IEEE TRANSACTIONS ON CIRCUITS AND SYSTEMS—II: EXPRESS BRIEFS, VOL. 62, NO. 5, MAY 2015 511
Characterizing the Convergence of a Distributed
Consensus Algorithm via Relative Hull
Yao Chen, Member, IEEE
Abstract—This brief proposed a novel geometric structure
called relative hull. By using this newly introduced concept, a novel
consensus algorithm of multiagent systems was established. It has
been strictly proved that such an algorithm contains a much larger
convergence region with respect to the widely investigated average
consensus algorithms. Furthermore, applications of this algorithm
to consensus of multiagent systems with compasses and consensus
on a torus demonstrated the effectiveness and generality of the
proposed geometric structure.
Index Terms —Consensus, multiagent systems, relative hull.
I. INTRODUCTION
C
ONSENSUS of multiagent systems implies the conver-
gence of the states to some identical values for a group
of dynamical agents [9], [11], [15], [18]. Such a behavior has
been widely applied in multisource data fusion, clock synchro-
nization of sensor networks, computer graphics, and distributed
computation and optimization [4]–[8], [17], [19], [20], [23],
[24]. In addition to the engineering applications of consensus,
the intensive investigations on consensus of multiagent systems
also provide plausible explanations for the popular emergent
behavior of social entities [10] and natural flocks [17].
The most famous average consensus algorithm updates the
states of agents via the convex hull of the neighboring agents
[1]–[3], [12], [16], [21], [22], and such a kind of system is
essentially a linear one and cannot be applied in characterizing
the complex dynamical behaviors of some nonlinear multiagent
systems [14]. Consequently, an interesting question arises of
how to extend the concept of convex hull to a more general
form that also guarantees the consensus of multiagent systems.
Facing this interesting problem, a novel concept called relative
hull will be proposed in this brief. Since the relative hull of any
given set is much larger than that of the convex hull, we can
say that relative hull is a more general concept with respect
to the existing convex hull. By using this newly introduced
concept, a novel consensus algorithm of multiagent systems
will be established at the same time.
Manuscript received September 9, 2014; accepted November 29, 2014. Date
of publication December 29, 2014; date of current version April 23, 2015.
This work was supported in part by the National Natural Science Foundation
of China under Grant 61304157. This brief was recommended by Associate
Editor J. Lu.
The author is with the Department of Computing, The Hong Kong Polytech-
nic University, Hung Hom, Hong Kong, and also with the School of Electronic
and Information Engineering, Beijing Jiaotong University, Beijing 100044,
China (e-mail: chenyao07@gmail.com).
Color versions of one or more of the figures in this brief are available online
at http://ieeexplore.ieee.org.
Digital Object Identifier 10.1109/TCSII.2014.2386258
As one of the application examples, this brief takes multi-
agent system with compasses as an example to verify the effec-
tiveness of the proposed algorithm. We choose a simplified
discrete-time model for characterizing the dynamical behavior
of multiagents with compasses [13], and by compass we mean
that all the agents share a common direction during the evolu-
tion process. Under the assistance of such a compass, each agent
can choose the movement direction in a larger convergence
region, demonstrating the effectiveness and generality of the
relative hull. Furthermore, the main results have also been ap-
plied to the consensus of multiagent system on a torus structure,
which is difficult to be resolved in the traditional framework.
The organization of this brief is as follows. Section II intro-
duces the concept of relative hull and main results. Section III
proves the main theorems. Section IV takes multiagent system
with compasses and torus consensus as two examples to demon-
strate the effectiveness of the proposed concept and algorithms.
Section V concludes this brief.
II. P
ROBLEM FORMULATION AND MAIN RESULTS
A sequence of graphs G(t)=(V,E(t))|
∞
t=0
is called finally
connected if each G(t) is undirected, and
G
∞
= lim
t→∞
k≥t
G(k)
is connected.
A sequence of graphs G(t)=(V, E(t))|
∞
t=0
is called jointly
(strong) connected if there exists T>0 such that
G
k
=
kT +1
t=(k−1)T
G(t)
is (strongly) connected for any k ≥ 1.
1
Given any set S ⊆ R
n
, the convex hull of S is denoted by
conv(S).ForsetS = {x
i
}
m
i=1
⊆ R
n
with a finite number of
elements, the α-convex hull of S is defined by
conv
α
(S)=
m
i=1
μ
i
x
i
:
m
i=1
μ
i
=1,μ
i
≥ α
.
Given a convex set K ⊆ R
n
,forS ⊂ K and Θ ⊆ R
n
, define
the relative hull of S with respect to Θ as
H (K, Θ,S)=K
θ∈Θ
{x : θ
T
x ≤ sup θ
T
S}
.
1
Connectivity means the graph contains a spanning tree.
1549-7747 © 2014 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission.
See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.