Physicists find that a little disorder makes complex networks more stable, not less

For decades, engineers building power grids, materials scientists designing metamaterials and theorists modeling brains and ecosystems have shared a common instinct: make the parts as alike as possible, and the system will hold together. A study published Sept. 17 in the journal Science by physicists at Northwestern University argues that instinct is often wrong.

The team, led by physics professor Adilson Motter, developed a mathematical framework showing that deliberately introducing differences — what physicists call disorder, heterogeneity or asymmetry — among the components of a network or the connections between them can make the whole system more stable, not less. The effect held up across five very different kinds of networks the researchers modeled: power grids, neuronal circuits, flocking and drone-swarm systems, architected materials and ecological food webs.

What the numbers show

The Northwestern University announcement of the findings describes two distinct routes by which disorder helps: variation among a network's nodes, and variation among the links that connect them. In the models the team tested, moderate amounts of either kind of difference improved a network's ability to absorb small disturbances and return to a stable state. Push the disorder too far, however, and the same systems became less stable, not more — meaning the benefit follows a curve rather than a straight line, with an optimum somewhere between rigid uniformity and chaos.

One striking result: even disorder introduced at random, with no attempt to engineer where or how much variation to add, improved stability compared with a perfectly uniform baseline in many of the models. That suggests real-world systems do not need to be finely tuned to benefit — they may already be gaining stability simply because their components are not identical.

Why uniformity became the default assumption

The idea that sameness equals stability traces in part to simplified mathematical tools long used to study synchronization, such as the Kuramoto model, which represents each node in a network with only a single dynamical variable. According to the Northwestern release, those stripped-down models cannot capture the richer, multi-variable behavior of real components — and in doing so, they inadvertently erased a stabilizing effect that more realistic systems actually exhibit. Postdoctoral researcher Arthur Montanari, a co-first author of the paper along with graduate student Pietro Zanin, put it plainly: "Although these models have provided important insights, they can miss effects that emerge in real systems, where components and interactions have richer dynamics."

The new work builds on scattered earlier evidence that disorder can help. A 2020 study in Nature Physics found that power generators synchronized more reliably when they differed slightly from one another rather than running as identical units, and a 2025 Nature Communications study co-authored by Montanari found a similar stabilizing effect in models of flocking and drone swarms. What had been missing was a general theory explaining when and why the effect appears — and how far it can be pushed before it backfires.

The paper also speaks to an older puzzle in ecology. In 1972, the biologist Robert May showed mathematically that large, complex ecosystems should become less stable as the number of interacting species grows — a conclusion sharply at odds with the observed persistence of species-rich systems such as coral reefs and rainforests, a mismatch ecologists have argued over for more than 50 years. The Northwestern team's framework suggests that variation in the strength of interactions between species, such as between pollinators and the flowers they visit, may be part of what allows real, diverse ecosystems to persist where May's uniform models predicted collapse.

"Real systems are rarely uniform. Birds differ in personalities, neurons vary in shape and even our social relationships can be asymmetric." — Arthur Montanari, Northwestern University

Who stands to use this

The clearest near-term audience is power-grid engineering. Electrical grids are already becoming less uniform as utilities add wind turbines, rooftop solar and battery storage alongside traditional generators of varying size and behavior — a shift often treated as an engineering headache to be managed rather than a source of resilience. The new framework offers a formal basis for the opposite view: that some of this heterogeneity could be preserved or even designed in deliberately to make grids more robust to disturbances, rather than smoothed away in pursuit of uniformity.

The researchers point to similar openings in architected materials, where varying the shape, size or orientation of repeating structural units could unlock mechanical properties uniform lattices cannot achieve, and in multi-agent robotics, where swarms of drones or other autonomous vehicles might be built or programmed with intentional variation rather than identical units. Neuroscientists modeling how the brain computes and ecologists studying food-web resilience were also named, in a summary of the findings carried by Mirage News, among the fields the team sees as natural test beds for the theory.

Reaction and independent review

The paper appears alongside a companion Perspective in the same issue of Science, commissioned from Raissa D'Souza, a professor at the University of California, Davis and a member of the journal's Board of Reviewing Editors, who works on nonlinear dynamics and network theory. Commentaries of that kind are typically invited by Science's editors specifically to put a new result in context for specialists outside the immediate research group, a step that functions as a form of independent vetting ahead of publication.

Motter, who directs Northwestern's Center for Network Dynamics, framed the result as resolving a gap between theory and observation that has nagged network scientists for years. "Disorder can stabilize networks, but only when the node dynamics are rich enough," he said, according to the university's announcement, adding that earlier work had turned up a "growing number of cases" of disorder improving real-world systems without anyone establishing how general the phenomenon was. Montanari, describing the trade-off the team identified, said: "If you make the system more homogeneous, you lose stability. But if you increase disorder too much, you also lose stability."

What happens next

The study, distributed to reporters through EurekAlert, was funded by the Army Research Office and the National Science Foundation, including support through the NSF-Simons National Institute for Theory and Mathematics in Biology, a joint program between the NSF and the Simons Foundation. The Motter group has also released an interactive tool that lets other researchers adjust the framework's parameters and watch how different levels and placements of disorder affect a model network's stability.

The immediate task for engineers and biologists, according to the paper's authors, is translating a general mathematical principle into system-specific guidance: how much disorder helps a particular power grid, drone swarm or ecosystem, and where it should be concentrated rather than spread evenly. The framework identifies that a stabilizing window exists and describes its mathematical boundaries, but applying it to a specific real grid or a specific material will require follow-up work matching the theory to real components, whose dynamics are typically more complicated than the simplified models used to establish the general result. Laboratory and field tests of the principle in physical hardware, rather than in simulation, are expected to follow as engineers in the affected fields take up the framework.