Information Sciences 703 (2025) 121886

Contents lists available at ScienceDirect

Information Sciences
journal homepage: www.elsevier.com/locate/ins

Timer-based distributed coordination for achieving asymptotic

consensus in directed communication networks
,∗
Shu Liu a, , Huaguang Zhang a,b, , Jiayue Sun a,c
a College of Information Science and Engineering, Northeastern University, Shenyang, 110004, China
b State Key Laboratory of Synthetical Automation for Process Industries, Northeastern University, Shenyang, 110004, China
c Collective Intelligence & Collaboration Laboratory, Beijing, 100072, China

A R T I C L E I N F O A B S T R A C T

Keywords: This paper addresses the consensus problem of linear multi-agent systems (MASs) in directed
Consensus graphs using a timer-based event-triggered control algorithm. The proposed distributed algorithm
Directed communication networks allows each agent to update its control law and event monitoring condition using only relative state
Distributed coordination
information from neighboring agents at discrete event instants. This algorithm minimizes reliance
Event-triggered algorithm
Linear multi-agent systems
on global information and significantly reduces communication overhead, thereby enhancing both
efficiency and scalability. A common challenge in event-based control algorithms is the potential
for Zeno behavior, where an i­finite number of events could occur within a finite time, making
the system impractical. While conventional algorithms avoid Zeno behavior by ensuring non­
zero time intervals between events, they often fail to address the issue of excessively short event
intervals. Our algorithm overcomes this limitation by establishing a strictly positive lower bound
for the interval between events for each agent, thereby not only avoiding Zeno behavior but also
ensuring practical applicability and robustness of the control strategy. Through simulation studies,
we validate the efficacy of our algorithm in achieving asymptotic consensus in linear MASs over
directed graphs.

1. Introduction

In the past decade, cooperative control for multi-agent systems (MASs) has become a central focus of research. The significance
of this field is highlighted by the remarkable adaptability and versatility exhibited by groups when handling complex scenarios, as
shown in [1,2]. Cooperative control focuses on solving the consensus problem, where MASs achieve synchronization through inter­
agent information exchange within a communication network [3,4]. This foundational concept extends its practicality to a multitude
of domains, including microgrids [5--7], multiple autonomous surface vehicles [8], and multi-robot coordination [9,10].
However, traditional consensus control algorithms frequently face limitations related to the computational and communication
resources of each agent. Continuous communication among agents is noted as a resource-intensive process, diminishing the efficiency
of MASs [11]. Event-triggered techniques have emerged as a promising solution to these challenges [12]. Event-triggered algorithms,
which initiate pred­fined operations based on specific criteria or events, offer a more efficient alternative to traditional time-triggered
algorithms by avoiding issues of invalid sampling within MASs [13].

* Corresponding author at: College of Information Science and Engineering, Northeastern University, Shenyang, 110004, China.
E-mail address: [email protected] (H. Zhang).

https://doi.org/10.1016/j.ins.2025.121886
Received 14 January 2024; Received in revised form 4 June 2024; Accepted 23 January 2025
Available online 27 January 2025
0020-0255/© 2025 Published by Elsevier Inc.
S. Liu, H. Zhang and J. Sun Information Sciences 703 (2025) 121886

The incorporation of event-triggered algorithms in MAS consensus control, initially introduced by [14], has led to the develop­
ment of various event-based consensus algorithms from different perspectives, as demonstrated in [13,15], and other references. For
instance, event-based consensus strategies have been developed for single and double integrator MASs in works such as [14,16].
The achievement of consensus control for linear MASs subsequently involved the design of appropriate event-triggered strategies, as
detailed in the works of [17--21]. Notably, these event-based algorithms incorporate information about the structure of MAS com­
munication networks, such as the minimum non-zero eigenvalue of Laplacian matrices or agent scale. This information is commonly
referred to as global information. However, the reliance on global information means these consensus algorithms cannot be considered
truly distributed. Moreover, computing eigenvalues for high-dimensional matrices consumes significant computational resources. This
presents a challenge when handling Laplacian matrix-related information in cases where MASs consist of a large number of agents. To
address this issue, [22,23] proposed fully distributed event-based algorithms for undirected and directed graphs, respectively. These
algorithms use adaptive control techniques to circumvent the i­fluence of global information on achieving consensus. Notably, in
directed graphs, distributed event-based consensus algorithms often require the initial values of adaptive parameters for each agent
to satisfy a un­fied global condition, where all agents initially have access to common parameters. This condition is crucial for es­
tablishing Lyapunov stability in directed graphs, as explained in [23] (Remark 4), [24], and [25]. However, these non-free adaptive
parameters prevent the algorithms from operating in a truly distributed manner, thereby limiting the scalability of MASs, as discussed
in [26] (Remark 5).
The assessment of designed event-triggered consensus algorithms often revolves around their ability to eliminate Zeno behavior,
ensuring that the time gap between consecutive events remains greater than zero. This criterion is emphasized in the work of [15]. Tra­
ditional algorithms for ensuring the absence of Zeno behavior in event-triggered algorithms involve providing rigorous mathematical
proofs that guarantee a positive time gap between any two consecutive events (see, e.g., [3,22,23,25--27]). This implies that event­
triggering intervals can potentially become remarkably close to each other. It’s important to emphasize that these event-triggered
algorithms are typically executed on electronic devices with a fixed minimum operational frequency, such as those in software with
fixed step sizes. In scenarios where the required event response speed is greater than zero but falls below the electronic device’s
minimum sampling frequency, it is important to recognize the potential challenges faced by event-based algorithms, as mentioned
in [28]. In such cases, even without Zeno behavior, these algorithms may miss events or degrade overall system performance due to
the electronic devices’ inability to process data quickly enough. This challenge is a shared concern among dynamic event-triggered
strategies, as highlighted in Remark 9 of [23]. To address this issue, one potential strategy involves substituting the dynamic threshold
within the event-triggering condition with a small positive constant. However, this adaptation comes with a trade-off it ensures only
that consensus errors remain bounded, rather than achieving the desired asymptotic convergence. For an in-depth exploration of this
approach, please consult the works of [22] (Remark 6), [23] (Remark 9), [29] (Remark 3).
Initially, [30] introduced timer-based event-driven algorithms for single-integrator systems under undirected graphs. These al­
gorithms were designed to handle situations where adjacent event intervals can be arbitrarily close, effectively eliminating Zeno
behavior. Furthermore, these results were extended to linear MASs by [31]. It’s essential to underline that traditional timer-based
event-driven algorithms involve the reciprocal of measurement error (see, e.g., (15) in [5]; (8) in [31]; (11) in [30]). In these al­
gorithms, measurement errors are reset to zero at each event and increment non-strictly until the next event. This indicates that
agents need to perform real-time computations involving massive numerical values at the right-hand limit of each event to determine
whether event conditions are met. However, this process may lead to arithmetic ove­flow errors, which occur when computed values
exceed the numerical range that the electronic device can represent. This can lead to program instability, reduced performance, or
inaccurate results. Notably, all these timer-based event-driven works are based on undirected graphs. In practical MASs, information
exchange between agents is often one-sided due to variations in sensing and communication capabilities [32]. Therefore, considering
directed graphs is a more realistic and reasonable approach. As highlighted by [15,23,25,27], extending undirected graph results
to directed graphs poses significant challenges. The primary reason is that the Laplacian matrix of directed graphs lacks symmetry
compared to undirected graphs. Furthermore, asymmetry complicates the development of distributed event protocols, as it hinders
the application of Abel transformation for constructing error-based event monitoring conditions (see, e.g., (9) in [22]; (A.5) in [31]).
This research addresses the consensus control challenge in linear multi-agent systems (MASs) within directed graphs. The motiva­
tion stems from the need to overcome practical limitations of existing event-based algorithms, which often fail when event intervals
become too short relative to electronic device operational frequencies, even in the absence of Zeno behavior. Our goal is to develop
a novel timer-based event-triggered control algorithm that reduces communication overhead and ensures practical applicability. The
primary contributions of this paper can be summarized as follows:

• Avoiding I­finitesimal Event Intervals. Our algorithm constructs bounded event descent rates to eliminate Zeno behavior,
ensuring that adjacent event intervals for each agent cannot be arbitrarily close. Remarkably, we achieve asymptotic consensus
without sacr­ficing a distinct positive lower bound on event intervals, addressing a common concern present in other studies
[22,23].
• Flexible Adaptive Parameter Initialization. Unlike conventional distributed event-based consensus protocols, our algorithm
allows agents to autonomously select their initial adaptive parameters, enhancing scalability and enabling true distributed oper­
ation. This unique feature distinguishes our algorithm from constrained algorithms [23--25].
• Applicability to Directed Graphs. Our algorithm, incorporating adaptive control and eliminating the need for global infor­
mation, is specifically designed for general directed graphs. Unlike consensus protocols tailored for undirected graphs (e.g.,
[22,26,29,31]), our algorithm resolves the asymmetry introduced by directed communication pathways.

2
S. Liu, H. Zhang and J. Sun Information Sciences 703 (2025) 121886

• Avoiding Arithmetic Ove­flow Errors. Our novel timer-based event-triggered algorithm significantly reduces the magnitude of
generated numerical values compared to traditional algorithms, effectively preventing arithmetic ove­flow errors. This feature
enhances the stability and reliability of our control algorithm.

The structure of this paper is as follows: Section 2 introduces the preliminary and problem statement, covering essential topics
such as graph theory, the mathematical model of linear MASs, our fundamental assumption, and lemma. In Section 3, we present the
main results of this study and provide the theoretical proofs. Section 4 offers a simulation example, demonstrating the effectiveness
of our algorithm. Finally, Section 5 provides the conclusion for this paper.

2. Preliminaries and problem statement

2.1. Notation

Let ℝ, ℝ>0 , and ℤ≥0 denote the set of real, positive real, and positive integer numbers, respectively. ℝ𝑛 and ℝ𝑛×𝑚 represent
𝑛-dimensional real vectors and 𝑛 × 𝑚 dimensional real matrices, respectively. 𝐼𝑛 is the 𝑛 × 𝑛 identity matrix. The symbol ‖ ⋅ ‖ denotes
the Euclidean norm, while ⊗ sign­fies the Kronecker product. For a symmetric matrix 𝑋 , 𝜆2 (𝑋) and 𝜆𝑚𝑎𝑥 (𝑋) represent the non-zero
minimum and maximum eigenvalues of matrix 𝑋 , respectively. diag{⋯} represents a block-diagonal matrix. For a function 𝑓 , we
d­fine 𝑓 (𝑡+ ) = lim𝜄→𝑡+ 𝑓 (𝜄).

