A Design Method of Distributed Algorithms via Discrete-time Blended Dynamics Theorem

TL;DR

This paper introduces a discrete-time blended dynamics theorem for designing distributed algorithms, enabling predictable behavior of heterogeneous multi-agent systems through multi-step coupling, demonstrated in three network applications.

Contribution

It extends the blended dynamics theorem to discrete-time systems using multi-step coupling, broadening its applicability for designing distributed algorithms.

Findings

Predicts behavior of heterogeneous multi-agent systems.

Enables design of node dynamics for specific tasks.

Demonstrates applications in network size estimation, PageRank, and degree sequence computation.

Abstract

We develop a discrete-time version of the blended dynamics theorem for the use of designing distributed computation algorithms. The blended dynamics theorem enables to predict the behavior of heterogeneous multi-agent systems. Therefore, once we get a blended dynamics for a particular computational task, design idea of node dynamics for individual heterogeneous agents can easily occur. In the continuous-time case, prediction by blended dynamics was enabled by high coupling gain among neighboring agents. In the discrete-time case, we propose an equivalent action, which we call multi-step coupling in this paper. Compared to the continuous-time case, the blended dynamics can have more variety depending on the coupling matrix. This benefit is demonstrated with three applications; distributed estimation of network size, distributed computation of the PageRank, and distributed computation of the degree sequence of a graph, which correspond to the coupling by doubly-stochastic, column-stochastic, and row-stochastic matrices, respectively.