continuity of s(, ) with respect to , there exists () > 0 such that
0 < < (), it holds that s(, ) < 0 since s(, 0) < 0. Therefore, we have the following result.
Theorem 5.2. Let Abe defined by (2.3) and (2.4). Then for each > 0, there exists () > 0 such that s(, ) < 0 provided
that 0 < < (). Hence the closed loop system (1.1) is exponentially stable.
From Theorem 5.2 we know that one can choose smaller such that the system (1.1) is exponentially stable. However
we cannot obtain an estimate for () or the relation between and . Therefore, howto choose the gain parameters is still
an important question in practice. In the next subsection, we shall give some simulations to show the relationship between
and that make the closed-loop system (1.1) stable exponentially.
5.2. Simulation
In this subsection, we give some simulations for the eigenvalues of the system (1.1) to show the relation between
and .
(1) The first simulation is to check the stability of the system for (
1
,
2
) < (
1
,
2
).
Set = (1, 1) and = (5, 5) or set = (0.1, 0.1) and = (2, 2). By Matlab scientific calculation, we obtain the
finitely many number spectral points of the system (1.1), whose distributions are shown in Figs. 2 and 3, respectively. In
Figs. 2 and 3, the notation denotes all spectral points.
866 Y.L. Chen, G.Q. Xu / J. Math. Anal. Appl. 409 (2014) 851867
Fig. 3. The spectrum graph.
Fig. 4. The relation between and the spectrum.
The simulation results (see Figs. 2 and 3) show that all eigenvalues are located on the left hand side of the imaginary
axis. Hence the system (1.1) is exponentially stable with such feedback gains. These results also show that the feedback
controllers are feasible for (
1
,
2
) reasonable large.
(2) The second simulation is to describe change of s(, ) with = (
1
,
2
).
In order to study the behavior of the rightmost eigenvalue of the system with the change of gain parameter , we keep
= (
1
,
2
) fixed and change the value of .
Set = (1, 1), we take of the form = (1, 1) where R. Let change from 0 to 20 and take step length h = 0.1,
calculate the value of s(, ). We obtain the calculation results whose graphics are shown in Fig. 4. But we cannot obtain
an analytic expression between and the maximal real part of the spectrum since the figure is so complex.
The simulation result (see, Fig. 4) shows that all eigenvalues of the system (1.1) are in the left half-plane. In this case,
the maximal real part of spectrum of Ais still negative. This implies that the system is still exponentially stable. However,
when changes in interval [6, 8], s(, ) has a singular variety. The rightmost eigenvalue approximates to 0.2 when is
about 7.
6. Conclusion
We studiedthe stability problemof a robot systemcomposedof two EulerBernoulli beams withnon-collocatedfeedback
controllers. The advantage of this class of the non-collocated feedback controllers is simpler and more feasible in practice.
With the help of the Riesz basis approach and the asymptotic analysis technique, we proved the exponential stability of
uniform EulerBernoulli beam equations under non-collocated boundary feedback controllers with suitable choice of the
feedback gains. The key point of this kind of controllers is the suitable choice of the feedback gains. Although we do not
obtain an analytic expression of relation between (
1
,
2
) and (
1
,
2
), we have proved that such a relation exists.
Y.L. Chen, G.Q. Xu / J. Math. Anal. Appl. 409 (2014) 851867 867
The simulation result (see, Fig. 4) shows that for the variables and of the form = (1, 1) and = (1, 1), R,
all eigenvalues of the system(1.1) are in the left half-plane. Hence the corresponding closed loop systems are exponentially
stable. We observe that and need not have this form. Our further work will determine (
1
,
2
) for certain fixed
= (
1
,
2
).
Acknowledgments
The authors would like to thank the referees for their useful and helpful comments and suggestions.
References
[1] R.H. Cannon, E. Schmitz, Initial experiments on the end-point control of a flexible one-link robot, Int. J. Robot. Res. 3 (3) (1984) 6275.
[2] S.K. Chang, I. Lasiecka, R. Triggiani, Finite element compensators for thermoelastic systems with boundary control and point observation, Numer.
Funct. Anal. Optim. 20 (56) (1999) 419435.
[3] Y.L. Chen, Z.J. Han, G.Q. Xu, Exponential stability of string system with variable coefficients under non-collocated feedback controls, Asian J. Control
13 (1) (2011) 148163.
[4] P.A. Chodavarapu, M.W. Spong, On noncollocated control of a single flexible link, in: Pro. IEEE International Conference on Robotics and Automation,
Minneapolis, Minnesota, 1996, pp. 11011106.
[5] M.S. de Queiroz, C.D. Rahn, Boundary control and noise in distributed parameter systems: an overview, Mech. Syst. Signal Process. 16 (1) (2002) 1938.