2.2. Graph theory

The communication network consists of N agents and is represented as a directed graph denoted by  = ( , ), where  =
{1, ..., 𝑁} represents the nodes corresponding to individual agents, and  ⊆  ×  represents the set of directed edges indicating
the communication pathways. An edge (𝑖, 𝑗) ∈  sign­fies a one-way communication link from agent 𝑖 to agent 𝑗 , indicating that
agent 𝑗 can receive information from agent 𝑖. The direction of information flow among agents is d­fined by the adjacency matrix
 = [𝑎𝑖𝑗 ] ∈ ℝ𝑁×𝑁 , 𝑖, 𝑗 = 1, … , 𝑁 , where 𝑎𝑖𝑗 = 1(𝑖 ≠ 𝑗) if (𝑗, 𝑖) ∈  and 𝑎𝑖𝑗 = 0 otherwise. Additionally, it is assumed that (𝑖, 𝑖) ∉  . The
∑𝑁
Laplacian matrix of  is denoted as  = [𝑙𝑖𝑗 ] ∈ ℝ𝑁×𝑁 , 𝑖, 𝑗 = 1, … , 𝑁 , where 𝑙𝑖𝑖 = 𝑗=1 𝑎𝑖𝑗 and 𝑙𝑖𝑗 = −𝑎𝑖𝑗 for 𝑖 ≠ 𝑗 . A directed network
 is considered strongly connected if there exists a directed path between any pair of agents.

2.3. Mathematical model of linear MASs

Consider a linear MASs consisting of N agents (N ≥ 2). The dynamics of each agent are described as below:

𝑥̇ 𝑖 (𝑡) = 𝐴𝑥𝑖 (𝑡) + 𝐵𝑢𝑖 (𝑡), 𝑖 = 1, ..., 𝑁, (1)

where 𝑥𝑖 (𝑡) ∈ ℝ𝑛
and 𝑢𝑖 (𝑡) ∈ ℝ𝑚
are the state and control input of the i-th agent, respectively. 𝐴 ∈ ℝ𝑛×𝑛
and 𝐵 ∈ ℝ𝑛×𝑚
represent
the known real constant matrices, and the pair (𝐴, 𝐵) is stabilizable. By the well-known algebraic Riccati equation, there exists a
symmetric positive definite matrix 𝑃 satisfying:

𝐴𝑇 𝑃 + 𝑃 𝐴 − 𝑃 𝐵𝐵 𝑇 𝑃 + 𝐼𝑛 = 0. (2)
This paper has two control objectives. Firstly, it aims to devise a novel timer-based distributed event-triggered algorithm to achieve
asymptotic consensus in linear MASs under general directed graphs. The second objective is for the designed algorithm to eliminate
Zeno behavior while avoiding i­finitesimally small event intervals.

Definition 1. Asymptotic consensus for linear MASs (1) is considered achieved if, for any initial values 𝑥𝑖 (𝑡0 ) ∈ ℝ𝑛 :

lim ‖𝑥𝑖 (𝑡) − 𝑥𝑗 (𝑡)‖ = 0, 𝑖, 𝑗 ∈  . (3)

𝑡→+∞

Definition 2. For linear MASs (1), each agent can prevent triggering intervals from becoming arbitrarily close by satisfying the
following condition:

min 𝑡𝑖 − 𝑡𝑖𝑘 ≥ 𝜏𝑖 > 0, 𝑖 ∈ , (4)

𝑘∈ℤ≥0 𝑘+1

where 𝑡𝑖𝑘 represents the k-th event triggering moment for agent 𝑖, and 𝜏𝑖 ∈ ℝ denotes the positive minimum time interval for agent 𝑖.

Remark 1. The presence of Zeno behavior implies that the number of events triggered in a finite time approaches i­finity. The absence
of Zeno behavior has become a crucial criterion for validating the correctness of event-based algorithms, as observed in works such
as [22,23,25,26]. However, practical implementations of these algorithms can pose potential issues. Although it can be theoretically
proven that event intervals are positive (𝑡𝑖𝑘+1 − 𝑡𝑖𝑘 > 0), the adjacent intervals may still be arbitrarily small (𝑡𝑖𝑘+1 − 𝑡𝑖𝑘 → 0+ ). This
situation may cause difficulties for physical devices with finite operating speeds to meet the response speed required by the algorithms.

3
S. Liu, H. Zhang and J. Sun Information Sciences 703 (2025) 121886

As noted by [23], this issue has become a common limitation of dynamic event triggering. Consequently, Definition 2 offers a more
realistic perspective than merely excluding Zeno behavior.

Assumption 1. The communication graph  is directed, and strongly connected.

Remark 2. Assumption 1 plays a pivotal role in addressing the leaderless consensus problem within directed graphs, as emphasized
by previous research [23,33]. It’s worth noting that undirected graphs [31,22,26] and weight-balanced graphs [28,15] represent
specific instances of strongly connected graphs. Therefore, Assumption 1 can be regarded as a more relaxed and universally applicable
condition. By stipulating that the graph is strongly connected, this assumption ensures that within the MASs, every agent possesses
a directed pathway for communication with any other agent. This condition forms the bedrock of the consensus framework in this
research domain.

Lemma 1 (see [34]). Under Assumption 1, the Laplacian matrix  of the graph  possesses a unique positive left eigenvector 𝑟 ∈ ℝ𝑛
represented as 𝑟 ≜ [𝑟1 , ..., 𝑟𝑁 ]𝑇 . This eigenvector is associated with zero eigenvalues and sati­fies the conditions 𝑟𝑇  = 0 and 𝟏𝑇 𝑟 = 1. Let
𝑅 ≜ diag{𝑟1 , ..., 𝑟𝑁 } and consider 𝜍 ∈ ℝ𝑛 as any positive vector. Then, the following inequality holds for any vector 𝑥 ∈ ℝ𝑛 :
𝜆2 (𝑅 + 𝑇 𝑅)
min 𝑥𝑇 (𝑅 + 𝑇 𝑅)𝑥 > > 0.
𝜍 𝑇 𝑥=0,𝑥𝑇 𝑥=1 𝑁

3. Timer-based event-triggered algorithm

This section introduces a novel distributed algorithm leveraging timer-based event-triggered mechanisms to achieve asymptotic
consensus in directed graph-based MASs. The algorithm includes an adaptive event-triggered control law designed to minimize
communication and computational overhead by dynamically adjusting control inputs based on consensus error measurements. Addi­
tionally, we incorporate a precise timer-based event monitoring condition rooted in Lyapunov stability analysis, acting as a countdown
timer with well-defined upper and lower limits to trigger communication updates strategically. We establish a strictly positive lower
bound on the triggering interval for each agent, ensuring that communication updates occur at appropriate intervals to maintain
network stability and convergence towards consensus.

3.1. Adaptive event-triggered control law

This subsection introduces an adaptive event-triggered control law designed to achieve asymptotic consensus in directed graphs.
This control law is based on the state error of each agent and dynamically adjusts control inputs to reduce communication and
computational load. For agent 𝑖, we d­fine consensus error 𝜁𝑖 (𝑡) as follows:
𝑁
∑ ( )
𝜁𝑖 (𝑡) ≜ 𝑎𝑖𝑗 𝑥𝑖 (𝑡) − 𝑥𝑗 (𝑡) , 𝑖 ∈ . (5)
𝑗=1

Rewrite (5) in a compact form as

( )
𝜁(𝑡) =  ⊗ 𝐼𝑛 𝑥(𝑡), (6)
where 𝜁(𝑡) = [𝜁1𝑇 (𝑡), ..., 𝜁𝑁
𝑇 (𝑡)]𝑇
and 𝑥(𝑡) = [𝑥𝑇1 (𝑡), ..., 𝑥𝑇𝑁 (𝑡)]𝑇 .
Note that the Laplacian matrix  has a unique zero eigenvalue, and
the corresponding eigenvector is 𝟏. It’s evident that 𝜁(𝑡) = 0 if and only if 𝑥1 (𝑡) = ⋯ = 𝑥𝑁 (𝑡). Therefore, when 𝜁(𝑡) asymptotically
converges to zero, the states in (1) asymptotically achieve consensus.
The control law, 𝑢𝑖 (𝑡), is given by:
( )
𝑢𝑖 (𝑡) = 𝑐𝑖 (𝑡) + 𝑑̂𝑖 (𝑡) 𝐾 𝜁̂𝑖 (𝑡), (7)

where 𝜁̂𝑖 (𝑡) = 𝜁𝑖 (𝑡𝑖𝑘 ) and 𝑑̂𝑖 (𝑡) = 𝑑𝑖 (𝑡𝑖𝑘 ) for 𝑡 ∈ [𝑡𝑖𝑘 , 𝑡𝑖𝑘+1 ). The dynamics of 𝑐𝑖 (𝑡) are governed by the equation:

𝑐̇ 𝑖 (𝑡) = 𝜅𝑖 𝜁̂𝑖𝑇 (𝑡)Γ𝜁̂𝑖 (𝑡), 𝑐𝑖 (0) > 0, (8)

where 𝜅𝑖 is a positive constant, and 𝑐𝑖 (𝑡) accumulates the effects of the consensus error 𝜁(𝑡) on the control gain. The term 𝑑𝑖 (𝑡) =
𝜁𝑖𝑇 (𝑡)𝑃 𝜁𝑖 (𝑡) represents the instantaneous effect of the consensus error. The matrices 𝐾 and Γ are determined as 𝐾 = −𝐵 𝑇 𝑃 and
Γ = 𝐾 𝑇 𝐾 , with 𝑃 obtained from the algebraic Riccati equation in (2).

Remark 3. The selection of the initial value for 𝑐𝑖 (𝑡) is no longer bound by global variables, offering the freedom for flexible choices
without common parameter constraints (𝑐𝑖 (0) can assume any positive value). In other words, agents are no longer required to access
common parameters at the initial moment, distinguishing this approach from previous works on directed graphs (as noted in Remark
4 of [23], and [24]). This flexibility also means that the design of 𝑐𝑖 (𝑡) no longer depends on global information (global variables).
Consequently, it demonstrates notable scalability and plug-and-play capability.

