Congreso Nacional de Control Automático 2017 487

Monterrey, Nuevo León, Mexico, Octubre 4-6, 2017

On the Leader-following Consensus of

Distributed Order Multi-agent Systems
G. Nava-Antonio ˚ G. Fernández-Anaya ˚
E. G. Hernández-Martı́nez ˚˚ J. Jamous-Galante ˚
E.D. Ferreira-Vazquez ˚˚˚ J.J. Flores-Godoy ˚˚˚˚
˚
Departamento de Fı́sica y Matemáticas, Universidad Iberoamericana,
Ciudad de México, México
˚˚
Departamento de Ingenierı́as, Universidad Iberoamericana, Ciudad
de México, México
˚˚˚
Departamento de Ingenierı́a Eléctrica, Facultad de Ingenierı́a y
Tecnologı́as, Universidad Católica del Uruguay, 11600, Montevideo,
Uruguay
˚˚˚˚
Departamento de Matemática, Facultad de Ingenierı́a y
Tecnologı́as, Universidad Católica del Uruguay, 11600, Montevideo,
Uruguay

Abstract: This work presents an extension of recent results about the distributed coordination
of multi-agent systems. The novelty of the present paper lies in the inclusion of the distributed
order derivative in the dynamics of the studied problems, something which has not been
investigated yet, to our best knowledge. By applying a generalization of the Lyapunov direct
method and properties of inequalities, we give sufficient conditions to guarantee leader-following
consensus of certain distributed order multi-agent systems with single and double integrator
dynamics. As particular cases of our main results, we get back the already known fractional
counterpart of the theory discussed in this article.

Keywords: Fractal systems; Lyapunov method; Dynamic systems; Asymptotic stability;

Leader-following consensus; Distributed fractional calculus.

1. INTRODUCTION In addition to the articles already cited, it is worth

Recently, the study of multi-agent systems and their
distributed coordination has gained the attention of re-
searchers. This field analyzes the accomplishment of group
objectives through the local interaction of the agents.
A key concept in distributed coordination is consensus,
which determines if the dynamics of the agents will
converge to a certain desired value. The relevance of
consensus of multi-agent systems is clear when noticing
its diverse applications: description of sensor networks,
formation of drones, interaction of various satellites, etc.
See Bosse and Pantke (2013); Wu et al. (2010).
Consensus of multi-agent systems has been mostly studied
from the framework of integer order calculus. The con-
sensus problem of integer order systems with single and
double integrator dynamics was analyzed in Djaidja and
Wu (2015) and Liu et al. (2015). Furthermore, research
has explored models with time delays (Qin et al. (2015))
or external disturbances (Hu et al. (2015)), and controllers
have been designed accordingly (Yang et al. (2015)).
Some of the ideas discussed in the previous paragraph
have been extended to systems with fractional order dy-
namics. For instance: Yu et al. (2015) explored the leader-
following consensus of fractional order nonlinear multi-
agent systems described by directed graphs; and Cao
et al. (2010) investigated the distributed coordination of
networked multi-agent systems with fractional dynamics.
highlighting the work done in Ren et al. (2015), which
presents a method for determining if nonlinear fractional
multi-agent systems, with single and double integrator
dynamics, achieve leader-following consensus. That paper
is the foundation of the present work, where we will
generalize leader-following consensus results for systems
with distributed order derivatives.
In summary, distributed order calculus generalizes the
fractional and integer order differential operators. Besides
being mathematically interesting, the distributed order
derivative has been proposed to model certain physical
systems in a more adequate fashion (see Naber (2004)). A
conceptual interpretation of this new operator is provided
in Lorenzo and Hartley (2002); there, it is argued that the
differentiation order in a given problem can depend on
anisotropic properties of the physical system in question,
where each differential element should have its own order
of differentiation.
Motivated by the above discussion, we present methods
that allow proving if certain distributed order multi-agent
systems achieve leader-following consensus. We consider
the cases where single and double integrator dynamics are
involved. By choosing appropriately the weight function
of the distributed derivative, we recover Theorems 11 and
13 of Ren et al. (2015). As far as we are concerned, the
present work is the first one that studies the consensus
problem of multi-agent systems using distributed order The next Theorem generalizes the Lyapunov direct
calculus. method for nonlinear time-varying distributed order sys-
tems.
