Consensus Driven by the Geometric Mean

Herbert Mangesius, Dong Xue, Sandra Hirche

Abstract

Consensus networks are usually understood as arithmetic mean driven dynamical averaging systems. In applications, however, network dynamics often describe inherently non-arithmetic and non-linear consensus processes. In this paper, we propose and study three novel consensus protocols driven by geometric mean averaging: a polynomial, an entropic, and a scaling-invariant protocol, where terminology characterizes the particular non-linearity appearing in the respective differential protocol equation. We prove exponential convergence to consensus for positive initial conditions. For the novel protocols we highlight connections to applied network problems: The polynomial consensus system is structured like a system of chemical kinetics on a graph. The entropic consensus system converges to the weighted geometric mean of the initial condition, which is an immediate extension of the (weighted) average consensus problem. We find that all three protocols generate gradient flows of free energy on the simplex of constant mass distribution vectors albeit in different metrics. On this basis, we propose a novel variational characterization of the geometric mean as the solution of a non-linear constrained optimization problem involving free energy as cost function. We illustrate our findings in numerical simulations.

Introduction

In network theory, the concept of "consensus" describes the process by which a group of autonomous agents reaches a common value through local interactions. Traditionally, this is modeled as a linear system where agents move toward the arithmetic mean of their neighbors. While this framework is robust, it is often insufficient for systems that operate on multiplicative or power-law scales, such as gene expression in biology, compound growth in finance, or specific dynamics in power grids.

This paper addresses the challenge of designing and proving the stability of continuous-time protocols that drive a network toward a geometric consensus. By shifting the target from the arithmetic to the geometric mean, the authors provide a more natural mathematical framework for non-linear data relationships.

Proposed Consensus Protocols

The authors introduce three distinct non-linear protocols, each defined by the specific nature of the non-linearity in their differential equations:

  1. Polynomial Protocol: This protocol is structured similarly to chemical reaction kinetics. It models the rate of change as a balance between "in-flows" and "out-flows," represented by products of states raised to specific weights. It bridges the gap between social/robotic consensus and biochemistry.
  2. Entropic Protocol: Driven by negative relative entropy (information divergence), this protocol moves agents based on the logarithmic difference between their current state and those of their neighbors. It is shown to converge exactly to the weighted geometric mean of the initial states.
  3. Scaling-Invariant Protocol: This is a log-linear update rule that mimics the behavior of a standard linear consensus system but operates within logarithmic coordinates.

Theoretical Framework and Convergence

To establish the stability of these systems, the authors utilize a "Virtual Graph" perspective. By transforming non-linear, time-invariant equations into "virtual" linear time-varying consensus systems, they apply established stability results from linear theory. This approach allows for a rigorous proof that for any positive initial conditions, all three protocols achieve global exponential convergence to a consensus state.

Furthermore, the research reveals that these protocols are not merely heuristic; they are gradient flows of free energy. The network essentially performs a form of optimization, "sliding down" an energy surface toward the most efficient state. This leads to a novel variational characterization: the geometric mean is the unique solution that minimizes free energy on a manifold where the product of all components remains constant.

Numerical Validation

In the absence of empirical datasets, the theoretical findings were validated through numerical simulations across various synthetic graph structures:

  • Strongly connected balanced digraphs: Where incoming and outgoing influences are equal.
  • Non-balanced digraphs: Used to test the "weighted" versions of the geometric mean.
  • Complete graphs: Utilized to study the specific bounds of the polynomial protocol.

Results confirmed that while the Entropic Protocol targets the weighted geometric mean, the Polynomial Protocol's consensus value remains bounded between the arithmetic mean and the "arithmetic-geometric mean" (a value related to elliptic integrals).

Conclusion and Future Work

This work expands the toolkit of network consensus by providing a rigorous foundation for geometric averaging. However, certain limitations remain. The protocols currently require strictly positive initial conditions ($x > 0$), which, while appropriate for chemical concentrations or financial indices, may limit broader applications. Additionally, the exact analytical consensus value for the polynomial protocol remains a conjecture, likely involving complex elliptic integrals. Future research will focus on extending this variational framework to time-varying graph topologies and more radical network switching scenarios.

Summary updated

Abstract

Consensus networks are usually understood as arithmetic mean driven dynamical averaging systems. In applications, however, network dynamics often describe inherently non-arithmetic and non-linear consensus processes. In this paper, we propose and study three novel consensus protocols driven by geometric mean averaging: a polynomial, an entropic, and a scaling-invariant protocol, where terminology characterizes the particular non-linearity appearing in the respective differential protocol equation. We prove exponential convergence to consensus for positive initial conditions. For the novel protocols we highlight connections to applied network problems: The polynomial consensus system is structured like a system of chemical kinetics on a graph. The entropic consensus system converges to the weighted geometric mean of the initial condition, which is an immediate extension of the (weighted) average consensus problem. We find that all three protocols generate gradient flows of free energy on the simplex of constant mass distribution vectors albeit in different metrics. On this basis, we propose a novel variational characterization of the geometric mean as the solution of a non-linear constrained optimization problem involving free energy as cost function. We illustrate our findings in numerical simulations.
BibTeX
@article{mangesius2018consensus,
  author = {Herbert Mangesius and Dong Xue and Sandra Hirche},
  title = {{Consensus Driven by the Geometric Mean}},
  journal = {IEEE Trans. Control. Netw. Syst.},
  year = {2018},
  doi = {10.1109/tcns.2016.2598665},
}
Figure