4
S. Liu, H. Zhang and J. Sun Information Sciences 703 (2025) 121886

3.2. Timer-based event monitoring condition

This subsection explores the intricacies of a timer-based event monitoring condition used to determine when to trigger events for
updating control inputs. Specifically, we d­fine the monitoring criteria and its i­fluencing factors to ensure events are triggered at
appropriate times, thus avoiding excessive communication and computational overhead.
For agent 𝑖, the measurement error is d­fined as:

𝑒𝑖 (𝑡) ≜ 𝜁̂𝑖 (𝑡) − 𝜁𝑖 (𝑡), 𝑡 ∈ [𝑡𝑖𝑘 , 𝑡𝑖𝑘+1 ). (9)

The event sequence {𝑡𝑖𝑘 }𝑘∈ℤ≥0 for agent 𝑖 is determined by

{ }
𝑡𝑖𝑘+1 = inf 𝑡 > 𝑡𝑖𝑘 | 𝜓𝑖 (𝑡) ≤ 0 ∧ 𝑒𝑖 (𝑡) ≠ 0 , (10)
where the function 𝜓𝑖 sati­fies the differential equation:
[ ]2
𝜓̇ 𝑖 (𝑡) = − 𝜂𝑖 (𝑡) − 𝜓𝑖2 (𝑡) − 𝑐𝑖 (𝑡) + 𝑑𝑖 (𝑡) − [𝑑𝑖 (𝑡) + 𝑑̂𝑖 (𝑡)]𝑐̇ 𝑖 (𝑡) (11)

with the initial condition 𝜓𝑖 (𝑡𝑖+

𝑘
) = 𝛾𝑖 ∈ ℝ>0 . Moreover, the time-varying adaptive parameter 𝜂𝑖 ∈ ℝ>0 is governed by 𝜂̇ 𝑖 (𝑡) = 𝜎𝑖 ‖𝑒𝑖 (𝑡)‖2 ,
and 𝜂𝑖 (0) > 0 for any 𝜎𝑖 ∈ ℝ>0 .

Remark 4. The expression in (11) indicates that 𝜓̇ 𝑖 (𝑡) < 0, which implies that the event monitoring condition, 𝜓𝑖 (𝑡), acts as a timer
with well-defined upper and lower limits. It triggers an event when counting down from the initial value of 𝛾𝑖 to zero. Importantly,
after each event triggering, 𝜓𝑖 (𝑡) is promptly reset to 𝛾𝑖 , ensuring the condition 0 ≤ 𝜓𝑖 (𝑡) ≤ 𝛾𝑖 is maintained throughout the process.
The design of 𝜓̇ 𝑖 (𝑡) is rooted in Lyapunov stability analysis, and its inclusion guarantees the asymptotic convergence of the consensus
error 𝜁(𝑡) to zero.

1
Remark 5. Traditional timer-based event triggering algorithms involve expressions such as (as seen in (15) of [5]; (8) of [31];
‖𝑒𝑖 (𝑡)‖2
(11) of [30]). In these algorithms, the measurement error 𝑒𝑖 (𝑡) is reset to zero at each event moment and then gradually increases
until the next event moment. This essentially compels the agent to execute computationally demanding operations at the right limit
of each event instant to determine whether the event condition is sati­fied. This algorithm carries a substantial risk of encountering
arithmetic ove­flow errors, potentially resulting in program crashes, reduced performance, or inaccurate outcomes. In contrast, the
proposed timer-based event monitoring condition in (11) effectively mitigates this prevalent issue.

The principal objective of employing event-triggered algorithms in MASs is to reduce the need for continuous communication
among agents. However, it’s worth noting that 𝑑𝑖 (𝑡) in (11) demands constant monitoring of the relative information between agent
𝑖 and its neighbors. To address this limitation, we turn to the consensus error prediction model introduced in [23] to establish the
following predictive mechanism:
𝑁
∑
𝜁̇ 𝑖 (𝑡) = 𝑙𝑖𝑗 𝑥̇ 𝑗 (𝑡) = 𝐴𝜁𝑖 (𝑡) + 𝑙𝑖𝑖 [𝑐𝑖 (𝑡) + 𝑑̂𝑖 (𝑡)]𝐵𝐾 𝜁̂𝑖 (𝑡)
𝑗=1
𝑁
∑
− 𝐵𝐾 𝑎𝑖𝑗 [𝑐𝑗 (𝑡) + 𝑑̂𝑗 (𝑡)]𝜁̂𝑗 (𝑡),
𝑗=1

for 𝑡 ∈ [𝑡𝑖𝑘 , 𝑡𝑖𝑘+1 ). Hence, we can derive a r­fined expression for the consensus error 𝜁𝑖 (𝑡) as follows:

⎡ 𝑡 ⎤
⎢ 𝐴(𝑡−𝑡𝑖 ) ̂ ⎥
𝜁𝑖 (𝑡) = ⎢𝑒 𝑘 +𝑙
𝑖𝑖 ∫ [𝑐𝑖 (𝜎) + 𝑑𝑖 (𝜎)]𝑒
𝐴(𝑡−𝜎)
𝐵𝐾𝑑𝜎 ⎥ 𝜁̂𝑖 (𝑡)
⎢ ⎥
⎣ 𝑡𝑖𝑘 ⎦
𝑡
𝑁
∑
− 𝑒𝐴(𝑡−𝜎) 𝐵𝐾 𝑎𝑖𝑗 [𝑐𝑗 (𝜎) + 𝑑̂𝑗 (𝜎)]𝜁̂𝑗 (𝜎)𝑑𝜎. (12)
∫
𝑗=1
𝑡𝑖𝑘

While the event monitoring condition (11) requires agent 𝑖 to continuously collect relative state information from its neighbors, this
challenge is addressed by utilizing (12) to predict 𝑑𝑖 (𝑡). When computing the consensus error 𝜁𝑖 (𝑡) using (12), the update mechanism
for (11) unfolds as follows: i) At the event instants of agent 𝑖, 𝑐̇ 𝑖 (𝑡) and 𝑑̂𝑖 (𝑡) are updated by measuring relative information with its
neighbors. ii) At the event instants of agent 𝑗 , where (𝑗, 𝑖) ∈  , 𝑑𝑖 (𝑡) is updated by receiving 𝜁̂𝑗 (𝑡) from agent 𝑗 . Thus, the use of (12)
ensures that the proposed event-triggered algorithm operates with discontinuous communication.
To facilitate the subsequent derivations, we introduce the following variables: 𝑐(𝑡) = diag{𝑐1 (𝑡), … , 𝑐𝑁 (𝑡)}, 𝑑(𝑡) = diag{𝑑1 (𝑡), … , 𝑑𝑁 (𝑡)},
̂ = diag{𝑑̂1 (𝑡), … , 𝑑̂𝑁 (𝑡)}, 𝜁(𝑡)
𝑑(𝑡) ̂ = [𝜁̂𝑇 (𝑡), … , 𝜁̂𝑇 (𝑡)]𝑇 , 𝑒(𝑡) = [𝑒𝑇 (𝑡), … , 𝑒𝑇 (𝑡)]𝑇 , and 𝜓(𝑡) = diag{𝜓1 (𝑡), … , 𝜓𝑁 (𝑡)}.
1 𝑁 1 𝑁

5
S. Liu, H. Zhang and J. Sun Information Sciences 703 (2025) 121886

By substituting (7) into (1), we derive the subsequent closed-loop dynamics:

𝜁̇ (𝑡) = (𝐼𝑁 ⊗ 𝐴)𝜁(𝑡) + [(𝑐(𝑡) + 𝑑(𝑡))

̂ ⊗ 𝐵𝐾]𝜁(𝑡)
̂ (13)
Consider the ascending sequence {𝑡𝑙 }𝑙∈ℤ>0 , comprising the event instances of all agents. For 𝑡 ∈ [𝑡𝑙 , 𝑡𝑙+1 ), as deduced from (9) and
(13), the following relationship emerges:

𝑒(𝑡) ̂ ⊗ 𝐵𝐾]𝜁(𝑡)
̇ = −(𝐼𝑁 ⊗ 𝐴)𝜁(𝑡) − [(𝑐(𝑡) + 𝑑(𝑡)) ̂ (14)

3.3. Consensus analysis

This subsection conducts a consensus analysis by employing the adaptive event-triggered control law introduced in Section 3.1
and the event monitoring condition d­fined in Section 3.2. By appropriately selecting control parameters, we demonstrate that the
algorithm can achieve asymptotic consensus in directed graphs. Please note that, for the sake of brevity, we will omit the explicit
time dependence of symbols in the forthcoming Lyapunov stability analysis.

Theorem 1. Consider the linear MASs described by (1) on the graph . Assuming that Assumption 1 is sati­fied, and the adaptive control
law given by (7) along with the event monitoring condition as spec­fied in (10) is implemented, the following properties hold:
i) The consensus error 𝜁(𝑡) converges asymptotically to 0. The adaptive parameters 𝑐𝑖 (𝑡) and 𝜂𝑖 (𝑡) will converge to finite values.
ii) Any consecutive events for agent 𝑖 are separated by strictly positive minimum time intervals, denoted as 𝜏𝑖 , given by:
( )
1 𝛾𝑖
𝜏𝑖 = √ arctan √ > 0. (15)
𝜇𝑖 𝜇𝑖

Here, 𝜇𝑖 = (𝑐𝑖𝑚 + 𝑑𝑖𝑚 )2 + 2𝑑𝑖𝑚 𝑐̇ 𝑖𝑚 + 𝜂𝑖𝑚 , where 𝑐𝑖𝑚 = max𝑡∈[0,∞) 𝑐𝑖 (𝑡) > 0, 𝑑𝑖𝑚 = max𝑡∈[0,∞) 𝑑𝑖 (𝑡) > 0, and 𝑐̇ 𝑖𝑚 = max𝑡∈[0,∞) 𝑐̇ 𝑖 (𝑡) > 0, for
𝑖 = 1, … , 𝑁 .

