In a 'scale-free' network of many interconnected nodes, like the Internet, most of the nodes are connected to a relatively small number of other nodes. Only a very small minority have a large number of connections. It is therefore extremely unlikely that randomly failing links would have a catastrophic effect on the whole network. In contrast, an intelligent attack on the few highly connected nodes could be devastating.

Barabasi and co-workers studied the effect that removing random nodes from a scale-free network had on the ability of the remaining nodes to communicate with each other, and the degree to which the network became fragmented. They found that the network's performance remained constant, even after they had removed as many of 5% of the nodes, and that it was resistant to fragmentation. But when the team simulated an intelligent attack by targeting the highly connected nodes, it was a different story: the network became fragmented very quickly, and with 5% of the nodes missing, its ability to communicate was halved.

The team applied tools and ideas from statistical mechanics to the Internet. "I believe that for many complex systems, we have to first understand the topology that describes how the diverse constituents interact with each other", Barabasi told PhysicsWeb. "This is fundamentally a physics problem, since it involves randomness and self-organization living side by side, and that is best addressed by the tools of statistical mechanics."

Error tolerance may come at the expense of reduced robustness in a scale-free network like the Internet, but Barabasi and colleagues point out that this peculiar feature can be exploited in scale-free systems such as metabolic networks, where drug design can target vulnerable points. But it is not a promising development for the Internet.