Differences Among a Network's Parts Can Make It More Stable, Not Less

Physicists at Northwestern University report that a standard conclusion in network science, that variation among a network's parts makes the whole less stable, is an artifact of simplified models and fails when the parts have richer internal behavior.

The authors write that their framework presents disorder as a general resource for stabilizing complex systems rather than a liability. Arthur N. Montanari, Pietro Zanin and Adilson E. Motter, all of the Center for Network Dynamics at Northwestern, published the work, titled Disorder-promoted stability, in Science on September 17, 2026.

Earlier work on network dynamics found that differences among nodes, the individual units a network connects, inhibit stability. The authors say this sits awkwardly with how common uneven natural and engineered systems are. Their paper traces the conclusion to model reductions made for mathematical convenience and reports that it breaks down when each node is described by several internal variables rather than one.

In those systems, the mathematics describing how a small disturbance grows or dies away loses a symmetry that the simplified models take for granted, a property the authors call non-Hermiticity. Variation among the nodes can then increase stability rather than erode it, the team reports, even when the parameters are picked at random.

The paper reports the effect in models of neural, power-grid and material networks. It also reports a stabilizing effect in models of ecological networks, where each node carries a single variable but the interactions are nonreciprocal, so one species affects another more than it is affected in return. These are model systems, not measurements of any particular brain, power grid or ecosystem.

The paper lists the National Science Foundation, the Army Research Office and the Simons Foundation as funders.