⚡ electrical engineering

Decentralized design of consensus protocols with minimal communication links based on directed spanning tree

This paper proposes a fully decentralized consensus protocol design for multi-agent systems that utilizes a directed spanning tree-based linear transformation to convert the consensus problem into a decentralized output stabilization task, thereby achieving consensus with minimal communication links and providing a necessary and sufficient criterion based on decentralized fixed modes.

Original authors: Yangzhou Chen, Lanhao Zhao

Published 2026-03-18

📖 5 min read🧠 Deep dive

CC BY 4.0

Original authors: Yangzhou Chen, Lanhao Zhao

Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

Imagine a group of friends trying to agree on a meeting time. In the world of Multi-Agent Systems (MAS), these "friends" are robots, drones, or computers that need to coordinate their actions to reach a common goal (called consensus).

Usually, to agree, everyone talks to everyone else they know. If you have 100 friends, that's a lot of phone calls! This creates a traffic jam of data, slows things down, and if one connection fails, the whole plan might collapse.

This paper proposes a smarter, simpler way to get everyone on the same page. Here is the breakdown using everyday analogies:

1. The Problem: The "Group Chat" Nightmare

In traditional systems, every agent (robot) tries to listen to all its neighbors at once.

The Analogy: Imagine a huge conference room where everyone is shouting their opinion to everyone else. It's chaotic, loud, and inefficient. If you have to listen to 10 people to make a decision, you get overwhelmed.
The Goal: The authors want to reduce the noise. They want to find the minimum number of connections needed to get the job done, while still letting every robot design its own strategy without asking a "boss" (central computer) for permission.

2. The Solution: The "Family Tree" Strategy

The paper suggests organizing the group not as a chaotic circle, but as a Directed Spanning Tree (DST).

The Analogy: Think of a family tree or a corporate hierarchy.
- There is one Root (the Grandparent/CEO).
- Everyone else has exactly one parent above them.
- Information flows down the tree.
The Magic: In this structure, you don't need to talk to your cousins or your neighbors' neighbors. You only need to talk to your direct parent.
- If everyone listens to their parent, and the parent listens to their parent, the message eventually reaches everyone.
- This is the "minimal communication link." It's the most efficient path possible.

3. The Transformation: Turning "Agreement" into "Stability"

The tricky part is that the robots are moving and changing. How do we mathematically prove they will agree?

The Analogy: Imagine a line of dancers. Instead of trying to calculate the distance between every single pair of dancers (which is a nightmare), the authors invent a new way of looking at the dance.
They transform the problem: Instead of asking "Are we all at the same spot?", they ask "Is the distance between a dancer and their partner shrinking to zero?"
By focusing only on the parent-child pairs (the fundamental edges of the tree), they turn a complex group problem into a simple "stability" problem. It's like checking if a single domino falling will knock over the next one, rather than calculating the physics of the whole pile.

4. The "Decentralized" Design: Everyone is the Boss

Usually, to make a group of robots work together, a super-computer calculates the settings for everyone and sends them out.

The Paper's Innovation: This paper allows every robot to design its own settings independently.
The Analogy: Instead of a conductor telling every musician exactly how to play, the conductor gives them a simple rule: "Listen to the person in front of you and adjust your speed so you don't bump into them."
Each robot calculates its own "gain" (how hard to push or pull) based only on its parent. They don't need to know the whole group's plan. This makes the system incredibly robust; if one robot breaks, the others can still figure out their own settings.

5. The "Root" Problem: What if the Boss is Lost?

There's a catch. In a perfect tree, the top person (the Root) has no parent. But in the real world, the "Root" might not be the actual leader of the original group.

The Analogy: Imagine the "Grandparent" of the family tree is actually a child in the original family. They don't have a parent to listen to!
The Fix: The authors figured out how to handle this. Even if the Root doesn't have a parent, they can still calculate a special setting using a mathematical trick (called the Gerschgorin Circle Theorem, which is like drawing a safety zone on a map). As long as everyone stays inside their safety zone, the whole group stays stable.

Summary: Why This Matters

Less Data: Robots talk to fewer people, saving battery and bandwidth.
More Freedom: Each robot can be unique and calculate its own rules.
More Reliability: If the network gets crowded or a link breaks, the "tree" structure is simple enough to keep the group moving.

In a nutshell: The authors found a way to organize a chaotic crowd into a neat family tree, allowing everyone to just listen to their parent. This lets the whole group agree on a plan quickly, with minimal talking, and without needing a central boss to micromanage everyone.

1. Problem Statement

The paper addresses two critical challenges in the consensus control of Multi-Agent Systems (MASs):

Decentralized Design: Existing consensus protocols often rely on centralized parameter tuning or require all agents to use identical control gain matrices, limiting flexibility and scalability. The authors seek a method where each agent can independently design its own feedback gain.
Minimal Communication: Traditional distributed protocols often require agents to communicate with all their neighbors simultaneously, leading to high data traffic and potential network congestion. The goal is to achieve consensus using the minimum number of communication links possible.

