20 SEP 2026 — Engineers building networked systems generally try to make the parts as alike as possible. A Northwestern University paper in Science, published on 17 September, argues that uniformity is often what makes those systems fragile. The claim is not that mess is good. It is that a moderate, deliberately placed amount of variation can stabilise a network that identical components would destabilise.
What the framework claims
Adilson Motter, Arthur Montanari and Pietro Zanin published a mathematical framework for identifying when heterogeneity helps. Their paper appeared on 17 September under the title "Disorder-promoted stability".
Disorder here means variation among the components of a network or among the connections between them. The framework sets out when that variation increases stability instead of eroding it.
Motter is the Charles E. and Emma H. Morrison Professor of Physics and Astronomy at Northwestern's Weinberg College. Montanari and Zanin, a postdoctoral researcher and a graduate student, are co-first authors.
Where it was tested
The framework was applied across models of power grids, neurons, flocking animals, drone swarms, architected materials and ecological food webs.
That spread is the point. A result about power grids alone would be an engineering finding. The same mathematics holding across neurons and flocks is a claim about networks as a class, independent of what the nodes happen to be.
The condition that matters
The finding comes with a restriction the headline version tends to lose. "Disorder can stabilize networks, but only when the node dynamics are rich enough," Motter said. The individual units have to have enough internal behaviour for variation between them to do anything useful. In a network whose nodes are simple, the effect does not appear, which is why it went unnoticed in the simplified models this field has often relied on.
There is also a ceiling. Increase disorder too much, Motter noted, and you also lose stability. The claim is about a moderate and well-placed amount, not a direction of travel.
Links versus nodes
One exception is worth separating out. When the disorder sits in the connections rather than in the nodes, even networks with simple node dynamics can benefit. That has practical weight because links are frequently the cheaper thing to vary. Changing the components of a power grid or a swarm is expensive; changing how strongly they are coupled is often a matter of configuration. The paper does not set out how to design beneficial disorder patterns, which the authors name as future work.
What the release does not give
The announcement published no quantitative thresholds. It gives no figure for how much disorder is optimal, no measured stability gain, and no numbers for any of the six system classes. The Northwestern release describes the effect qualitatively.
The authors also state that the optimal level of disorder varies by system, that excessive disorder destabilises, and that simplified mathematical models can miss the stabilising effect entirely.
A general condition has been identified and demonstrated across model systems, but how much variation to build into any particular grid or swarm remains unanswered. Treat the paper as a result about when to look, not yet as a design rule.
View author profile → · Editorial policy
