Stable coexistence in indefinitely large systems of competing species

TL;DR

This paper demonstrates that in certain ecological networks, large systems of competing species can remain stable even with strong interactions, challenging the universal instability suggested by earlier models.

Contribution

The study shows stability in large competitive Lotka-Volterra systems on specific networks, refining May's bounds by linking stability to network degree distribution.

Findings

Stability persists in large systems with non-vanishing interactions.

Network structure, especially degree distribution, influences stability.

Large degree nodes can cause instability, not the system size.

Abstract

The Lotka-Volterra system is a set of ordinary differential equations describing growth of interacting ecological species. This model has gained renewed interest in the context of random interaction networks. One of the debated questions is understanding how the number of species in the system, $n$, influences the stability of the model. Robert May demonstrated that large systems become unstable, unless species-species interactions vanish. This outcome has frequently been interpreted as a universal phenomenon and summarised as "large systems are unstable". However, May's results were performed on a specific type of graphs (Erd\H{o}s-R\'enyi), whereas we explore a different class of networks and we show that the competitive Lotka-Volterra system maintains stability even in the limit of large $n$, despite non-vanishing interaction strength. We establish a lower bound on the interspecific interaction strength, formulated in terms of the maximum and minimum degrees of the ecological network, rather than being dependent upon the network's size. For values below this threshold, coexistence of all species is attained in the asymptotic limit. In other words, the outlier nodes with large degree cause instability, rather than the large number of species in the system. Our result refines May's bound, by showing that the type of network model is relevant and can lead to completely different results.