The core problem is to design a distributed control protocol for linear MASs that guarantees asymptotic state consensus using only the information from neighbors defined by a Directed Spanning Tree (DST), rather than the full communication graph.

2. Methodology

The proposed methodology relies on a linear transformation approach to convert the consensus problem into a decentralized output stabilization problem.

A. Problem Transformation via DST

Graph Representation: The communication topology is modeled as a weighted directed graph $G$ . The authors assume the existence of a Directed Spanning Tree (DST) $T$ within $G$ , with a designated root node (node $N$ ).
Fundamental Edge States: Instead of tracking the state of every agent, the system is transformed to track the fundamental edge states ( $y_i = x_{k_i} - x_i$ ), where $k_i$ is the parent of node $i$ in the DST.
Linear Mapping: Using Lemma 1, the authors prove that any edge state in the full graph can be linearly represented by the fundamental edge states on the DST. This allows the reduction of the $N$ -agent system into a system of $N-1$ fundamental edge dynamics.

B. Decentralized Output Stabilization

The consensus problem is equivalently transformed into a decentralized output-feedback stabilization problem.
The system is modeled as:
$\dot{y} = A^* y + B^* u$
$z = C^* y$
where $y$ is the fundamental edge state vector, and the output $z$ represents the information available to each agent.
Stability Criterion: The system achieves consensus if and only if the closed-loop system is asymptotically stable. This is analyzed using the concept of Decentralized Fixed Modes (DFM). A necessary and sufficient condition is derived: the system achieves consensus if the DFM set does not contain any eigenvalues of the open-loop system with non-negative real parts.

C. Protocol Design with Minimal Links

The paper proposes two specific design cases based on the role of the DST root node in the original graph:

Case 1: Root Node is a Root in Original Graph
- If the DST root node has no incoming edges in the original topology, it requires no control input ( $u_N = 0$ ).
- Other agents only use information from their specific parent in the DST.
- Result: The closed-loop system matrix becomes an upper triangular block matrix. Consensus is achieved if each local gain matrix $K_i$ stabilizes the subsystem $A - w_{i,k_i} B K_i$ (Hurwitz condition).
Case 2: Root Node is Not a Root in Original Graph
- If the DST root node receives information from other nodes in the original graph, it must process this information.
- The control law for the root node involves a sum of edge states from its neighbors.
- Stability Analysis: Since the system matrix is no longer strictly triangular, the authors use the Gerschgorin Circle Theorem (extended to block matrices) to derive sufficient conditions.
- Decentralized Design: The conditions allow each agent to design its gain $K_i$ based on local information and the known gains of its immediate neighbors on the DST, ensuring all Gerschgorin circles lie in the left half-plane.

3. Key Contributions

Minimal Communication Topology: The paper demonstrates that consensus can be achieved using only the links defined by a DST, which is the theoretical minimum number of links required for connectivity in a directed graph. This significantly reduces communication overhead compared to protocols using full neighbor sets.
Fully Decentralized Design Procedure: Unlike traditional methods that often require a common gain matrix for all agents or centralized computation, this approach allows each agent to independently calculate its unique feedback gain matrix based on local parameters and the specific structure of the DST.
DFM-Based Necessary and Sufficient Conditions: The authors establish a rigorous theoretical link between the consensus problem and Decentralized Fixed Modes, providing a precise criterion for solvability.
Gerschgorin-Based Local Design: For the complex case where the root node receives external information, the paper provides a practical, decentralized design method using Gerschgorin circles, avoiding the need for global eigenvalue analysis.

4. Results and Verification

The theoretical results are validated through two numerical examples:

Example 1 (Root is a Root Node): A 4-agent system where the DST root (Node 4) has no incoming edges. The authors designed distinct gain matrices ( $K_1, K_2, K_3$ ) for the other agents. Simulations showed that all agent states converged to a common value, confirming that using only DST links is sufficient.
Example 2 (Root is Not a Root Node): A 6-agent system where the DST root (Node 6) receives information from Node 1.
- The authors applied the Gerschgorin circle conditions to design gains sequentially (starting from the root's neighbors).
- Simulations confirmed that the system achieved consensus even with the root node receiving external data.
- A comparative simulation showed that if the root node's gain was set to zero (ignoring external data), the other agents still tracked the root, demonstrating the robustness of the DST-based structure.

5. Significance

Efficiency: By minimizing communication links, the proposed method reduces bandwidth usage, energy consumption, and the risk of network congestion, making it highly suitable for large-scale or resource-constrained MASs (e.g., sensor networks, UAV swarms).
Scalability and Flexibility: The fully decentralized design allows for the addition or removal of agents without requiring a global redesign of the control parameters. Each agent operates with its own unique gain, offering greater flexibility than homogeneous control strategies.
Theoretical Foundation: The work bridges the gap between graph theory (DST existence) and control theory (decentralized stabilization), offering a systematic framework for designing minimal-link consensus protocols.

In conclusion, this paper provides a robust framework for designing consensus protocols that are both communication-efficient (using minimal links) and computationally decentralized (allowing independent gain design), addressing key bottlenecks in modern multi-agent system deployment.