The Social Reactor
or, the strange mathematics of cities and slime molds
You can stand on a busy corner in any major city in the world — Manhattan or Mumbai or Tokyo or São Paulo — and watch ten thousand people pass through your line of sight in an hour, and not one of them is following a plan you can see. They are all going somewhere. They are all on time, mostly. The traffic flows, the trains run, the food is delivered to the restaurants, the electricity stays on, the garbage gets carried away, the babies get born and the dying get attended to and the music plays. Nobody is in charge. Nobody could be. The city is too large to be steered.
Cities are the most preposterous thing humans have built. They have no precedent in the natural world — not really. A termite mound is impressive but it caps out at maybe three meters tall and a few thousand termites; we routinely run cities of twenty million. A beaver pond has a particular geometry but the beaver does not keep adding to it. A coral reef has structure but it does not have rush hour. The city, once started, almost never stops growing until it runs out of water or political will, and even then it usually grows anyway. It is a kind of object the universe had not previously invented, and we are the first species to invent it.
For most of the twentieth century, urban planners and sociologists studied cities the way you would study a person — as a particular case with a particular history, where you read the local newspaper and interview the mayor and try to understand the personality of the place. Paris has its character; Detroit has its decline; Mumbai has its monsoons. This is fine for journalism. It is useless for science, because it offers no way to compare a thing to itself.
Around 2007, two researchers at the Santa Fe Institute — the physicist Geoffrey West and the urban scientist Luis Bettencourt — tried something different. They went looking for scaling laws.
A scaling law is the relationship between the size of a system and some property of the system, expressed as a power. The general form is:
Where N is the size (here, population), Y is whatever you are measuring, and β is the exponent that says how the property changes as the system grows. The interesting thing about scaling laws is what happens at the special value of β: when β = 1, the property is exactly proportional to size (a city twice as big has twice as much). When β < 1, the property grows slower than size (a city twice as big has less than twice as much). When β > 1, the property grows faster than size (a city twice as big has more than twice as much).
What West and Bettencourt found, looking at data from thousands of cities across many countries and many decades, was that cities scale according to surprisingly clean laws — and that there are exactly three regimes, depending on what you measure.
The first regime is sublinear, with β ≈ 0.85. Infrastructure quantities — total road length, total electrical wiring, total gas station count, total water pipe footage — grow more slowly than population. A city twice as big needs only about 85% more infrastructure. This is the economy of scale. Bigger cities are more efficient per person in their physical bones. New York City has fewer miles of road per resident than Boise has. Every megacity inherits this discount.
The second regime is linear, with β ≈ 1. Basic individual quantities — number of homes, water consumption per capita, employment count — grow proportionally to population. There’s no city-scale advantage or disadvantage; if you double the population, you double the houses. Boring, expected, and what most people assume should govern everything.
The third regime is superlinear, with β ≈ 1.15. And this is the surprising one, the one West and Bettencourt are now famous for. Socioeconomic outputs — GDP, total wages, patents filed, R&D employment, restaurant variety, even average walking speed on the sidewalk — grow faster than population. A city twice as big has more than twice as much. By the time you go from a town of 100,000 to a metropolis of 10,000,000, you have a hundredfold population but roughly two hundredfold GDP. The big city is more productive per person than the small one. Every city has this productivity bonus, and the size of the bonus is set by the same exponent in every country.
Crime is also superlinear. AIDS cases are superlinear. Patents and pollution and innovation and infectious disease all grow with the same exponent. Whatever it is that big cities do to people, they do more of it per person, in both the directions we like and the directions we don’t.
Compare this to biology. The pioneering work on biological scaling was done by Max Kleiber in the 1930s. He measured the metabolic rate of organisms across the entire size range from mice to elephants and found that metabolic rate scales as M3/4. An elephant is a few thousand times heavier than a mouse but uses metabolic energy only a few hundred times faster. Bigger animals are more efficient per unit of mass. Every animal in the kingdom inherits this discount.
The biological exponent is sublinear. The infrastructure exponent for cities is sublinear. There is reason to believe these are the same phenomenon — the math falls out of the geometry of fractal distribution networks (blood vessels, sewage pipes, electrical grids, the bronchial tree of the lung) optimizing the cost of moving stuff between a central source and a distributed periphery. West has spent thirty years arguing this point, and the math is good.
But the socioeconomic exponent for cities is superlinear, and that is a different story. There is no animal that gets more productive per cell as it grows. The only way to keep growing in biology is by inventing entirely new cell types and entirely new organs, and even then the metabolic discount keeps cutting against you. Cities don’t pay that bill. They get more productive faster than they get bigger. They are running on a different gradient than the one biology runs on.