2. PRELIMINARY CONCEPTS Theorem 6. Fernández-Anaya et al. (2017). Let x “ 0 be
an equilibrium point for the system (4). Assume that there
In this section some basic definitions and properties of exists a Lyapunov function V pxptq, tq satisfying
the distributed order derivative are presented, as well as
some concepts of Lyapunov stability, graph theory, and α1 }x}a ď V pxptq, tq ď α2 }x}ab , (6)
consensus of multi-agent systems.
C bpαq
2.1 Distributed order calculus and Lyapunov stability 0 Dt V pxptq, tq ď ´α3 }x}ab , (7)
where α P p0, 1q and a, b, αi ą 0, i “ 1, 2, 3. If the roots of
Definition 1. Aguila-Camacho et al. (2014). The Caputo Bpsq ` α3 {α2 “ 0 are in the open left-half complex plane,
fractional derivative of order α P R` is defined as follows and bpαq is such that L´1 t1{pBpsq ` α α3
qu ě 0, @t ě 0,
2
żt then the origin of the system is asymptotically stable.
C α 1 f pmq pζq
t0 Dt f ptq “ dζ, t ą t0 (1)
A useful property for building appropriate Lyapunov
Γpm ´ αq t0 pt ´ ζqα´m`1
ş8 functions for distributed order systems is presented next:
where m “ mintk P N | k ą αu and Γprq “ 0 tr´1 e´t dt. Lemma 7. Fernández-Anaya et al. (2017). Let xptq P Rn .
Definition 2. Jiao et al. (2012). The distributed order Then, for t ě t0 and @α P p0, 1s, the following relationship
derivative with respect to the density function (weight holds:
function) bpαq ě 0 is defined as follows: 1 C bpαq T C bpαq
żm t0 Dt x ptqxptq ď xptqT t0 Dt xptq. (8)
C bpαq
2
t0 Dt xptq “ bpαqxpαq ptqdα, (2)
m´1 2.2 Graph theory fundamentals
C α
where m ´ 1 ă α ă m and xpαq ptq “ t0 Dt xptq. Now, we will discuss some concepts of graph theory (see
The Laplace transform of a distributed order derivative is Ren and Beard (2008)). Graphs can be used to describe
żm «
m´1
the interaction topology of multi-agent systems. A graph
bpαq
ÿ G “ pV, Wq is defined by its vertices, contained in the
Lt C
0 Dt xptqu “ bpαq sα Xpsq ´ sα´1´k ˆ
m´1 set V “ tv1 , v2 , . . . , vn u, and its edges, elements of the
k“0
ff
m´1
set W Ď V 2 . In our case, each agent corresponds to a
ÿ 1 vertex. Throughout this text, we will consider undirected
xpkq p0` q dα “ BpsqXpsq ´ Bpsqxpkq p0` q, (3)
sk`1 graphs, where each edge is an unordered pair pvi , vj q “
k“0
pvj , vi q which signifies that agent j can receive the state
şm information of agent i and vice versa; on this situation
where Bpsq “ m´1 bpαqsα dα. Since we will work with
it is said that agent i and agent j are neighbors. The
α P r0, 1s then m “ 1.
neighbors of agent i are denoted by Ni . The adjacency
Hypothesis 3. In this text we will focus on distributed matrix A “ raij s P Rnˆn and the Laplacian matrix
order systems of the form L “ rlij s P Rnˆn are used to represent algebraically
C bpαq
the interaction graph; these are defined as ař ij “ 1 if
t0 Dt xptq “ f pxptq, tq, (4) pvi , vj q P W, aij “ 0 if pvi , vj q R W, lii “ jPNi aij ,
and lij “ ´aij for i ‰ j.
where f pxptq, tq P L1 r0, 8s, xptq P R is such that
C α
t0 Dt xptq ă M for t P r0, 8q and @α P p0, 1q, bpαq is an The following Lemma presents a property of multi-agent
absolutely integrable function on the interval α P r0, 1s systems that will be useful in the proofs of our main
ş1
and satisfies that 0 bpαqsα dα ‰ 0, for Repsq ą 0. This results.
kind of system has a unique solution given in the next Lemma 8. Ren and Cao (2010). If an interaction graph
Theorem. is undirected and connected, then the matrix H “
Theorem 4. Ford and Morgado (2012). If Hypothesis 3 is L ` B will be symmetric and positive definite, where
satisfied, then system (4) has the unique solution: B “diagtb1 , b2 , . . . , bn u, i P t1, 2, . . . , nu, and bi ě 0 does
˜ « ff¸ not always equal 0.
