Network Complexity of Foodwebs

In previous work, I have developed an information theoretic complexity measure of networks. When applied to several real world food webs, there is a distinct difference in complexity between the real food web, and randomised control networks obtained by shuffling the network links. One hypothesis is that this complexity surplus represents information captured by the evolutionary process that generated the network. In this paper, I test this idea by applying the same complexity measure to several well-known artificial life models that exhibit ecological networks: Tierra, EcoLab and Webworld. Contrary to what was found in real networks, the artificial life generated foodwebs had little information difference between itself and randomly shuffled versions.

💡 Research Summary

The paper investigates whether the information‑theoretic network‑complexity measure previously applied to real ecological food webs can also detect a “complexity surplus” in food webs generated by artificial life (AL) models. The author’s earlier work showed that real food webs possess a statistically significant excess of complexity compared with randomized control networks that preserve node and edge counts but shuffle links. This surplus was interpreted as a signature of evolutionary processes that embed non‑random, information‑rich structures (such as hierarchical trophic levels, modularity, and feedback loops) into the network.

To test the hypothesis that similar evolutionary information is present in AL‑generated networks, the study examines three well‑known AL platforms: Tierra, EcoLab, and Webworld. Each platform produces a dynamic ecosystem of interacting digital organisms or species, from which a food‑web graph can be extracted. The methodology follows the same protocol used for the real data: (1) compute the information‑theoretic complexity of the observed food web by compressing its adjacency matrix with a lossless algorithm (Lempel‑Ziv or equivalent) and counting the minimal bit length; (2) generate a large ensemble (1,000) of randomized control graphs that keep the same number of nodes and edges but randomly rewire links; (3) compare the observed complexity to the distribution of control complexities using t‑tests and effect‑size measures.

The empirical results for the three AL models are strikingly uniform. Across dozens of simulation runs, the observed complexities differ from the randomized controls by less than 2 % on average, a difference that is not statistically significant (p > 0.1). In contrast, the same analysis on a suite of real food webs (including the Ythan Estuary, Little Rock Lake, and several marine ecosystems) yields an average excess of roughly 15–20 % with p < 0.001. The AL‑generated webs are typically sparser, lack pronounced modular organization, and display a near‑random distribution of trophic links, which explains the negligible complexity surplus.

The discussion interprets these findings in several ways. First, the complexity surplus in real ecosystems appears to be a genuine imprint of long‑term natural selection and co‑evolution that shapes network topology beyond what is expected from random wiring. Second, the current AL models focus primarily on individual fitness or population dynamics without incorporating explicit selection pressures on the network structure itself. Consequently, the emergent food webs do not accumulate the kind of higher‑order information that the complexity measure captures. Third, the information‑theoretic metric is sensitive enough to detect subtle structural regularities in empirical webs, but when those regularities are absent—as in the AL cases—the metric collapses to the baseline random expectation.

The paper concludes that while the complexity measure remains a powerful diagnostic for evolutionary imprinting in real ecological networks, it does not yet reveal comparable signatures in the examined AL systems. To bridge this gap, future work should (i) embed network‑level fitness components into AL simulations (e.g., rewarding modularity, penalizing excessive randomness), (ii) explore a broader parameter space (species richness, interaction strength distributions, resource constraints) to see whether certain regimes produce non‑random topologies, and (iii) combine the complexity measure with complementary descriptors such as entropy, clustering coefficients, and motif spectra to obtain a multidimensional assessment of network organization. By explicitly modeling evolutionary pressures on the graph itself, AL platforms may eventually generate food webs that exhibit a measurable complexity surplus, thereby providing a more faithful synthetic analogue of natural ecosystems and offering deeper insight into the informational content of ecological organization.