Proof. Let
𝑁 𝑁
𝐴∗ ∑ ∑ (𝜂𝑖 − 𝐵∗ )2
𝑉𝜓 = 𝜓𝑖 𝑒𝑇𝑖 𝑒𝑖 + , (16)
2 𝑖=1 𝑖=1
4𝜎𝑖
where 𝐴∗ and 𝐵∗ are positive constants, designed later. Taking the time derivative of 𝑉𝜓 for 𝑡 ∈ [𝑡𝑙 , 𝑡𝑙+1 ), we obtain

𝑁 𝑁 𝑁
𝐴∗ ∑ ∑ 1 ∑
𝑉̇ 𝜓 = 𝜓̇ 𝑖 𝑒𝑇𝑖 𝑒𝑖 + 𝐴∗ 𝜓𝑖 𝑒𝑇𝑖 𝑒̇ 𝑖 + (𝜂 − 𝐵∗ )𝜂̇ 𝑖
2 𝑖=1 𝑖=1
2𝜎𝑖 𝑖=1 𝑖
𝑁
𝐴∗ ∑
= 𝜓̇ 𝑒𝑇 𝑒 − 𝐴∗ 𝑒𝑇 (𝜓 ⊗ 𝐼𝑛 )(𝐼𝑁 ⊗ 𝐴)𝜁
2 𝑖=1 𝑖 𝑖 𝑖
[ ] ∑𝑁 𝑁
𝐵 ∑ 𝑇
̂ ⊗ 𝐵𝐾 𝜁̂ + 1
− 𝐴∗ 𝑒𝑇 (𝜓 ⊗ 𝐼𝑛 ) (𝑐 + 𝑑) 𝜂𝑖 𝑒𝑇𝑖 𝑒𝑖 − ∗ 𝑒 𝑒. (17)
2 𝑖=1 2 𝑖=1 𝑖 𝑖
Using the well-known Young’s inequality and considering the fact that 0 ≤ 𝜓𝑖 ≤ 𝛾𝑖 , we can proceed to transform the second part of
(17) as follows:

− 𝐴∗ 𝑒𝑇 (𝜓 ⊗ 𝐼𝑛 )(𝐼𝑁 ⊗ 𝐴)𝜁
𝑁
1 ∑ 2 1
≤ 𝐴2∗ 𝜓𝑖 ‖𝑒𝑖 ‖2 + 𝜆𝑚𝑎𝑥 (𝐴𝑇 𝐴)𝜁 𝑇 𝜁
2 𝑖=1
2

∑ 𝑁 𝑁
1 1∑ 2 1
≤ (𝐴2∗ − 1)𝛾 2 ‖𝑒𝑖 ‖2 + 𝜓𝑖 ‖𝑒𝑖 ‖2 + 𝜆𝑚𝑎𝑥 (𝐴𝑇 𝐴)𝜁 𝑇 𝜁, (18)
2 𝑖=1
2 𝑖=1
2

where 𝐴∗ ≥ 1 and 𝛾 = max{𝛾1 , … , 𝛾𝑁 }. For any positive constant 𝜋1 , we can transform the third part of (17) as follows using Young’s
inequality.
[ ]
̂ ⊗ 𝐵𝐾 𝜁̂
− 𝐴∗ 𝑒𝑇 (𝜓 ⊗ 𝐼𝑛 ) (𝑐 + 𝑑)

𝜋1 2 2 ∑
𝑁 [ ]
1 ̂𝑇 ̂ 𝑇 (𝑐 + 𝑑)
̂ ⊗ 𝐾 𝑇 𝐵 𝑇 𝐵𝐾 𝜁.
̂
≤ 𝐴∗ 𝛾 ‖𝑒𝑖 ‖2 + 𝜁 (𝑐 + 𝑑) (19)
2 𝑖=1
2𝜋1
Note that the second part of (19) can be transformed as follows.

6
S. Liu, H. Zhang and J. Sun Information Sciences 703 (2025) 121886
[ ]
1 ̂𝑇 ̂ 𝑇 (𝑐 + 𝑑)
̂ ⊗ 𝐾 𝑇 𝐵 𝑇 𝐵𝐾 𝜁̂
𝜁 (𝑐 + 𝑑)
2𝜋1
1 [ ]
≤ 𝜆𝑚𝑎𝑥 (𝑇 )𝜆𝑚𝑎𝑥 (𝐵 𝑇 𝐵)𝜁̂𝑇 (𝑐 + 𝑑)2 ⊗ Γ 𝜁̂
𝜋1
[ ]
1
+ 𝜆𝑚𝑎𝑥 (𝑇 )𝜆𝑚𝑎𝑥 (𝐵 𝑇 𝐵)𝜁̂𝑇 (𝑑 − 𝑑) ̂
̂ 2 ⊗ Γ 𝜁. (20)
𝜋1
Since 𝜁̂ = 𝜁 + 𝑒, we can verify that
[ ]
𝜁̂𝑇 (𝑐 + 𝑑)2 ⊗ Γ 𝜁̂
𝑁
∑
[ ]
≤ 2𝜁 𝑇 (𝑐 + 𝑑)2 ⊗ Γ 𝜁 + 2𝜆𝑚𝑎𝑥 (Γ) (𝑐𝑖 + 𝑑𝑖 )2 ‖𝑒𝑖 ‖2 . (21)
𝑖=1

Utilizing the definitions of 𝑑𝑖 , 𝑑̂𝑖 , and 𝑐̇ 𝑖 , we can apply Young’s inequality to transform the second part of inequality (20) as follows.
[ ]
̂ 2 ⊗ Γ 𝜁̂
𝜁̂𝑇 (𝑑 − 𝑑)
𝑁 [
∑ √ √ √ ]2
= 𝑒𝑇𝑖 𝑃 ( 𝑃 𝜁𝑖 + 𝑃 𝜁̂𝑖 ) 𝜁̂𝑖𝑇 Γ𝜁̂𝑖
𝑖=1
𝑁
∑ (√ √ √ √ )
≤ 𝜆𝑚𝑎𝑥 ( 𝑃 𝜁𝑖 + 𝑃 𝜁̂𝑖 )(𝜁𝑖𝑇 𝑃 + 𝜁̂𝑖𝑇 𝑃 ) 𝑒𝑇𝑖 𝑃 𝑒𝑖 𝜁̂𝑖𝑇 Γ𝜁̂𝑖
𝑖=1
𝑁
∑ √ √ √ √
= (𝜁𝑖𝑇 𝑃 + 𝜁̂𝑖𝑇 𝑃 )( 𝑃 𝜁𝑖 + 𝑃 𝜁̂𝑖 )𝑒𝑇𝑖 𝑃 𝑒𝑖 𝜁̂𝑖𝑇 Γ𝜁̂𝑖
𝑖=1

∑𝑁
2
≤ 𝜆𝑚𝑎𝑥 (𝑃 ) (𝑑𝑖 + 𝑑̂𝑖 )𝑐̇ 𝑖 ‖𝑒𝑖 ‖2 , (22)
𝜅 𝑖=1

where 𝜅 = min{𝜅1 , … , 𝜅𝑁 }. By substituting (21) and (22) into (20), and then (20) into (19), and finally (19) and (18) into (17), we
obtain the following inequality.
𝑁 𝑁
𝐴∗ ∑ 1∑ 1
𝑉̇ 𝜓 ≤ 𝜓̇ 𝑖 ‖𝑒𝑖 ‖2 + 𝜂 ‖𝑒 ‖2 + 𝜆𝑚𝑎𝑥 (𝐴𝑇 𝐴)𝜁 𝑇 𝜁
2 𝑖=1 2 𝑖=1 𝑖 𝑖 2
𝑁
( ) 𝑁
1∑ 2 𝜋1 + 1 2 2 𝐵 ∗ 𝛾 2 ∑
+ 𝜓 ‖𝑒 ‖ + 2
𝐴∗ 𝛾 − − ‖𝑒 ‖2
2 𝑖=1 𝑖 𝑖 2 2 2 𝑖=1 𝑖
2 [ ]
+ 𝜆 (𝐵 𝑇 𝐵)𝜆𝑚𝑎𝑥 (𝑇 )𝜁 𝑇 (𝑐 + 𝑑)2 ⊗ Γ 𝜁
𝜋1 𝑚𝑎𝑥
∑𝑁
2
+ 𝜆𝑚𝑎𝑥 (𝐵 𝑇 𝐵)𝜆𝑚𝑎𝑥 (𝑇 )𝜆𝑚𝑎𝑥 (Γ) (𝑐𝑖 + 𝑑𝑖 )2 ‖𝑒𝑖 ‖2
𝜋1 𝑖=1

∑𝑁
2
+ 𝜆𝑚𝑎𝑥 (𝐵 𝑇 𝐵)𝜆𝑚𝑎𝑥 (𝑇 )𝜆𝑚𝑎𝑥 (𝑃 ) (𝑑𝑖 + 𝑑̂𝑖 )𝑐̇ 𝑖 ‖𝑒𝑖 ‖2 . (23)
𝜅𝜋1 𝑖=1
Let
𝑁
∑ 𝑁
1 1∑
𝑉𝜁 = 𝑘∗ 𝑟𝑖 (𝑐𝑖 𝑑𝑖 + 𝑑𝑖2 ) + 𝑘 𝑟 (𝑐 − 𝐶∗ )2 , (24)
𝑖=1
2 2 𝑖=1 ∗ 𝑖 𝑖
where 𝑟𝑖 is d­fined in Lemma 1, while 𝑘∗ and 𝐶∗ represent positive constants that will be designed later. The time derivative of 𝑉𝜁
along (13) for 𝑡 ∈ [𝑡𝑙 , 𝑡𝑙+1 ] is expressed as:

𝑁
∑ 𝑁
∑
𝑉̇ 𝜁 = 𝑘∗ 𝑟𝑖 (𝑐̇ 𝑖 𝑑𝑖 + 𝑐𝑖 𝑑̇𝑖 + 𝑑𝑖 𝑑̇𝑖 ) + 𝑘∗ 𝑟𝑖 (𝑐𝑖 − 𝐶∗ )𝑐̇ 𝑖
𝑖=1 𝑖=1
𝑁
∑ 𝑁
∑
= 𝑘∗ 𝑟𝑖 (𝑐𝑖 + 𝑑𝑖 )𝑑̇𝑖 + 𝑘∗ 𝑟𝑖 (𝑐𝑖 + 𝑑𝑖 − 𝐶∗ )𝑐̇ 𝑖
𝑖=1 𝑖=1
[ ] [ ]
̂ ⊗ Γ 𝜁̂
= 𝑘∗ 𝜁 𝑇 𝑅(𝑐 + 𝑑) ⊗ (𝐴𝑇 𝑃 + 𝑃 𝐴) 𝜁 − 2𝑘∗ 𝜁 𝑇 (𝑐 + 𝑑)𝑅(𝑐 + 𝑑)

7
S. Liu, H. Zhang and J. Sun Information Sciences 703 (2025) 121886

𝑁
∑
+ 𝑘∗ 𝜅𝑖 𝑟𝑖 (𝑐𝑖 + 𝑑𝑖 − 𝐶∗ )𝜁̂𝑖𝑇 Γ𝜁̂𝑖 . (25)
𝑖=1

By applying the fact 𝜁̂ = 𝜁 + 𝑒, we obtain:

[ ]
̂ ⊗ Γ 𝜁̂
− 2𝑘∗ 𝜁 𝑇 (𝑐 + 𝑑)𝑅(𝑐 + 𝑑)
= −2𝑘∗ 𝜁 𝑇 [(𝑐 + 𝑑)𝑅(𝑐 + 𝑑) ⊗ Γ] 𝜁 − 2𝑘∗ 𝜁 𝑇 [(𝑐 + 𝑑)𝑅(𝑐 + 𝑑) ⊗ Γ] 𝑒
[ ]
̂ ⊗ Γ 𝜁.
+ 2𝑘∗ 𝜁 𝑇 (𝑐 + 𝑑)𝑅(𝑑 − 𝑑) ̂ (26)

Combining the definitions of 𝜁 with the well-known Spectral theorem [40], we arrive at:
[ √ ][ ]
𝜁 𝑇 (𝑐 + 𝑑) ⊗ Γ (𝑐 + 𝑑)−1 𝑟 ⊗ 𝐼𝑛
√
= 𝑥𝑇 (𝑇 𝑟 ⊗ Γ)
= 0.
As all elements in (𝑐 + 𝑑)−1 𝑟 are positive, we can proceed to transform the first part of (26) based on Lemma 1.