1
xptq “ xp0q ` f ˚ L ´1
ş1 ptq , (5) 2.3 Consensus of multi-agent systems
0
bpαqpsqα dα
The definition of a Lipschitz function, relevant in the
where
ş8 ˚ is the convolution operator, i.e. gptq ˚ hptq “ study of multi-agent systems, is presented next.
´8
gpτ qhpt ´ τ qdτ . Definition 9. Podlubny (1999). Let kpt, xq be a real-
valued continuous function, defined in the domain G Ď R.
Note that Theorem 4 can be easily extended for the case kpt, xq is said to satisfy the Lipschitz condition with re-
when xptq P Rn . spect to x in the domain G if,
Definition 5. The constant x0 is said to be an equilibrium
point of the system (4), if f px0 , tq “ 0, @t. |kpt, x1 q ´ kpt, x2 q| ď θ|x1 ´ x2 |, (9)
where θ is defined as the Lipschitz constant.
The next definitions present the concept of leader-
following consensus for multi-agent systems with single
and double integrator dynamics.
Definition 10. Ren et al. (2015). A multi-agent system of
n agents with single integrator dynamics achieves leader-
following consensus if, for any initial condition, its solu-
tion satisfies
lim }xi ptq ´ xr ptq} “ 0, i P t1, 2, . . . , nu , (10)
tÑ8
where xi ptq is the state of the ith agent and xr ptq is the
state of the virtual leader.
Definition 11. Ren et al. (2015). A multi-agent system of
n agents with double integrator dynamics achieves leader-
following consensus if it satisfies the following equalities,
for any initial condition:
lim }xi0 ptq ´ xr0 ptq} “ 0, lim }xi1 ptq ´ xr1 ptq} “ 0,
tÑ8 tÑ8
i P t1, 2, . . . , nu , (11)
where xi0 ptq, xi1 ptq are the states of the ith agent and
xr0 ptq, xr1 ptq are the states of the virtual leader.
Before presenting the main results of the present paper,
we will make a couple of clarifications: hereinafter, we
will consider that all agents are in a one-dimensional
space; there is not loss of generality in doing so, since
all results presented will be valid for m dimensions using
the Kronecker product. Additionally,
a in this work we will
consider the norms: }Z}ř“ z12 ` z22 ` . . . ` zn2 , Z P Rn ,
n
and }B}1 “max1ďjďn t i“1 |bij |u, B “ rbij s P Rnˆn , for
vectors and for matrices respectively.
3. MAIN RESULTS
In this section we present two Theorems that allow to
where aij is the pi, jqth entry of the corresponding adja-
cency matrix A P Rnˆn , β ě 0 and bi is a nonnegative
constant that does not always equal zero.
In the following, we will use the matrices H “ L ` B,
where B “ diagtbi , b2 , . . . , bn u and L is the Laplacian
matrix associated with the interaction graph. We will refer
to the eigenvalues of H as λi , i P t1, 2, . . . , nu. Note that
the Jordan canonical form of H satisfies Λ “ P ´1 HP ,
where P “ rpij s P Rnˆn and P ´1 “ rp̄ij s P Rnˆn .
Sufficient conditions to guarantee the leader following
consensus of distributed order multi-agent systems with
single integrator dynamics are given next.
Theorem 12. Consider a multi-agent system with a fixed
interaction graph G, which is undirected and connected.
Suppose that the function f pt, xq in (12) is Lipschitz in x
with Lipschitz constant θ, and that bpαq fulfills the con-
ditions stated in Theorem 6. Additionally, if the following
inequality is satisfied
› ›
β }P }1 ›P ´1 ›1
ą , (14)
l min1ďiďn tλi u
where l “ 2θ, then system (12) achieves leader-following
consensus, using the control input (13).