[6] B.Z. Guo, Riesz basis property and exponential stability of controlled EulerBernoulli beamequations with variable coefficients, SIAMJ. Control Optim.
40 (6) (2002) 19051923.
[7] B.Z. Guo, J.M. Wang, K.Y. Yang, Dynamic stabilization of an EulerBernoulli beam under boundary control and non-collocated observation, Systems
Control Lett. 57 (9) (2008) 740749.
[8] B.Z. Guo, C.Z. Xu, The stabilization of a one-dimensional wave equation by boundary feedback with noncollocated observation, IEEE Trans. Automat.
Control 52 (2) (2007) 371377.
[9] E. Hendrickson, I. Lasiecka, Finite dimensional approximations of boundary control problems arising in partially observed hyperbolic systems, Dyn.
Contin. Discrete Impuls. Syst. 1 (1) (1995) 101142.
[10] G. Ji, I. Lasiecka, Partially observed analytic systems with fully unbounded actuators and sensors-FEM algorithms, Comput. Optim. Appl. 11 (2) (1998)
111136.
[11] W. Lacarbonara, H. Yabuno, Closed-loop nonlinear control of an initially imperfect beamwith non-collocated input, J. Sound Vib. 273 (2004) 695711.
[12] I. Lasiecka, Galerkin approximations of infinite dimensional compensators for flexible structures with unbounded control action, Acta Appl. Math. 28
(2) (1992) 101133.
[13] I. Lasiecka, Finite element approximation of compensator design for analytic generators with fully unbounded controls/observations, SIAM J. Control
Optim. 33 (1) (1995) 6788.
[14] L.Y. Liu, K. Yuan, Noncollocated passivity-based control of a single-link flexible manipulator, Robotica 21 (2003) 117135.
[15] A. Pazy, Semigroups of Linear Operators and Applications to Partial Differential Equations, Springer-Verlag, Berlin, 1983.
[16] S.G. Ruan, J.J. Wei, On the zeros of transcendental functions with applications to stability of delay differential equations with two delays, Dyn. Contin.
Discrete Impuls. Syst. Ser. A Math. Anal. 10 (2003) 863874.
[17] J.H. Ryu, D.S. Kwon, B. Hannaford, Control of a flexible manipulator with noncollocated feedback: time domain passivity approach, IEEE Trans. Robot.
20 (4) (2003) 121134.
[18] J.H. Ryu, D.S. Kwon, B. Hannaford, Control of a flexible manipulator with noncollocated feedback: time-domain passivity approach, IEEE Trans. Robot.
20 (2004) 776780.
[19] M.A. Shubov, Riesz basis property of root vectors of non-self-adjoint operators generated by aircraft wing model in subsonic airflow, Math. Methods
Appl. Sci. 23 (8) (2000) 15851615.
[20] V.A. Spector, H. Flashner, Modelling and design implications of non-collocated control in flexible systems, ASME J. Dyn. Syst. Meas. Control 112 (1990)
186193.
[21] J.M. Wang, G.Q. Xu, S.P. Yung, Exponential stability of variable coefficients Rayleigh beams under boundary feedback controls: a Riesz basis approach,
Systems Control Lett. 51 (1) (2004) 3350.
[22] S.T. Wu, Virtual passive control of flexible arms with collocated and noncollocated feedback, J. Robot. Syst. 18 (11) (2001) 645655.
[23] G.Q. Xu, Boundary feedback exponential stabilization of a Timoshenko beam with both ends free, Internat. J. Control 78 (4) (2005) 286297.
[24] G.Q. Xu, D.X. Feng, S.P. Yung, Riesz basis property of the generalized eigenvector system of a Timoshenko beam, IMA J. Math. Control Inform. 21 (8)
(2004) 6583.
[25] G.Q. Xu, Z.J. Han, S.P. Yung, Riesz basis property of serially connected Timoshenko beams, Internat. J. Control 80 (3) (2007) 470485.
[26] G.Q. Xu, S.P. Yung, The expansion of semigroup and criterion of Riesz basis, J. Differential Equations 210 (1) (2005) 124.
[27] B. Yang, C.D. Mote Jr., On time delay in noncolocated control of flexible mechanical systems, ASME J. Dyn. Syst. Meas. Control 114 (1992) 409415.
[28] R.M. Young, An Introduction to Nonharmonic Fourier Series, Academic Press, Inc., London, 2001.
[29] K. Yuan, L.Y. Liu, On the stability of zero dynamics of a single-link flexible manipulator for a class of parametrized outputs, J. Robot. Syst. 20 (10) (2003)
581586.