Gardner volumes and self-organization in a minimal model of complex ecosystems
Frederik J. Thomsen, Johan L.A. Dubbeldam, Rudolf Hanel · 2026
Read Gardner volumes and self-organization in a minimal model of complex ecosystems by Frederik J. Thomsen, Johan L.A. Dubbeldam, Rudolf Hanel (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2601-00707.txt.
Introduction
Frederik J Thomsen and two co-authors study how large ecosystems sort themselves out, in a 2026 paper. Their model is a stripped-down one: many species with set links, in a random web. Each species grows or shrinks by a simple rule, and none can drop below zero.
The authors show that the states the system can sit in form shrinking zones. These zones match a tool from neural network theory called Gardner volumes. As the share of live species falls, the zones shrink fast. The authors tie that shrink to shifts in the link web over time. The tests say whether all life dies out or growth runs wild. For the wild case, a small tweak to the model leads to a calm end state instead. The work links ecosystem fate to the size of its room to move.
Catalog record
| CREATOR | Frederik J. Thomsen, Johan L.A. Dubbeldam, Rudolf Hanel |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 86 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2601-00707
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
Gardner volumes and self-organization in a minimal model of complex ecosystems Preprint: APS/123-QED Frederik J. Thomsen, Johan L.A. Dubbeldam, Rudolf Hanel Abstract We study self-organization in a minimally nonlinear model of large random ecosystems. Populations evolve over time according to a piecewise linear system of ordinary differential equations subject to a non-negativity constraint resulting in discrete time extinction and revival events. The dynamics are generated by a random elliptic community matrix with tunable correlation strength. We show that, independent of the correlation strength, solutions of the system are confined to subsets of the phase space that can be cast as time-varying Gardner volumes from the theory of learning in neural networks. These volumes decrease with the diversity (i.e. the fraction of extant species) and become exponentially small in the long-time limit. Using standard results from random matrix theory, the changing diversity is then linked to a sequence of contractions and expansions in the spectrum of the community matrix over time, resulting in a sequence of May-type stability problems determining whether the total population evolves toward complete extinction or unbounded growth. In the case of unbounded growth, we show the model allows for a particularly simple nonlinear extension in which the solutions instead evolve towards a new attractor. I Introduction Complex systems consisting of many interacting components emerge naturally in the study of ecological communities. Depending on the context, what these components represent ranges from the populations of animal Hofbauer and Sigmund 1998 , molecular Segel 1984 or microbial Smith and Waltman 1995 species to groups of individuals in an epidemic Brauer et al. 2012 . The list of processes governing the interactions between components is enormous, including competition, cooperation, predation, mutualism, and commensalism Harte 2011 . Dynamical systems used to describe a subset of these processes are often cast either as variants of generalised Lotka-Volterra differential equations Lotka 1920 ; Volterra 1926 or as (piecewise) linear differential equations. The latter type arises more specifically in gene-regulatory or biochemical reaction networks, where the components represent molecular or microbial species, such as in the works of Glass-Kauffman Glass and Kauffman 1973 and Jain-Krishna Jain and Krishna 1998 ; however, (piecewise) linear systems can also appear as a linearisation in the neighborhood of an attractor Stokić et al. 2008 ; Belga Fedeli et al. 2021 . The central mathematical object in generating the dynamics for both types of system is the community matrix that encodes the interactions between components. When the ecological community is large and the corresponding community matrix high-dimensional, precise knowledge of the interactions is impossible to obtain. This has popularized the statistical physics approach to ecology pioneered by Robert May May 1972 , in which the community matrix is assumed to be random. This randomness has since been leveraged in countless studies to derive emergent macroscopic properties for the long-time dynamics. For example, phase diagrams Bunin 2017 , the typical number of equilibria Ros et al. 2023 and the number of surviving species at equilibrium Clenet et al. 2023 ; Serván et al. 2018 , as well as abundance distributions for non-equilibrium solutions Arnoulx de Pirey and Bunin 2024 ; Mallmin et al. 2024 , have been investigated; for a recent review see also Akjouj et al. 2024 . Independently of the model context, the interacting components often only have physical significance when their values are nonnegative. In contrast to Lotka-Volterra systems, linear systems must therefore generally be formulated in a piecewise, discontinuous manner to self-consistently enforce that solutions remain nonnegative at all times Casey et al. 2006 ; Jain and Krishna 2001 . In the random matrix approach, aligning the feasibility constraint with existence criteria for equilibrium solutions is already an interesting problem. For Lotka-Volterra systems, classical existence results for globally attracting, non-negative equilibrium solutions require strong dissipativity assumptions on the dynamics Takeuchi 1996 , which translate to strong assumptions on the spectrum of the community matrix Clenet et al. 2022 . Positive equilibria in particular, where all species coexist, have been shown to require precise fine-tuning of the interactions Grilli et al. 2017 . A mechanism to explain how, nevertheless, large ecological communities stably coexist, is to view them as survivors selected from a much larger species pool, either by a sequence of invasion processes (see e.g. Morton et al. 1996 ) or by initializing the system in a point of coexistence and evaluating the species remaining in the long-time limit Serván et al. 2018 ; Bunin 2017 . In the latter case, for Lotka-Volterra systems, the community matrix evaluated at a stable equilibrium is a submatrix of the full set of possible interactions, “pruned” by the dynamics. More specifically, Servan et al. Serván et al. 2018 show that the expected number of coexisting species in a globally attracting equilibrium is 1/2 of the total species pool. In Pettersson et al. 2020 , the authors instead consider changes to the community matrix after continuation of the equilibrium in the standard deviation of the interactions. A related mechanism of self-organization specific to piecewise linear differential equations was explored in a so-called minimally nonlinear model for chemical reaction networks Stokić et al. 2008 . In contrast to Lotka-Volterra systems, in this model, species may go extinct in finite time and be revived at a later time through catalytic reactions with the extant species. The revival mechanism here replaces classical species invasion criteria (see e.g. Akjouj et al. 2024 ). Rather than at an equilibrium and under strong dissipativity assumptions, the submatrix for the remaining active species thus changes over time and can be re-evaluated after each extinction or revival event. The system was shown to self-organize near an “inflated edge of chaos” Hanel et al. 2010 , characterized by a broad plateau of vanishing maximal Lyapunov exponent, where always approximately 1/2 of the total species pool remains active. In this work we revisit the the minimally nonlinear model for random community matrices parametrized by a correlation parameter, \xi . This parameter smoothly interpolates between non-reciprocal ( \xi<0 ) and reciprocal ( \xi>0 ) interactions. Our focus lies on determining analytically the mechanisms behind the systems ability to self-organize. Because the system is piecewise linear, the dynamics can be decomposed into angular and radial variables, measuring the species composition and total population, respectively. The dynamics of the angular variables determine the sequence of changes made to the community matrix through revival and extinction events, independently of the total population dynamics. Our main result is that solutions for the angular dynamics are confined to subsets of the phase-space which can be cast as time-dependent “Gardner volumes” from the theory of learning in neural networks Gardner 1988 ; Amit et al. 1989 . These volumes are shown to decrease with the system diversity (given by the fraction of extant species) and become exponentially small as the diversity approaches a critical threshold. Numerically we find that solutions evolve to this critical regime in the long-time limit, thereby becoming highly sensitive to perturbations. While the long-time behavior of solutions depends on the correlation strength, the volumes themselves are independent of this parameter. When the sequence of extinction and revival events is finite, the community matrix is adapted in such a way that the leading eigenvector