Proof. Substituting (13) in (12) produces
«
n
ÿ
C bpαq
0 D t x i ptq “f pt, xi ptqq ´ β aij pxi ptq ´ xj ptqq
j“1
ff
` bi pxi ptq ´ xr ptqq . (15)
Introducing the change of variables yi ptq “ xi ptq ´ xr ptq,
i P t1, 2, . . . , nu, we obtain
C bpαq
0 Dt yi ptq “f pt, yi ptq ` xr ptqq ´ f pt, xr ptqq (16)
# +
ÿn

determine if certain distributed order multi-agent sys- ´β aij pyi ptq ´ yj ptqq ` bi yi ptq .
tems, with corresponding fixed undirected graphs, achieve j“1
leader-following consensus. First, we study the case of
We can rewrite (16) in vector form as
systems with single integrator dynamics and then with
double integrator dynamics. C bpαq
0 Dt Y ptq “ F pt, Y ptqq ´ βHY ptq, (17)
A distributed order multi-agent system with single in-
tegrator dynamics can be described by the differential where Y ptq “ ry1 ptq, y2 ptq, . . . , yn ptqsT and F pt, Y ptqq “
equations rf pt, y1 ptq ` xr ptqq ´ f pt, xr ptqq, f pt, y2 ptq ` xr ptqq ´
f pt, xr ptqq, . . . , f pt, yn ptq ` xr ptqq ´ f pt, xr ptqqsT .
C bpαq
0 Dt xi ptq “ f pt, xi ptqq ` ui ptq, i P t1, 2, . . . , nu ,
Multiplying (17) by P ´1 on the left hand side and defining
C bpαq
0 Dt xr ptq “ f pt, xr ptqq, (12) Zptq “ P ´1 Y ptq yields
C bpαq
where xr ptq is the state of the virtual leader of the system, 0 Dt Zptq “ P ´1 F pt, Y ptqq ´ βP ´1 HP P ´1 Y ptq
and xi ptq, ui ptq, f pt, xi ptqq are the state, control input and “ P ´1 F pt, P Zptqq ´ βΛZptq, (18)
inherent nonlinear dynamics of the ith agent, respectively.
whereřZptq “ rz1 ptq, z2 ptq, . . . , zn ptqsT , Fř pt, P Zptqq “
In order to achieve leader-following consensus, we will use n n
rf pt, k“1 p1k zk ptq`xr ptqq´f pt,řxr ptqq, f pt, k“1 p2k zk ptq
the following controller: n
`xr ptqq ´ f pt, xr ptqq, . . . , f pt, k“1 pnk zk ptq ` xr ptqq ´
#
n
ÿ f pt, xr ptqqsT . Assuming that system (18) satisfies the hy-
ui ptq “ ´ β aij pxi ptq ´ xj ptqq (13) pothesis of Theorem 4, it has a unique solution Zptq.
j“1
+ Note that, since the undirected graph G is connected and
considering that bi ě 0 @i, H is symmetric and positive
` bi pxi ptq ´ xr ptqq , definite (see Lemma 8). Hence Λ “ diagtλ1 , λ2 , . . . , λn u,
λi ą 0, i P t1, 2, . . . , nu.
Now, we will demonstrate that system (18) is asymptot- where 0n is the zero vector of size n. Recalling that
ically stable. Consider the following Lyapunov candidate yi “ xi ptq ´ xr ptq, i P t1, 2, . . . , nu, we have determined
function, which satisfies inequality (6), for c ď 12 : that limtÑ8 |xi ptq ´ xr ptq| “ 0, and consequently the
multi-agent system achieves leader following consensus.
2n
2 1ÿ 2 2
c }Zptq} ď V pt, Zptqq “ z ď }Zptq} . (19) Our next result deals with the leader-following consensus
2 i“1 i
of distributed order multi-agent systems with double-
We can analyze the distributed order derivative of integrator dynamics. A multi-agent system with this kind
V pt, Zptqq by using Lemma 7: of behavior can be described by the differential equations
n C bpαq
C bpαq 1 ÿ C bpαq 2 0 Dt xi0 ptq “ xi1 ptq, (24)
D t V pt, Zptqq “ D zi ptq (20) C bpαq
0
2 i“1 0 t 0 Dt xi1 ptq “ f pt, xi0 ptq, xi1 ptqq ` ui ptq,
#
n n
ÿ
C bpαq
ÿ where i P t1, 2, . . . , nu, and xi0 ptq, xi1 ptq are the states
ď zi ptq 0 Dt zi ptq “ zi ptq ´ βλi zi ptq
of the ith agent. In this case, the virtual leader double
i“1 i“1
n
«
n
ff + integrator dynamics is given by
ÿ ÿ
p̄ij f pt, pjk zk ptq ` xr ptqq ´ f pt, xr ptqq C bpαq
` 0 Dt xr0 ptq “ xr1 ptq, (25)
j“1 k“1 C bpαq
0 Dt xr1 ptq “ f pt, xr0 ptq, xr1 ptqq,
# ˇ
n
ÿ n
ÿ ˇ ÿn
βλi zi2 ptq |zi ||p̄ij |ˇf pt, pjk zk ptq
ˇ
ď ´ ` where xr0 ptq and xr1 ptq are the states of the virtual leader.
i“1 j“1
ˇ
k“1
ˇ+
ˇ In order to achieve leader-following consensus in system
` xr ptqq ´ f pt, xr ptqqˇ .