− 2𝑘∗ 𝜁 𝑇 [(𝑐 + 𝑑)𝑅(𝑐 + 𝑑) ⊗ Γ] 𝜁

𝑘∗ 𝜆2 (𝑅 + 𝑇 𝑅) 𝑇 [ ]
≤− 𝜁 (𝑐 + 𝑑)2 ⊗ Γ 𝜁. (27)
𝑁
Utilizing Young’s inequality, we find positive values 𝜋2 and 𝜋3 that allow for the transformation of the second and third parts of (26)
as follows.

− 2𝑘∗ 𝜁 𝑇 [(𝑐 + 𝑑)𝑅(𝑐 + 𝑑) ⊗ Γ] 𝑒

∑𝑁
[ ] 1
≤ 𝜋2 𝑘2∗ 𝜆𝑚𝑎𝑥 (𝑅𝑇 𝑅)𝜁 𝑇 (𝑐 + 𝑑)2 ⊗ Γ 𝜁 + 𝜆𝑚𝑎𝑥 (Γ) (𝑐𝑖 + 𝑑𝑖 )2 ‖𝑒𝑖 ‖2 , (28)
𝜋2 𝑖=1

and
[ ]
̂ ⊗ Γ 𝜁̂
2𝑘∗ 𝜁 𝑇 (𝑐 + 𝑑)𝑅(𝑑 − 𝑑)
[ ] [ ]
1 ̂ 2 ⊗ Γ 𝜁̂
≤ 𝜋3 𝑘2∗ 𝜆𝑚𝑎𝑥 (𝑅𝑇 𝑅)𝜁 𝑇 (𝑐 + 𝑑)2 ⊗ Γ 𝜁 + 𝜁̂𝑇 (𝑑 − 𝑑)
𝜋3
∑𝑁
[ ] 2
≤ 𝜋3 𝑘2∗ 𝜆𝑚𝑎𝑥 (𝑅𝑇 𝑅)𝜁 𝑇 (𝑐 + 𝑑)2 ⊗ Γ 𝜁 + 𝜆𝑚𝑎𝑥 (𝑃 ) (𝑑𝑖 + 𝑑̂𝑖 )𝑐̇ 𝑖 ‖𝑒𝑖 ‖2 . (29)
𝜅𝜋3 𝑖=1

Note that the third part of (25) can be transformed as follows.

𝑁
∑
𝑘∗ 𝜅𝑖 𝑟𝑖 (𝑐𝑖 + 𝑑𝑖 − 𝐶∗ )𝜁̂𝑖𝑇 Γ𝜁̂𝑖
𝑖=1
𝑁
∑
= 𝑘∗ 𝜅𝑖 𝑟𝑖 (𝑐𝑖 + 𝑑𝑖 )(𝑒𝑇𝑖 + 𝜁𝑖𝑇 )Γ(𝑒𝑖 + 𝜁𝑖 )
𝑖=1
𝑁 [ ]
∑ 1 ̂𝑇 1
+ 𝑘∗ 𝜅𝑖 𝑟𝑖 𝐶∗ (𝜁 − 𝑒𝑇𝑖 )Γ(𝜁̂𝑖 − 𝑒𝑖 ) − 𝜁𝑖𝑇 Γ𝜁𝑖 − 𝜁̂𝑖𝑇 Γ𝜁̂𝑖
𝑖=1
2 𝑖 2
𝑁
∑ 𝑁
∑
≤2 𝑘∗ 𝜅𝑖 𝑟𝑖 (𝑐𝑖 + 𝑑𝑖 )𝑒𝑇𝑖 Γ𝑒𝑖 + 2 𝑘∗ 𝜅𝑖 𝑟𝑖 (𝑐𝑖 + 𝑑𝑖 )𝜁𝑖𝑇 Γ𝜁𝑖
𝑖=1 𝑖=1
𝑁
∑ 𝑁
1∑
+ 𝑘∗ 𝜅𝑖 𝑟𝑖 𝐶∗ 𝑒𝑇𝑖 Γ𝑒𝑖 − 𝑘 𝜅 𝑟 𝐶 𝜁 𝑇 Γ𝜁
𝑖=1
2 𝑖=1 ∗ 𝑖 𝑖 ∗ 𝑖 𝑖
𝑁
∑ 𝑁
∑
≤ 𝜆𝑚𝑎𝑥 (Γ) (𝑐𝑖 + 𝑑𝑖 )2 ‖𝑒𝑖 ‖2 + 𝑘2∗ 𝜅 2 𝑟2 𝜆𝑚𝑎𝑥 (Γ) ‖𝑒𝑖 ‖2
𝑖=1 𝑖=1
𝑁
∑ 𝜅 ( )
+ 2𝑘∗ 𝜅𝜁 𝑇 [𝑅(𝑐 + 𝑑) ⊗ Γ]𝜁 + 𝑘∗ 𝜅𝐶∗ 𝑟𝜆𝑚𝑎𝑥 (Γ) ‖𝑒𝑖 ‖2 − 𝑘∗ 𝑟𝐶∗ 𝜁 𝑇 𝐼𝑁 ⊗ Γ 𝜁, (30)
𝑖=1
2

8
S. Liu, H. Zhang and J. Sun Information Sciences 703 (2025) 121886

where 𝜅 = max{𝜅1 , … , 𝜅𝑁 }, 𝑟 = max{𝑟1 , … , 𝑟𝑁 }, and 𝑟 = min{𝑟1 , … , 𝑟𝑁 }. Substituting (27)-(29) into (26), and then substituting (26)
and (30) into (25), we obtain:
[ ] 𝑘 𝜆 (𝑅 + 𝑇 𝑅) 𝑇 [ ]
𝑉̇ 𝜁 ≤𝑘∗ 𝜁 𝑇 𝑅(𝑐 + 𝑑) ⊗ (𝐴𝑇 𝑃 + 𝑃 𝐴) 𝜁 − ∗ 2 𝜁 (𝑐 + 𝑑)2 ⊗ Γ 𝜁
𝑁
∑𝑁 ∑𝑁
1 2
+ (1 + )𝜆𝑚𝑎𝑥 (Γ) (𝑐𝑖 + 𝑑𝑖 )2 ‖𝑒𝑖 ‖2 + 𝜆𝑚𝑎𝑥 (𝑃 ) (𝑑𝑖 + 𝑑̂𝑖 )𝑐̇ 𝑖 ‖𝑒𝑖 ‖2
𝜋2 𝑖=1
𝜅𝜋3 𝑖=1
[ ]
+ (𝜋2 + 𝜋3 )𝑘2∗ 𝜆𝑚𝑎𝑥 (𝑅𝑇 𝑅)𝜁 𝑇 (𝑐 + 𝑑)2 ⊗ Γ 𝜁
( ) 𝑁
∑
+ 𝑘2∗ 𝜅 2 𝑟2 + 𝑘∗ 𝜅𝐶∗ 𝑟 𝜆𝑚𝑎𝑥 (Γ) ‖𝑒𝑖 ‖2
𝑖=1
[( 𝜅 ) ]
+ 𝜁𝑇 2𝑘∗ 𝜅𝑅(𝑐 + 𝑑) − 𝑘∗ 𝑟𝐶∗ 𝐼𝑁 ⊗ Γ 𝜁. (31)
2
Finally, let’s consider the following Lyapunov function candidate:

𝑉 = 𝑉 𝜓 + 𝑉𝜁 . (32)
It’s evident that 𝑉 is a positive definite function. Combining (23) and (31), we can deduce 𝑉̇ as presented in (33),
[( ) ]
1
𝑉̇ ≤𝜁 𝑇 − 𝜛1 (𝑐 + 𝑑)2 + 2𝑘∗ 𝜅𝑅(𝑐 + 𝑑) − 𝜅𝑘∗ 𝑟𝐶∗ 𝐼𝑁 ⊗ Γ 𝜁
2
∑ 𝑁
[ ] 1
+ 𝑘∗ 𝜁 𝑇 𝑅(𝑐 + 𝑑) ⊗ (𝐴𝑇 𝑃 + 𝑃 𝐴) 𝜁 + 𝜆𝑚𝑎𝑥 (𝐴𝑇 𝐴)𝜁 𝑇 𝜁 − 𝜛3 ‖𝑒𝑖 ‖2
2 𝑖=1
𝑁 [
∑ ] 𝑁
𝐴 ∑
+ 𝜛2 (𝑐𝑖 + 𝑑𝑖 )2 + (𝑑𝑖 + 𝑑̂𝑖 )𝑐̇ 𝑖 + 𝜂𝑖 + 𝜓𝑖2 ‖𝑒𝑖 ‖2 + ∗ 𝜓̇ ‖𝑒 ‖2 , (33)
𝑖=1
2 𝑖=1 𝑖 𝑖
where
⎧ 𝜛 = 𝑘∗ 𝜆2 (𝑅+𝑇 𝑅) − (𝜋 + 𝜋 )𝑘2 𝜆 (𝑅𝑇 𝑅)
3 ∗ 𝑚𝑎𝑥
⎪ 1 𝑁 2
⎪ − 𝜋2 𝜆𝑚𝑎𝑥 (𝐵 𝑇 𝐵)𝜆𝑚𝑎𝑥 (𝑇 )
⎪ 1 {
( )
⎪ 𝜛 = max 1 + 𝜋1 + 𝜋2 𝜆𝑚𝑎𝑥 (𝐵 𝑇 𝐵)𝜆𝑚𝑎𝑥 (𝑇 ) 𝜆𝑚𝑎𝑥 (Γ), 12 ,
⎨ 2
2 1
}
⎪ ( )
⎪ 2
+ 2
𝜆 (𝐵 𝑇 𝐵)𝜆 (𝑇 ) 𝜆 (𝑃 )
𝜅𝜋3 𝜅𝜋1 𝑚𝑎𝑥 𝑚𝑎𝑥 𝑚𝑎𝑥
⎪
⎪ 𝜛 = 𝐵∗ + 𝛾 2 − 𝜋1 +1 𝐴2 𝛾 2 − (𝑘2 𝜅 2 𝑟2 + 𝑘 𝜅𝐶 𝑟)𝜆 (Γ).
⎩ 3 2 2 2 ∗ ∗ ∗ ∗ 𝑚𝑎𝑥

