Topological constraints on self-organisation in locally interacting systems

TL;DR

This paper investigates how the topology of interaction networks influences the ability of locally interacting systems to achieve and maintain ordered, collective states, with implications for understanding biological and artificial systems.

Contribution

It establishes necessary topological conditions for spontaneous order in systems on planar graphs, using free energy scaling in three models to explain ordering capabilities.

Findings

Graph topology constrains spontaneous ordering in models.

Multiscale biological systems can organize into complex patterns.

Language models struggle with long sequences due to topological constraints.

Abstract

All intelligence is collective intelligence, in the sense that it is made of parts which must align with respect to system-level goals. Understanding the dynamics which facilitate or limit navigation of problem spaces by aligned parts thus impacts many fields ranging across life sciences and engineering. To that end, consider a system on the vertices of a planar graph, with pairwise interactions prescribed by the edges of the graph. Such systems can sometimes exhibit long-range order, distinguishing one phase of macroscopic behaviour from another. In networks of interacting systems we may view spontaneous ordering as a form of self-organisation, modelling neural and basal forms of cognition. Here, we discuss necessary conditions on the topology of the graph for an ordered phase to exist, with an eye towards finding constraints on the ability of a system with local interactions to maintain an ordered target state. By studying the scaling of free energy under the formation of domain walls in three model systems -- the Potts model, autoregressive models, and hierarchical networks -- we show how the combinatorics of interactions on a graph prevent or allow spontaneous ordering. As an application we are able to analyse why multiscale systems like those prevalent in biology are capable of organising into complex patterns, whereas rudimentary language models are challenged by long sequences of outputs.