[Paper Review] Consensus on Matrix-weighted Time-varying Networks

Lulu Pan, Haibin Shao+3 more|arXiv (Cornell University)|Jan 30, 2020

Neural Networks Stability and Synchronization25 references4 citations

TL;DR

This paper establishes necessary and sufficient conditions for average consensus on matrix-weighted time-varying networks by introducing the matrix-weighted integral network and analyzing its Laplacian null space. For periodic networks, it proves that average consensus is achieved if and only if the integral network over one period has a positive spanning tree, providing both algebraic and graph-theoretic conditions.

ABSTRACT

This paper examines the consensus problem on time-varying matrix-weighed undirected networks. First, we introduce the matrix-weighted integral network for the analysis of such networks. Under mild assumptions on the switching pattern of the time-varying network, necessary and/or sufficient conditions for which average consensus can be achieved are then provided in terms of the null space of matrix-valued Laplacian of the corresponding integral network. In particular, for periodic matrix-weighted time-varying networks, necessary and sufficient conditions for reaching average consensus is obtained from an algebraic perspective. Moreover, we show that if the integral network with period $T>0$ has a positive spanning tree over the time span $[0,T)$, average consensus for the node states is achieved. Simulation results are provided to demonstrate the theoretical analysis.

Motivation & Objective

• To address the lack of consensus conditions for time-varying matrix-weighted networks, where scalar-weighted network results do not apply due to matrix-valued edge weights.
• To extend consensus theory to networks with matrix-valued weights that capture multi-dimensional interdependencies among agents.
• To derive necessary and sufficient conditions for average consensus under mild switching assumptions on time-varying topologies.
• To provide both algebraic (null space of Laplacian) and graph-theoretic (positive spanning tree) conditions for consensus in periodic matrix-weighted networks.

Proposed method

• Introduces the matrix-weighted integral network over a time interval to aggregate switching dynamics and analyze consensus behavior.
• Defines the matrix-valued Laplacian of the integral network to characterize system dynamics and consensus convergence.
• Uses spectral analysis of the Laplacian matrix to derive conditions based on the null space, particularly focusing on the alignment of agent states with the consensus subspace.
• Applies graph-theoretic concepts such as positive spanning trees in the integral network to establish sufficient conditions for consensus.
• Employs the concept of positive definiteness and semi-definiteness of edge weight matrices to classify strong and weak connections in the network.
• Derives necessary and sufficient algebraic conditions for average consensus in periodic networks by analyzing the Laplacian over one period.

Experimental results

Research questions

• RQ1 What conditions ensure average consensus in time-varying matrix-weighted networks where edge weights are symmetric matrices?
• RQ2 How does the structure of the integral network—formed by aggregating time-varying topologies—determine consensus behavior?
• RQ3 What is the role of the null space of the matrix-valued Laplacian in determining consensus convergence?
• RQ4 For periodic matrix-weighted networks, what graph-theoretic condition is both necessary and sufficient for average consensus?
• RQ5 Can a positive spanning tree in the integral network over a period guarantee average consensus?

Key findings

• Average consensus is achieved if and only if the null space of the matrix-valued Laplacian of the integral network contains the consensus subspace, i.e., the vector of all ones in the extended state space.
• For periodic matrix-weighted time-varying networks with period $T>0$, a necessary and sufficient algebraic condition for average consensus is derived based on the Laplacian of the integral network over $[0,T)$.
• A graph-theoretic condition is established: if the integral network over one period has a positive spanning tree, then average consensus is achieved.
• The simulation results confirm that consensus is reached at $[0.6958, 0.3382]^T$, consistent with the theoretical prediction, even when individual network topologies do not individually support consensus.
• The null space analysis of individual Laplacians $L( ilde{oldsymbol{G}}_1)$, $L( ilde{oldsymbol{G}}_2)$, and $L( ilde{oldsymbol{G}}_3)$ does not guarantee consensus, but the integral network’s null space does, demonstrating the necessity of the integral network framework.
• The presence of positive definite matrices as edge weights contributes to the formation of a positive spanning tree in the integral network, which is critical for consensus.