It is easy to observe that 𝜛1 > 0 when we choose 𝜋2 + 𝜋3 sufficiently small and a large enough 𝜋1 . Additionally, there exists a
sufficiently large 𝐵∗ to make 𝜛3 > 0. Let’s d­fine 𝐴∗ = 2𝜛2 , with 𝐴∗ = 2𝜛2 ≥ 1, in line with the assumption for 𝐴∗ as stated in (18).
Now, when we combine (11) and (33), we obtain:
[( ) ]
1
𝑉̇ ≤𝜁 𝑇 − 𝜛1 (𝑐 + 𝑑)2 + 2𝑘∗ 𝜅𝑅(𝑐 + 𝑑) − 𝜅𝑘∗ 𝑟𝐶∗ 𝐼𝑁 ⊗ Γ 𝜁
2
1 [ ]
+ 𝜆𝑚𝑎𝑥 (𝐴 𝐴)𝜁 𝜁 + 𝑘∗ 𝜁 𝑅(𝑐 + 𝑑) ⊗ (𝐴𝑇 𝑃 + 𝑃 𝐴) 𝜁.
𝑇 𝑇 𝑇
(34)
2
(2𝜅+1)2 𝑟2
Please note that when we choose 𝐶∗ ≥ 𝑘∗ , we have:
2𝜅𝜛1 𝑟
[( 𝐶 ) ]
𝜁𝑇 − 𝜛1 (𝑐 + 𝑑)2 + 2𝑘∗ 𝜅𝑅(𝑐 + 𝑑) − ∗ 𝜅𝑘∗ 𝑟𝐼𝑁 ⊗ Γ 𝜁
2
[(
( )2
2𝜅 + 1
= 𝜁𝑇 − 𝜛1 𝑐 + 𝑑 − 𝑘 𝑟𝐼 − 𝑘∗ 𝑅(𝑐 + 𝑑)
2𝜛1 ∗ 𝑁
]
( (2𝜅 + 1)2 ) )
1
+ 𝑘2∗ 𝑟2 − 𝜅𝑘∗ 𝑟𝐶∗ 𝐼𝑁 ⊗ Γ 𝜁
4𝜛1 2
𝑇
[ ]
≤ −𝜁 𝑘∗ 𝑅(𝑐 + 𝑑) ⊗ Γ 𝜁. (35)
By substituting (35) into (34), we obtain:
[ ] 1
𝑉̇ ≤ 𝑘∗ 𝜁 𝑇 𝑅(𝑐 + 𝑑) ⊗ (𝐴𝑇 𝑃 + 𝑃 𝐴 − 𝑃 𝐵𝐵 𝑇 𝑃 ) 𝜁 + 𝜆𝑚𝑎𝑥 (𝐴𝑇 𝐴)𝜁 𝑇 𝜁
2
( )
1
≤ 𝜆 (𝐴𝑇 𝐴) − 𝑘∗ 𝑟𝑐𝑖 (0) 𝜁 𝑇 𝜁, (36)
2 𝑚𝑎𝑥
9
S. Liu, H. Zhang and J. Sun Information Sciences 703 (2025) 121886

𝜆𝑚𝑎𝑥 (𝐴𝑇 𝐴)
where we have used (2) and the fact 𝑐𝑖 (0) > 0. It’s evident that 𝑉 (𝑡) remains bounded for 𝑡 ∈ [𝑡𝑙 , 𝑡𝑙+1 ) if 𝑘∗ ≥ . Since 𝑉 (𝑡) is
2𝑟𝑐𝑖 (0)
continuous at time points 𝑡𝑙∈ℤ>0 (refer to Remark 6 for a comprehensive explanation), it follows that 𝑉 (𝑡) is monotonically decreasing
for 𝑡 ∈ [0, ∞). Consequently, 𝑉 (𝑡) is bounded when 𝑡 > 0. This implies that 𝜁𝑖 (𝑡), 𝜁̂𝑖 (𝑡), 𝑐𝑖 (𝑡), 𝑐̇ 𝑖 (𝑡), and 𝜂𝑖 (𝑡) are all bounded.
Next, we will demonstrate that each agent has a strictly positive lower bound for the time interval between adjacent events, with
asymptotic convergence of the consensus error 𝜁(𝑡) to zero.
For 𝑡 ∈ [𝑡𝑖𝑘 , 𝑡𝑖𝑘+1 ), from (11), we have

𝜓̇ 𝑖 (𝑡) ≥ −𝜇𝑖 − 𝜓𝑖2 (𝑡). (37)

Here, 𝜇𝑖 = (𝑐𝑖𝑚 + 𝑑𝑖𝑚 )2 + 2𝑑𝑖𝑚 𝑐̇ 𝑖𝑚 + 𝜂𝑖𝑚 , where 𝑐𝑖𝑚 = max𝑡∈[0,∞) 𝑐𝑖 (𝑡) > 0, 𝑑𝑖𝑚 = max𝑡∈[0,∞) 𝑑𝑖 (𝑡) > 0, and 𝑐̇ 𝑖𝑚 = max𝑡∈[0,∞) 𝑐̇ 𝑖 (𝑡) > 0, for
𝑖 = 1, … , 𝑁 . By employing the comparison lemma [41], we introduce a new variable:

𝜑̇ 𝑖 (𝑡) = −𝜇𝑖 − 𝜑2𝑖 (𝑡), (38)

with 𝜑𝑖 (𝑡𝑖+
𝑘
) = 𝛾𝑖 . It is evident that 𝜓𝑖 (𝑡) ≥ 𝜑𝑖 (𝑡), for 𝑡 ∈ [𝑡𝑖𝑘 , 𝑡𝑖𝑘+1 ). Integrating (38), results in the following relationship:
( )
1 𝛾 𝜑 (𝑡)
𝑡 − 𝑡𝑖𝑘 = √ arctan √ 𝑖 − arctan √𝑖 . (39)
𝜇𝑖 𝜇𝑖 𝜇𝑖

Considering that 𝜓𝑖 (𝑡) ≥ 𝜑𝑖 (𝑡) and 𝜓𝑖 (𝑡𝑖𝑘+1 ) = 0, we can obtain

1 𝛾
𝑡𝑖𝑘+1 − 𝑡𝑖𝑘 ≥ 𝜏𝑖 = √ arctan √ 𝑖 > 0. (40)
𝜇𝑖 𝜇𝑖
As stipulated by Definition 2, (40) provides co­firmation that each agent preserves a strictly positive lower bound on the time intervals
between events, effectively eliminating Zeno behavior. From (36), it can be observed that 𝑉̇ (𝑡) ≤ 0. According to Remark 6, it is known
that 𝑉 (𝑡) is continuous at each event time 𝑡𝑙 . Consequently, by applying the invariant set theorem for impulsive dynamical systems,
as generalized in [36] (Th.5.1), we can conclude that the solutions of the system (1), tend towards the invariant set {𝑥(𝑡) ∈ ℝ𝑁𝑛 ∶
𝜁(𝑡) = 0} for 𝑡 > 0. This implies that, for all initial conditions, the solutions of the system converge to this specific invariant set as
𝑡 → ∞, i.e., 𝜁(𝑡) → 0 as 𝑡 → ∞. This completes the proof. □

Remark 6. Due to the piecewise nature of 𝜓𝑖 (𝑡) and 𝑒𝑖 (𝑡) within 𝑉𝜓 (𝑡), it is necessary to consider the continuity of 𝑉 (𝑡) at event
times 𝑡𝑙 in the following two cases: i) When 𝑡𝑙 = 𝑡𝑖𝑘 , 𝜓𝑖 (𝑡+ 𝑙
) = 𝛾𝑖 , 𝑒𝑖 (𝑡+
𝑙
) = 0, 𝜓𝑖 (𝑡−
𝑙
) = 0, and 𝑒𝑖 (𝑡−
𝑙
) is a bounded constant. Conse­
quently, it holds that lim𝑡→𝑡+ 𝜓𝑖 (𝑡)𝑒𝑖 (𝑡)𝑒𝑖 (𝑡) = lim𝑡→𝑡− 𝜓𝑖 (𝑡)𝑒𝑖 (𝑡)𝑒𝑖 (𝑡) = 0. ii) When 𝑡𝑙 ≠ 𝑡𝑖𝑘 , it is evident that lim𝑡→𝑡+ 𝜓𝑖 (𝑡) × 𝑒𝑇𝑖 (𝑡)𝑒𝑖 (𝑡) =
𝑇 𝑇
𝑙 𝑙 𝑙
lim𝑡→𝑡− 𝜓𝑖 (𝑡)𝑒𝑇𝑖 (𝑡)𝑒𝑖 (𝑡). Hence, based on these two cases, it can be concluded that 𝑉 (𝑡) remains continuous at event times 𝑡𝑙 .
𝑙

Remark 7. It’s worth noting that the condition 𝑒𝑖 (𝑡) ≠ 0 in (10) plays a crucial role. If we were to remove this condition, the conclusion
of the consensus error 𝜁(𝑡) converging asymptotically to zero would still hold. However, when combined with (11), we observe that
in stable systems, 𝜓𝑖 (𝑡) would repetitively oscillate between 𝛾𝑖 and 0 at fixed intervals. This oscillation represents an unnecessary
consumption of computational resources. Therefore, the inclusion of 𝑒𝑖 (𝑡) ≠ 0 in (10) serves a purpose by preventing the event­
triggered algorithm from sampling needlessly when the consensus error already asymptotically converges to zero.

Remark 8. The intuitive understanding is that by increasing 𝛾𝑖 , 𝜓𝑖 (𝑡) takes more time to approach zero, resulting in an extended time
interval between consecutive events. However, in practice, ensuring longer intervals between events is not as simple as just increasing
𝛾𝑖 . This is because the adjustment of 𝛾𝑖 indirectly impacts the maximum values of 𝜁𝑖 (𝑡), subsequently i­fluencing 𝑐𝑖𝑚 , 𝑑𝑖𝑚 , and 𝑐̇ 𝑖𝑚 in
(15). Therefore, selecting an appropriate 𝛾𝑖 requires finding a balance between the transient performance of linear MASs (1) and the
time intervals between events. For a more detailed comparison, please refer to Table 1.

