Von Neuman and Predicting the Weather

I remember a talk that Von Neumann gave at Princeton around 1950, describing the glorious future which he then saw for his computers. Most of the people that he hired for his computer project in the early days were meteorologists. Meteorology was the big thing on his horizon. He said, as soon as we have good computers, we shall be able to divide the phenomena of meteorology cleanly into two categories, the stable and the unstable. The unstable phenomena are those which are upset by small disturbances, the stable phenomena are those which are resilient to small disturbances. He said, as soon as we have some large computers working, the problems of meteorology will be solved. All processes that are stable we shall predict. All processes that are unstable we shall control. He imagined that we needed only to identify the points in space and time at which unstable processes originated, and then a few airplanes carrying smoke generators could fly to those points and introduce the appropriate small disturbances to make the unstable processes flip into the desired directions. A central committee of computer experts and meteorologists would tell the airplanes where to go in order to make sure that no rain would fall on the Fourth of July picnic. This was John von Neumann's dream. This, and the hydrogen bomb, were the main practical benefits which he saw arising from the development of computers. {183}

The meteorologists who came to work with Von Neumann knew better. They did not believe in the dream. They only wanted to understand the weather, not to control it. They had a hard enough time trying to understand it. They tried especially hard to predict a particular hurricane which came up from the Gulf of Mexico and passed close by Princeton in the fall of 1949. Again and again they set the initial conditions in the Gulf of Mexico and tried to predict the hurricane for several years after it happened. So far as I remember, they never did succeed in getting the hurricane on the computer to end up anywhere near Princeton. Now another thirty-five years have gone by, and we have built four more generations of computers, and still we are not doing very well with the prediction of hurricanes. Nobody any longer believes seriously in the possibility of controlling the weather.

What went wrong? Why was Von Neumann's dream such a total failure? The dream was based on a fundamental misunderstanding of the nature of fluid motions. It is not true that we can divide fluid motions cleanly into those that are predictable and those that are controllable. Nature is as usual more imaginative than we are. There is a large class of classical dynamical systems, including non-linear electrical circuits as well as fluids, which easily fall into a mode of behavior that is described by the word "chaotic." A chaotic motion is generally neither predictable nor controllable. It is unpredictable because a small disturbance will produce exponentially growing perturbation of the motion. It is uncontrollable because small disturbances lead only to other chaotic motions and not to any stable and predictable alternative. Von Neumann's mistake was to imagine that every unstable motion could be nudged into a stable motion by small pushes and pulls applied at the right places. The same mistake is still frequently made by economists and social planners, not to mention Marxist historians.

Freeman J. Dyson, Infinite in All Directions (2004-07-22)

• chaos theory
• prediction