ˇ (24), we will use the following controller, based on the
ˇ analysis presented in Ren et al. (2015):
n
ÿ
Considering that f pt, xq is Lipschitz with respect to x, ui ptq “ ´ aij rpxi 0ptq ´ xj 0ptqq ` βpxi1 ptq ´ xj1 ptqqs
with θ “ 2l , it follows from (20) that j“1
#
n
ÿ ´ bi rpxi 0ptq ´ xr 0ptqq ` βpxi1 ptq ´ xr1 ptqqs , (26)
C bpαq
0 Dt V pt, Zptqq ď ´ βλi zi2 ptq (21)
i“1 where aij (i,j = 1,2,. . . , n) is the pi, jqth entry of corre-
sponding adjacency matrix A P Rnˆn , β ě 0 and di is a
ˇ ˇ+
n n n
ÿ l ˇ ÿ ˇ ÿ
pjk zk ˇ ď ´β λi zi2 ptq nonnegative constant that does not always equal zero.
ˇ ˇ
` |zi ||p̄ij | ˇ
j“1
2 ˇ
k“1
ˇ i“1
n n n
In the proof of the next result we will use the ˆ matrices B˙“
l ÿÿ ÿ 0nˆn ´In
` |zi ||p̄ij ||pjk ||zk |. diagtbi , b2 , . . . , bn u, H “ L ` B, and M “ P
2 i“1 j“1 k“1 H βH
Exchanging the indices i to j, j to k, k to i and doing R2nˆ2n , where In and 0nˆn are the n ˆ n identity matrix
some simple algebra produces and zero matrix, respectively. We will denote the Jordan
n canonical form of matrix M as ∆, which satisfies ∆ “
Q´1 M Q, where Q “ rqij s P R2nˆ2n and Q´1 “ rq̄ij s P
ÿ
C bpαq
0 Dt V pt, Zptqq ď ´β min tλi u zi2 ptq (22)
1ďiďn
i“1 R2nˆ2n . To refer to the eigenvalues of matrices H and
l
n
ÿ n
ÿ n
ÿ ÿn M we will use the notation λi and δi , i P t1, 2, . . . , nu
` |zj ||p̄jk ||pki ||zi | ď ´β min tλi u zi2 ptq respectively.
2 i“1 j“1 k“1 1ďiďn
i“1
# + # + The following Theorem allows to determine if a multi-
n n n n
l ÿ ÿ ÿÿ agent system with double integrator dynamics achieves
` max |p̄jk | max |pki | |zj ||zi |. leader-following consensus over a fixed undirected graph.
2 1ďkďn j“1 1ďiďn
k“1 i“1 j“1
řn řn řn 2 Theorem 13. Asume that in system (24) f px, tq is Lip-
Using the properties i“1 j“1 |zj ||zi | “ p i“1 |zi |q , schitz and the distribution function bpαq satisfies the
1
řn 2 řn 2 hypothesis of Theorem 6. If the fixed undirected graph
› |zi |q ď i“1 |zi | and the definitions of }P }1 ,
›2 p´1i“1 G is connected and the inequalities
›P › we obtain
1
« ff d
C bpαq
› ´1 › 4 2
0 Dt V pt, Zptqq ď ´ β min tλi u ` l }P } P ˆ 1` ěβě a (27)
› ›
1ďiďn 1 1
max1ďiďn tλi u min1ďiďn tλi u
« ff
n
ÿ › › 2
zi2 ď ´ β min tλi u ` l }P }1 ›P ´1 ›1 }Zptq} . min1ďiď2n tδi u
i“1
1ďiďn 2θ ă (28)
}Q}1 }Q´1 }1
(23)
are satisfied, then leader-following consensus is achieved,
It is clear
› from
› (23) that inequality (7) is satisfied if β{l ą using the control input (26).