Remark 9. In previous research, distributed event-triggered algorithms designed for directed graphs often necessitated the establish­
ment of multiple event detection inequalities (see [23,25,37]). A distinguishing feature of the distributed algorithms proposed in this
paper is their reliance on a single event monitoring condition. The simplicity and comprehensibility of single-event triggering make it
more straightforward to understand and analyze, reducing computational complexity and aiding in intuitive performance evaluation.
In contrast, multiple event triggering often necessitates the combination of multiple event monitoring conditions, increasing system
intricacy. This simplicity not only alleviates the computational load on processors but also broadens its adaptability across diverse
domains, thus leading to enhanced system efficiency, as elucidated in [19].

Remark 10. This paper has primarily focused on the consensus control of linear MASs, assuming that the system dynamics are
known. However, real-world applications often involve nonlinear and uncertain dynamics among agents, which can restrict the
direct applicability of our algorithm. Nonetheless, under specific conditions, our algorithms remain applicable. In particular, our
algorithm can be effectively employed when nonlinear systems can be linearized using feedback linearization techniques [6,38,39].
In such cases, the matrices 𝐴 and 𝐵 in (1) can be precisely determined, facilitating the implementation of our control algorithm.

10
S. Liu, H. Zhang and J. Sun Information Sciences 703 (2025) 121886

Fig. 1. Communication graph .

Furthermore, our techniques may be suitable for unknown linear systems with relatively straightforward dynamics, such as single­
input single-output minimum-phase systems with a relative degree of one [17,35]. It is important to emphasize, however, that these
conditions are stringent and do not universally apply to all nonlinear dynamic scenarios.

4. Simulation example

This section presents an evaluation of the proposed timer-based event-triggered algorithm through a simulation example imple­
mented in Simulink with a fixed step size of 1e-4. The simulation involves a system comprising six agents, each characterized by
dynamics described by (1). The matrices 𝐴 and 𝐵 governing these dynamics are specifically d­fined as: 𝐴 = [0, 1; 0, 0] and 𝐵 = [0; 1].
Furthermore, upon solving the algebraic Riccati equation, the resulting matrix 𝑃 is determined to be: 𝑃 = [1.7321, 1; 1, 1.7321].
The communication network  underlying this simulation comprises six agents and is depicted in Fig. 1. This graph is explicitly
ident­fied as a directed and strongly connected network, thereby satisfying Assumption 1.
The parameters in (7) and (11) are set as 𝜅𝑖 = 0.5, 𝑐𝑖 (0) = 1 and 𝜎𝑖 = 0.5. For convenience, we set 𝜓𝑖 (𝑡𝑖+ 𝑘
) = 𝛾𝑖 = 1.5 as the initial
value of the timer for each agent at event instants. The initial states for each agent are as follows: 𝑥1 (0) = [7, 1]𝑇 , 𝑥2 (0) = [8, 2]𝑇 ,
𝑥3 (0) = [9, 3]𝑇 , 𝑥4 (0) = [10, 4]𝑇 , 𝑥5 (0) = [11, 5]𝑇 and 𝑥6 (0) = [12, 6]𝑇 .
The evolution of the state variables for each agent over time, demonstrating the achievement of asymptotic consensus in the
linear multi-agent systems described by (1), is illustrated in Fig. 2. Furthermore, Figs. 3 and 4 depict the temporal evolution of the
adaptive parameters 𝑐𝑖 (𝑡) and 𝜂𝑖 (𝑡) for each agent, respectively. The convergence of these adaptive parameters 𝑐𝑖 (𝑡) and 𝜂𝑖 (𝑡) to specific
constants is consistent with the conclusions of Theorem 1.
In Fig. 5, the event triggering instances for each agent are visually represented. Additionally, Fig. 6 provides a detailed wor­flow
illustrating the event monitoring conditions specifically for agents 3 and 4 within the time interval 𝑡 ∈ [0, 2]. Further elucidation on
this wor­flow can be found in Remark 4, offering a comprehensive explanation of the underlying event-based mechanisms.
Fig. 7 presents a comparative analysis between the proposed algorithm and a traditional timer-based algorithm [5,31], operating
under identical conditions. The comparison evaluates the total accumulated values computed within a ten-second interval for both
10
algorithms, assessed using the integral ∫0 |𝜓(𝜏)|𝑑𝜏 ̇ . In the Fig. 7, the red bar chart illustrates the cumulative values obtained by
the proposed algorithm, while the blue bar chart depicts the results from the traditional timer-based algorithm. The numerical
values generated by the proposed algorithm exhibit significantly smaller magnitudes compared to those produced by the comparative
algorithm. Therefore, it can be concluded that the proposed algorithm provides a substantial advantage in mitigating numerical
ove­flow errors.
To investigate the impact of different 𝛾𝑖 values on the performance of (1), we conducted eight sets of comparative simulations
using a control variables approach, as outlined in Table 1. In this context, 𝑡𝑖 and 𝜏𝑖 (𝑖 = 1, … , 6) represent the actual minimum time
between events and the calculated minimum time between events based on (40), respectively. The total number of events is denoted as
10
𝑇𝑛 , and the system’s transient performance is assessed using the performance metric 𝐽 = ∫0 |𝜁(𝜏)|2 𝑑𝜏 . A higher value of 𝐽 indicates
poorer transient performance. Table 1 illustrates that an increase in 𝛾𝑖 results in a deterioration of the system’s transient performance.
As explained in Remark 8, 𝛾𝑖 indirectly affects 𝑐𝑖𝑚 , 𝑑𝑖𝑚 , and 𝑐̇ 𝑖𝑚 . Therefore, simply raising 𝛾𝑖 does not guarantee an increase in the
time between events for each agent. It is noteworthy that within these eight comparative sets, the actual minimum time between
events, denoted as 𝑡𝑖 , is significantly larger than the simulation step size of 1e-4. This observation underscores the effectiveness of
the proposed algorithm in avoiding continuous sampling and preventing events from occurring in rapid succession.
We conducted an additional simulation to assess the robustness of the proposed distributed algorithm when facing network
interruptions by severing the communication link between agent 1 and agent 2 at 1.5 s. The communication topology and parameters
remained consistent with the original experiment. As illustrated in Fig. 8, even after severing the communication between agent 1
and agent 2 at 1.5 seconds, the system still achieved consensus. This demonstrates the strong robustness of our proposed distributed
algorithm, capable of maintaining stability in the face of network disruptions.
Table 2 provides a comparison between this paper and existing relevant literature in key technical aspects, with a particular focus
on four core attributes: avoiding i­finitesimal event intervals (A1), flexible adaptive parameter initialization (A2), applicability to
directed graphs (A3), and avoidance of arithmetic ove­flow errors (A4). Through this comparison, we can clearly see that this study
demonstrates advantages in all of these aspects.

5. Conclusion

In this paper, the problem of consensus control for linear MASs was investigated, addressing several critical research gaps ident­fied
in the existing literature. Firstly, Zeno behavior was mitigated by constructing bounded event descent rates, ensuring that adjacent

11
S. Liu, H. Zhang and J. Sun Information Sciences 703 (2025) 121886

Table 1
Simulation results for different 𝛾𝑖 .

𝛾𝑖 𝑡1 ∕𝜏1 (s) 𝑡2 ∕𝜏2 (s) 𝑡3 ∕𝜏3 (s) 𝑡4 ∕𝜏4 (s) 𝑡5 ∕𝜏5 (s) 𝑡6 ∕𝜏6 (s) 𝑇𝑛 𝐽
1.5 0.0166/0.0141 0.0172/0.0129 0.0188/0.0106 0.0339/0.0130 0.1008/0.0389 0.0215/0.0158 352 10.2225
2.0 0.0213/0.0175 0.0224/0.0153 0.0166/0.0079 0.0475/0.0144 0.0775/0.0274 0.0304/0.0207 394 10.9617
2.5 0.0255/0.0209 0.0273/0.0167 0.0184/0.0074 0.0485/0.0142 0.0817/0.0299 0.0403/0.0256 515 12.5098
3.0 0.0317/0.0238 0.0327/0.0146 0.0264/0.0083 0.0408/0.0159 0.0447/0.0146 0.0512/0.0302 590 14.2305
3.5 0.0361/0.0287 0.0292/0.0109 0.0313/0.0071 0.0353/0.0103 0.0501/0.0191 0.0629/0.0332 683 15.1169
4.0 0.0402/0.0317 0.0252/0.0090 0.0244/0.0073 0.0316/0.0101 0.0267/0.0113 0.0749/0.0364 719 16.3637
5.0 0.0465/0.0344 0.0253/0.0083 0.0218/0.0068 0.0238/0.0080 0.0381/0.0129 0.0979/0.0411 874 19.1628
6.0 0.0516/0.0344 0.0154/0.0054 0.0161/0.0050 0.0201/0.0070 0.0163/0.0060 0.1164/0.0412 1023 22.7332

Fig. 2. State trajectories of each agent.

Fig. 3. Adaptive parameter 𝑐𝑖 .

Fig. 4. Adaptive parameter 𝜂𝑖 .

12
S. Liu, H. Zhang and J. Sun Information Sciences 703 (2025) 121886

Fig. 5. Triggering instants for each agent.

Fig. 6. State trajectories of 𝜓3 and 𝜓4 .

10
Fig. 7. Comparison of total accumulated values ∫0 |𝜓̇ 𝑖 (𝜏)|𝑑𝜏 for each agent: Proposed (Red) vs. Traditional timer-based algorithm (Blue).

event intervals for each agent cannot be arbitrarily close. This ensures the stability of the consensus process and prevents undesirable
behaviors. Secondly, the algorithm allows for flexible adaptive parameter initialization, enabling true distributed operation and
enhancing scalability. Agents are empowered to autonomously select their initial adaptive parameters, eliminating the need for
centralized coordination. Thirdly, an algorithm specifically applicable to directed graphs was designed, resolving the asymmetry
introduced by directed communication pathways. This expands the applicability of consensus control algorithms to a broader range
of network topologies. Lastly, arithmetic ove­flow errors were prevented through a novel timer-based event-triggered algorithm,
significantly reducing the magnitude of generated numerical values. This enhances the stability and reliability of the control algorithm,
especially in scenarios with large-scale systems.

13
S. Liu, H. Zhang and J. Sun Information Sciences 703 (2025) 121886

Fig. 8. Robustness evaluation with communication disruption.

Table 2
Advantages and disadvantages comparison table.

Related literature A1 A2 A3 A4
√ √
[17,19,22] × ×
√ √
[20,23--25,32,42] ×
√ ×
√
[5,31] √ × ×
√
[29,30,43,44] √ ×
√ ×
√ √
This paper

A1: Avoiding i­finitesimal event intervals. A2: Flexible adaptive parameter initialization.
A3: Applicability to directed graphs. A4: Avoiding arithmetic ove­flow errors.

