Power grids are inherently prone to big blackouts, say US scientists. Trying to make them more robust can make the problem worse.

There is a power cut every 13 days in the United States. Indeed, over the past 15 years, large-scale blackouts there have been more frequent than can be explained by random damage. So Benjamin Carreras of Oak Ridge National Laboratory in Tennessee and co-workers have devised mathematical models of a network of generators and consumers¹.

The models indicate that as soon as the demand for power exceeds a network's total capacity, the only way the network can cope is by 'load shedding'. In other words, when everyone switches on their washing machines at the same time, or when a storm breaks a local line, a partial blackout is inevitable.

At first, the extent of the blackout increases steadily as the demand rises. Then, as power demand shifts onto generators that are still working, this can overload other transmission lines, causing a sudden avalanche of load shedding as many lines fail. Carreras' team reckons that the eventual large-scale breakdown looks like what physicists call a 'critical point'.

For a simple network, where the connections look like branches of a tree, the switch is easy to see. But the US power grid doesn't really look like this. So the team tested their model on a network of 118 generators and loads representing a simplified version of the real grid. The behaviour is then a bit more erratic, but still shows the same basic properties: a two-stage breakdown ending in a sudden, catastrophic change.

So the blackouts seem to happen in much the same way irrespective of the precise structure of the power network. "It's difficult to do anything to change it," says Carreras. And safety measures might make things worse. If they are designed to eliminate small blackouts, they can hasten the build-up to the point where a big cascade of blackouts becomes inevitable.

Countries that are less developed than the United States, Carreras points out, typically suffer many small blackouts, but rarely incur big ones. In effect, the tension in the network gets released in many small jolts rather than a big paroxysm. But it would be hard to persuade power companies to permit such small failures, Carreras admits.