}P }1 ›P ´1 ›1 { min1ďiďn tλi u. Under this hypothesis, we
have shown that Zptq is asymptotically stable at its origin. Proof. Substituting the expression of the controller (26)
Since Y ptq “ P Zptq, it follows that limtÑ8 Y ptq “ 0n , in (24) we obtain
 C bpαq 2n
0 Dt xi0 ptq “ xi1 ptq, (29) C bpαq 1ÿ C bpαq 2
C bpαq 0 Dt V pt, Zptqq “ 0 Dt zi ptq (35)
2 i“1
0 Dt xi1 ptq “ f pt, xi0 ptq, xi1 ptqq
#
ÿn 2n
ÿ 2n
ÿ
aij rpxi0 ptq ´ xj0 ptqq ` βpxi1 ptq ´ xj1 ptqqs C bpαq
´ ď zi ptq 0 Dt zi ptq “ zi ptq ´ δi zi ptq
j“1 i“1 i“1
n „
´ bi rpxi0 ptq ´ xr0 ptqq ` βpxi1 ptq ´ xr1 ptqqs. ÿ
` q̄i,j`n f pt, yj0 ptq ` xr0 ptq, yj1 ptq ` xr1 ptqq
Making the change of variables yi0 ptq “ xi0 ptq ´ xr0 ptq, j“1
yi1 ptq “ xi1 ptq ´ xr1 ptq, i P t1, 2, . . . , nu results in +
C bpαq ´ f pt, xr0 ptq, xr1 ptqq .
0 Dt yi0 ptq “ yi1 ptq, (30)
C bpαq
0 Dt yi1 ptq “ f pt, xi0 ptq, xi1 ptqq ´ f pt, xr0 ptq, xr1 ptqq Since f pt, x0 , x1 q is assumed to be Lipschitz, with θ “ 2l ,
n
ÿ we can simplify (35) as follows
´ aij rpyi0 ptq ´ yj0 ptqq ` βpyi1 ptq ´ yj1 ptqqs «
2n
j“1
ÿ
C bpαq
0 Dt V pt, Zptqq ď ´ δi zi2 ptq (36)
´ bi pyi0 ptq ` βyi1 ptqq. i“1
ff
n
In vector form, we can rewrite (30) as ÿ l
ˆ ˙ ˆ ˙ ` |q̄i,j`n | |yj0 ptq ` yj1 ptq||zi ptq|
C bpαq Y0 ptq Y0 ptq 2
0 Dt “´M (31) j“1
Y1 ptq Y1 ptq 2n
«
n
ˇ
2n
ˆ ˙ ÿ
2
ÿ l ˇˇ ÿ
0nˆ1 “ ´ δi zi ptq ` |q̄i,j`n | ˇ qjk zk ptq
` , 2 ˇ k“1
F pt, Y0 ptq, Y1 ptqq i“1 j“1
ˇ ff
ÿ2n ˇ
where Yk ptq “ ry1k ptq, y2k ptq, . . . , ynk ptqsT pk “ 0, 1q, and ` qj`n,k zk ptqˇ|zi ptq| .
ˇ
F pt, Y0 ptq, Y1 ptqq “ rf pt, y10 ` xr0 ptq, y11 ptq ` xr1 ptqq ´ k“1
ˇ
f pt, xr0 ptq, xr1 ptqq, . . . , f pt, yn0 ptq ` xr0 ptq, yn1 ptq ` xr1 ptqq Doing some simple algebra and switching the indices i to
´ f pt, xr0 ptq, xr1 ptqqsT . k, and k to i:
Based on the analysis done in Song et al. (2010) the 2n 2n n 2n
C bpαq
ÿ
2 l ÿÿ ÿ
eigenvalues of M are 0 D t V pt, Zptqq ď ´ δ z
i i ptq ` |q̄i,j`n |ˆ
a i“1
2 i“1 j“1 k“1
λi ` β 2 λ2i ´ 4λi
δi “ , (32) 2n
2
ÿ
b |qjk ptq ` qj`n,k | |zk ptq||zi ptq| ď ´ min tδi u zi2 ptq
1ďiď2n
λj´n ´ β 2 λ2j´n ´ 4λj´n i“1
δj “ ,
# + # +
2n n
2 l ÿ ÿ
` max |q̄k,j`n | max |qji ptq ` qj`n,i | ˆ
where i P t1, 2, . . . , nu, j P tn ` 1, n ` 2, . . . , 2nu. It 2 1ďjďn k“1
1ďiď2n
j“1
is clear from (32) that all eigenvalues of M will be 2n ÿ
ÿ 2n
positive real numbers if β 2 λ2j´n ´ 4λj´n ě 0 and |zk ptq||zi ptq|. (37)
b
i“1 k“1
λj´n ´ β 2 λ2j´n ´ 4λj´n ě 0, which are equivalent to !ř )
2n › ›
b
4
a Considering that max1ďjďn k“1 |q̄k,j`n | “ ›Q´1 ›1 ,
1 ` max1ďiďn tλi u ě β ě 2{ min1ďiďn tλi u. There- !ř )
n řn
fore ∆ is a diagonal positive definite matrix, specifically max1ďiď2n j“1 |qji ptq ` qj`n,i | “ }Q}1 and i“1 |zi |2
řn řn
∆ “diagtδ1 , δ2 , . . . , δ2n u, δi ą 0, i P t1, 2, . . . , 2nu. ě 12 i“1 j“1 |zj ||zi |, it can be obtained form (37) the
Multiplying (31) by Q´1 on the left handˆside and following inequality:
˙ defining
2n
T ´1 Y0 ptq C bpαq
ÿ
Zptq “ rz1 ptq, z2 ptq, . . . , z2n ptqs “ Q , yields D V pt, Zptqq ď ´ min tδ u zi2 ptq (38)