We look forward to future studies that could extend our results to a broader range of nonlinear systems, especially those with
significant uncertainties. Additionally, exploring how our method could be adapted to more complex system dynamics will be an area
of deeper investigation in our future work.

CRediT authorship contribution statement

Shu Liu: Writing -- original draft, Methodology. Huaguang Zhang: Writing -- review & editing, Resources, Funding acquisition.
Jiayue Sun: Methodology.

Declaration of competing interest

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to
i­fluence the work reported in this paper.

Acknowledgements

This work was supported by the National Key R&D Program of China under grant 2018YFA0702200, the National Natural Science
Foundation of China (U23B20118), the Liaoning Revitalization Talents Program (XLYC1801005), the Natural Science Foundation of
Liaoning Province, China (2022JH25/10100008), and the Collective Intelligence & Collaboration Laboratory (Open Fund Project No.
QXZ23014401).

Data availability

No data was used for the research described in the article.

References

[1] Q. He, W. Liu, Formation control for linear multi-agent systems with asynchronously sampled outputs, Inf. Sci. 658 (2024) 119992.
[2] W. Ren, X. Zhang, Fully distributed containment control of second-order nonlinear multi-agent systems using disturbance observer with directed graph, Inf. Sci.
649 (2023) 119629.
[3] H. Zhang, J. Zhang, Leader-follower consensus control for linear multi-agent systems by fully distributed edge-event-triggered adaptive strategies, Inf. Sci. 555
(2021) 314--338.
[4] H. Zhang, J. Zhang, Y. Cai, S. Sun, J. Sun, Leader-following consensus for a class of nonlinear multi-agent systems under event-triggered and edge-event triggered
mechanisms, IEEE Trans. Cybern. 52 (8) (2022) 7643--7654.

14
S. Liu, H. Zhang and J. Sun Information Sciences 703 (2025) 121886

[5] Y.-Y. Qian, A.V.P. Premakumar, Y. Wan, Z. Lin, Y.A. Shamash, A. Davoudi, Dynamic event-triggered distributed secondary control of DC microgrids, IEEE Trans.
Power Electron. 37 (9) (2022) 10226--10238.
[6] M. Zhai, Q. Sun, R. Wang, H. Zhang, Containment-based multiple PCC voltage regulation strategy for communication link and sensor faults, IEEE/CAA J. Autom.
Sin. 10 (11) (2023) 2045--2055.
[7] B. Ning, Q.-L. Han, L. Ding, Distributed secondary control of AC microgrids with external disturbances and directed communication topologies: a full-order
sliding-mode approach, IEEE/CAA J. Autom. Sin. 8 (3) (2021) 554--564.
[8] Z. Peng, J. Wang, D. Wang, Q.-L. Han, An overview of recent advances in coordinated control of multiple autonomous surface vehicles, IEEE Trans. Ind. Inform.
17 (2) (2021) 732--745.
[9] Y. Xu, G. Deng, T. Zhang, H. Qiu, Y. Bao, Novel denial-of-service attacks against cloud-based multi-robot systems, Inf. Sci. 576 (2021) 329--344.
[10] X. Chen, W. Liang, X. Zhou, D. Jiang, X. Kui, K. Li, An efficient transmission algorithm for power grid data suitable for autonomous multi-robot systems, Inf. Sci.
572 (2021) 543--557.
[11] Y. Ju, D. Ding, X. He, Q.-L. Han, G. Wei, Consensus control of multi-agent systems using fault-estimation-in-the-loop: dynamic event-triggered case, IEEE/CAA
J. Autom. Sin. 9 (8) (2022) 1440--1451.
[12] W. Li, H. Zhang, Y. Zhou, Y. Wang, Bipartite formation tracking for multi-agent systems using fully distributed dynamic edge-event-triggered protocol, IEEE/CAA
J. Autom. Sin. 9 (5) (2022) 847--853.
[13] L. Ding, Q.-L. Han, X. Ge, X. Zhang, An overview of recent advances in event-triggered consensus of multiagent systems, IEEE Trans. Cybern. 48 (4) (2018)
1110--1123.
[14] D.V. Dimarogonas, E. Frazzoli, K.H. Johansson, Distributed event-triggered control for multi-agent systems, IEEE Trans. Autom. Control 57 (5) (2012) 1291--1297.
[15] C. Nowzari, E. Garcia, J. Cortés, Event-triggered communication and control of networked systems for multi-agent consensus, Automatica 105 (2019) 1--27.
[16] Y. Fan, G. Feng, Y. Wang, C. Song, Distributed event-triggered control of multi-agent systems with combinational measurements, Automatica 49 (2) (2013)
671--675.
[17] S. Liu, J. Sun, H. Zhang, M. Zhai, Fully distributed event-driven adaptive consensus of unknown linear systems, IEEE Trans. Neural Netw. Learn. Syst. 34 (10)
(2023) 8007--8016.
[18] W. Xu, D.W.C. Ho, L. Li, J. Cao, Event-triggered schemes on leader-following consensus of general linear multiagent systems under different topologies, IEEE
Trans. Cybern. 47 (1) (2017) 212--223.
[19] S. Luo, D. Ye, Adaptive double event-triggered control for linear multi-agent systems with actuator faults, IEEE Trans. Circuits Syst. I, Regul. Pap. 66 (12) (2019)
4829--4839.
[20] W. He, B. Xu, Q.-L. Han, F. Qian, Adaptive consensus control of linear multiagent systems with dynamic event-triggered strategies, IEEE Trans. Cybern. 50 (7)
(2020) 2996--3008.
[21] M. Li, Y. Long, T. Li, H. Liang, C.L.P. Chen, Dynamic event-triggered consensus control for input-constrained multi-agent systems with a designable minimum
inter-event time, IEEE/CAA J. Autom. Sin. 11 (3) (2024) 649--660, https://doi.org/10.1109/JAS.2023.123582.
[22] B. Cheng, Z. Li, Fully distributed event-triggered protocols for linear multiagent networks, IEEE Trans. Autom. Control 64 (4) (2019) 1655--1662.
[23] X. Li, Z. Sun, Y. Tang, H.R. Karimi, Adaptive event-triggered consensus of multiagent systems on directed graphs, IEEE Trans. Autom. Control 66 (4) (2021)
1670--1685.
[24] J. Zhang, H. Zhang, W. Li, Y. Wang, Adaptive event-triggered bipartite containment control of linear multiagent systems under directed topologies, IEEE Trans.
Control Netw. Syst. 10 (1) (2023) 516--525.
[25] C. Xia, C. Wang, Fully distributed event-triggered output feedback control for linear multi-agent systems with a derivable leader under directed graphs, Inf. Sci.
619 (2023) 562--577.
[26] W. Xu, W. He, D.W.C. Ho, J. Kurths, Fully distributed observer-based consensus protocol: adaptive dynamic event-triggered schemes, Automatica 139 (2022)
110188.
[27] R. Koike, T. Endo, F. Matsuno, Output-based dynamic event-triggered consensus control for linear multiagent systems, Automatica 133 (2021) 109863.
[28] J. Berneburg, C. Nowzari, Robust dynamic event-triggered coordination with a designable minimum interevent time, IEEE Trans. Autom. Control 66 (8) (2021)
3417--3428.
[29] M. Li, Y. Long, T. Li, C.L.P. Chen, Consensus of linear multi-agent systems by distributed event-triggered strategy with designable minimum inter-event time, Inf.
Sci. 609 (2022) 644--659.
[30] J. Berneburg, C. Nowzari, Distributed dynamic event-triggered coordination with a designable minimum inter-event time, in: 2019 American Control Conference
(ACC), Philadelphia, PA, USA, 2019, pp. 1424--1429.
[31] Y.-Y. Qian, Y. Wan, Design of distributed adaptive event-triggered consensus control strategies with positive minimum inter-event times, Automatica 133 (2021)
109837.
[32] J. Zhang, H. Zhang, Z. Ming, Y. Mu, Adaptive event-triggered time-varying output bipartite formation containment of multiagent systems under directed graphs,
IEEE Trans. Neural Netw. Learn. Syst. 34 (11) (2023) 8909--8922.
[33] H. Wang, W. Ren, W. Yu, D. Zhang, Fully distributed consensus control for a class of disturbed second-order multi-agent systems with directed networks,
Automatica 132 (2021) 109816.
[34] J. Mei, W. Ren, J. Chen, Distributed consensus of second-order multi-agent systems with heterogeneous unknown inertias and control gains under a directed
graph, IEEE Trans. Autom. Control 61 (8) (2016) 2019--2034.
[35] Z. Li, Z. Ding, Distributed adaptive consensus and output tracking of unknown linear systems on directed graphs, Automatica 55 (2015) 12--18.
[36] V. Chellaboina, S.P. Bhat, W.M. Haddad, An invariance principle for nonlinear hybrid and impulsive dynamical systems, Nonlinear Anal., Theory Methods Appl.
53 (3--4) (2003) 527--550.
[37] J. Zhang, H. Zhang, Y. Luo, Y. Liu, Adaptive event-triggered leader-follower consensus of linear multiagent systems under directed graph with nonzero leader
input, IEEE Trans. Circuits Syst. II, Express Briefs 69 (3) (2022) 1442--1446.
[38] B. Ali, D. Ali, L.L. Frank, M.G. Josep, Distributed cooperative secondary control of microgrids using feedback linearization, IEEE Trans. Power Syst. 28 (3) (2013)
3462--3470.
[39] M. Zhai, Q. Sun, R. Wang, B. Wang, S. Liu, H. Zhang, Fully distributed fault-tolerant event-triggered control of microgrids under directed graphs, IEEE Trans.
Netw. Sci. Eng. 9 (5) (2022) 3570--3579.
[40] B. David, Spectral Theory: Basic Concepts and Applications, Graduate Texts in Mathematics, Springer, 2020, p. 284.
[41] H. Khalil, Nonlinear Systems, third edition, Prentice Hall, 2002.
[42] S. Liu, J. Sun, H. Zhang, M. Zhai, Coordination for Lur’e multi-agent systems: fully distributed event-driven approach with single-event monitoring condition,
IEEE Trans. Circuits Syst. II, Express Briefs 69 (12) (2022) 4919--4923.
[43] S. Liu, B. Niu, H. Karimi, X. Zhao, Self-triggered fixed-time bipartite fault-tolerant consensus for nonlinear multiagent systems with function constraints on states,
Chaos Solitons Fractals 178 (2024) 114367.
[44] M.-N. Zhai, J. Sun, Distributed critical bus voltage regulation control for multi-microgrids with positive minimum interevent times, IEEE Trans. Ind. Inform.
20 (4) (2024) 5774--5783.

15