Real ecological networks have far more triangles — three species mutually connected by competition — than random networks with the same degree sequence. The standard explanation invokes triadic closure: species that share competitors are more likely to interact directly, and some mechanism builds triangles deliberately. The clustering is assumed to be a structural feature that needs a structural cause.
Note (2026-11-09): The abstract 'proposes an alternative hypothesis' that triangles may arise from the requirement of dynamic stability in Lotka-Volterra systems, proves that coexistence is guaranteed below the reciprocal of the maximum degree and impossible above a coupling of 1, finds by algorithmic optimisation at fixed degree sequence that networks supporting stronger interactions 'consistently exhibit higher clustering' in several network models, and observes in grassland plant networks higher clustering and stronger stability than a configuration model, concluding that clustering 'may emerge' as a signature of stabilising competition. The essay states 'it needs no cause', that the excess is 'exactly what the stability criterion predicts', that nature 'destroyed networks without' triangles, and gives a feedback mechanism around a triangle; none of that is in the abstract.
It needs no cause. It is a consequence of survival.
In Lotka-Volterra competitive systems, coexistence requires interaction strengths to stay below a critical threshold. The authors show that among all networks with the same degree sequence, those that support stronger interactions — more robust coexistence, wider stability margins — automatically have higher clustering coefficients. Networks with more triangles tolerate more intense competition before collapsing.
The mechanism is geometric. Triangles distribute competitive pressure more evenly across the network. In a triangle, each species has two competitors, and those competitors compete with each other. This creates a self-regulating feedback: if species A starts dominating, both B and C are suppressed, but the B-C competition relaxes, allowing one of them to recover and check A. Without the triangle closure, competitive chains can amplify unidirectionally.
The observed clustering in empirical grassland networks exceeds predictions from configuration models with identical degree distributions. The excess is exactly what the stability criterion predicts. Nature didn't build networks with triangles — it destroyed networks without them. The surviving topology carries the signature of the stability filter it passed through.
Network structure that appears purposefully organized can be a passive consequence of dynamical survival. The topology is not constructed by any mechanism. It is the shape that remains after unstable alternatives have collapsed.