Y1 ptq 0 t i
1ďiď2n
ˆ ˙ i“1
C bpαq 0nˆ1
0 Dt Zptq “ ´∆Zptq ` Q´1 . (33) 2n
F pt, Y0 ptqY1 ptqq › › ÿ
` l }Q}1 ›Q´1 ›1 zi2 ptq
i“1
If system (33) fulfills the conditions stated in Theorem 4, ˆ ˙
then it will have a unique Zptq, which we will prove to be
› ´1 › 2
ď´ min tδi u ´ l }Q}1 Q 1 }Zptq} .
› ›
asymptotically stable at its origin. Consider the following 1ďiď2n
Lyapunov candidate function, which clearly satisfies in- › ›
equality (6), for c ď 12 : If l ă min1ďiď2n tδi u { }Q}1 ›Q´1 ›1 , the coefficient of
2
}Zptq} in the right hand side of (38) will be nega-
2n
2 1ÿ 2 2 tive and consequently inequality (7) will be satisfied.
c }Zptq} ď V pt, Zptqq “ z ď }Zptq} . (34)
2 i“1 i Hence, according to Theorem 6, system (33) is asymp-
totically
ˆ ˙ stable at its origin. Furthermore, it follows from
Using the property introduced in Lemma 7 in (34), Y0 ptq
“ QZptq that limtÑ8 |xi1 ptq ´ xr1 ptq| “ 0, given
produces Y1 ptq
that yi0 ptq “ xi0 ptq ´ xr0 ptq, yi1 ptq “ xi1 ptq ´ xr1 ptq,
i P t1, 2, . . . , nu. This means that system (24) accom-
plishes second-order leader-following consensus.
Remark 14. Note that Theorems 12 and 13 extend the
main results of Ren et al. (2015); the contribution of our
work is that it includes distributed order dynamics, which
generalizes the fractional order case. If the distribution
function is chosen as bpαq “ δpα ´ βq we get back the
fractional versions of our results, which allow to determine
if fractional multi-agent systems achieve leader-following
consensus.
Remark 15. As far as we are concerned, research has not
found numerical methods to solve the following problems
with distributed order derivatives: linear non-autonomous
differential equations, non-linear differential equations,
systems of differential equations. Simulations for some
simple distributed order equations can be found in Di-
ethelm and Ford (2009); Katsikadelis (2014); Mashayekhi
and Razzaghi (2016). Considering this, we leave for future
work the numerical solution of examples that verify the
usefulness of the results presented in this work.
4. CONCLUSION
In this paper we introduced a couple of consensus results
for distributed order multi-agent systems. In Theorems
12 and 13, we presented methods to determine if leader-
following consensus is accomplished in linear or nonlin-
ear multi-agent systems with associated fixed undirected
graphs for the cases of single and double integrator dy-
namics, respectively.
As future work, a possible path would be to obtain
similar results for more general families of distributed
order multi-agent systems, perhaps using different kinds
of control inputs. Another potential objective would be
to attend the issue discussed in Remark 15 by designing
numerical methods to solve unexplored distributed order
problems. Consequently we would be able to verify, with
computer simulations, the theoretical results introduced
in the present article.
REFERENCES
Aguila-Camacho, N., Duarte-Mermoud, M.A., and Galle-
gos, J.A. (2014). Lyapunov functions for fractional or-
der systems. Commun. Nonlinear Sci. Numer. Simul.,
19, 2951–2957.
Bosse, S. and Pantke, F. (2013). Distributed computing
and reliable communication in sensor networks using
multi-agent systems. Prod. Eng., 7(1), 43–51.
Cao, Y., Li, Y., Ren, W., and Chen, Y. (2010). Dis-
tributed coordination of networked fractional-order sys-
tems. IEEE Transactions on Systems, Man, and Cy-
bernetics, Part B (Cybernetics), 40(2), 362–370.
Diethelm, K. and Ford, N.J. (2009). Numerical analysis
for distributed-order differential equations. J. Comput.
Appl. Math., 225(1), 96–104.
Djaidja, S. and Wu, Q. (2015). Leader-following consensus
for single-integrator multi-agent systems with multi-
plicative noises in directed topologies. International
Journal of Systems Science, 46(15), 2788–2798.
Fernández-Anaya, G., Nava-Antonio, G., Jamous-
Galante, J., Muñoz-Vega, R., and Hernández-Martı́nez,
E. (2017). Asymptotic stability of distributed order
nonlinear dynamical systems. Commun. Nonlinear
Sci. Numer. Simul., 48, 541–549.
Ford, N.J. and Morgado, M.L. (2012). Distributed order
equations as boundary value problems. Computers &
Mathematics with Applications, 64(10), 2973–2981.
Hu, M., Guo, L., Hu, A., and Yang, Y. (2015). Leader-
following consensus of linear multi-agent systems with
randomly occurring nonlinearities and uncertainties
and stochastic disturbances. Neurocomputing, 149, 884–
890.
Jiao, Z., Chen, Y., and Podlubny, I. (2012). Distributed-
Order Dynamic Systems Stability, Simulation, Applica-
tions and Perspectives. SpringerBriefs in Electrical and
Computer Engineering. Springer.
Katsikadelis, J.T. (2014). Numerical solution of dis-
tributed order fractional differential equations. J. Com-
put. Phys., 259, 11–22.
Liu, Z., Jin, Q., and Chen, Z. (2015). Distributed con-
tainment control for bounded unknown second-order
nonlinear multi-agent systems with dynamic leaders.
Neurocomputing, 168, 1138–1143.
Lorenzo, C.F. and Hartley, T.T. (2002). Variable order
and distributed order fractional operators. Nonlinear
dynamics, 29(1-4), 57–98.
Mashayekhi, S. and Razzaghi, M. (2016). Numerical solu-
tion of distributed order fractional differential equations
by hybrid functions. J. Comput. Phys., 315, 169–181.
Naber, M. (2004). Distributed order fractional sub-
diffusion. Fractals, 12(01), 23–32.
Podlubny, I. (1999). Fractional Differential Equations.
Academic Press, London.
Qin, W., Liu, Z., and Chen, Z. (2015). Impulsive observer-
based consensus control for multi-agent systems with
time delay. Int. J. Control, 88(9), 1789–1804.
Ren, G., Yu, Y., and Zhang, S. (2015). Leader-following
consensus of fractional nonlinear multiagent systems.
Mathematical Problems in Engineering, 2015.
Ren, W. and Beard, R.W. (2008). Distributed consensus
in multi-vehicle cooperative control. Springer.
Ren, W. and Cao, Y. (2010). Distributed coordination of
multi-agent networks: emergent problems, models, and
issues. Springer Science & Business Media.
Song, Q., Cao, J., and Yu, W. (2010). Second-order
leader-following consensus of nonlinear multi-agent sys-
tems via pinning control. Systems & Control Letters,
59(9), 553–562.
Wu, B., Wang, D., and Poh, E.K. (2010). Decentralized
control for satellite formation using local relative mea-
surements only. In Control and Automation (ICCA),
2010 8th IEEE International Conference on, 661–666.
IEEE.
Yang, Y., Yue, D., and Xue, Y. (2015). Decentralized
adaptive neural output feedback control of a class of
large-scale time-delay systems with input saturation.
Journal of the Franklin Institute, 352(5), 2129–2151.
Yu, Z., Jiang, H., and Hu, C. (2015). Leader-following
consensus of fractional-order multi-agent systems under
fixed topology. Neurocomputing, 149, 613–620.