Why? Because the productivity of a city is fundamentally about encounters — people meeting people, ideas meeting ideas, problems meeting unrelated solutions, the chance collisions of expertise that turn out to make a startup or a song or a cure. And encounters scale combinatorially. If you have N people, you have N(N−1)/2 possible pairs, which grows as N2. Per person, the number of available pairings grows linearly with N. The density of opportunity is itself a function of size, and bigger cities have access to a bigger combinatorial fan-out for every individual within them. This is why a clever person in Manhattan ends up doing things they would not have ended up doing in Albany, even if they brought the same brain to both. The brain met more other brains.
The same dynamics show up in places that are not cities at all.
In 2010, the Japanese mathematical biologist Toshiyuki Nakagaki ran one of the most charming experiments of the last fifty years. He took a slime mold — Physarum polycephalum, a single-celled but thousand-nuclei organism that crawls slowly across surfaces in search of food — and he placed oat flakes (slime mold candy) on a damp agar plate at locations that corresponded, scaled down, to the major train stations of Tokyo. He let the slime mold grow.
The slime mold is not a city planner. It has no nervous system, no representation of Tokyo, no understanding that the dots are stations or that connections matter. It does one thing: it grows tubes between food sources, and it pumps protoplasm through the tubes, and tubes that get used a lot get reinforced and tubes that don’t get used much shrink and disappear. That’s it. That’s the rule.
Within twenty-six hours, the slime mold had built a network connecting all the oat flakes. The network resembled the Tokyo subway map. Not in some loose metaphorical way. In a network-theory measurable way: similar total length, similar number of connections, similar redundancy in case of an outage, similar fault tolerance. Tokyo’s subway was designed by humans across a century by a sequence of engineers, and the slime mold matched it overnight by following one rule about tube reinforcement.
Mycelium — the root-like fungal network beneath a healthy forest floor — does the same kind of thing on the scale of a forest. It connects trees, routes nutrients between them, balances supply and demand across square kilometers. The result is a transport network that an engineer who tried to design it from scratch could not improve on without copying the design. Cities, slime molds, and mycelium are running the same trick. They have the same mathematics, the same patterns of redundancy and hub-and-spoke and short-paths-between-busy-nodes. They are different physical instantiations of the same design pattern.
What the universe wants in this case is networks that move stuff efficiently between many sources. There appear to be only a few good ways to do it, and any system — biological, fungal, or human-built — that needs to do it tends to converge on those ways without anybody designing them.
The Experiment
Below is a small city. Each dot is a person. Each person wanders within the bounded area, occasionally bumping into someone else; we’ll call those bumps interactions. Each interaction sparks briefly so you can see them happening. The simulation tracks how many interactions are happening per second, and how many are happening per person.
The point of this experiment is the relationship between population and interactions. The population slider controls how many people are in the city. As you turn the slider up, watch what happens to the per-capita interaction rate. Bigger cities don’t just have more interactions. They have more per person, because each person has more potential partners to bump into. That’s the superlinear bonus, in its purest form.
Things to try:
Start with population = 20. Let it run for ten or fifteen seconds and note the per-person interaction rate. Now move the slider to 60 and wait. Then 120. Then 240. The total interaction rate goes up dramatically; the per-person rate also goes up, more gradually. The per-person rate going up is the whole story. That’s the superlinear bonus.
Watch the scaling chart accumulate points across different populations. The line of points should trend upward to the right. The slope of that trend, on log-log axes, gives you β − 1. In a true linear scaling system β would be 1 and the slope would be flat — per-person rate independent of city size. In a real city, β is roughly 1.15, and the slope is gently positive. In our simulated city the slope tends to be a bit steeper because every person can interact with every other person, where in a real city you’re mostly only interacting with the slice you bump into in your daily orbit.
Try setting the bump radius very small. The interactions per person drop, because you mostly walk past each other without registering. Try setting it large. The interactions per person spike, because everyone is in everyone else’s neighborhood. Bump radius is a stand-in for the social-coupling tightness of the city — how easily ideas and germs and arguments propagate. Tighter coupling, more reactor.
Set the walk speed slow. Interaction rate drops because everyone is moving sluggishly through the same patches of space. Set it fast. Interaction rate climbs because everyone is hitting more of the city per second. Walking speed in real cities really does scale with city size — New Yorkers walk faster than people in Buffalo. They’re running the reactor harder.
Drop the population to 5 and watch how rare interactions become. A village of 5 has almost no encounters. A city of 400 has them constantly. The math is the same; only the density has changed.
The reason cities work, the reason they keep getting bigger, the reason every economy that has ever industrialized has produced megacities even when its government didn’t want them to, is that the social reactor has a built-in productivity dividend that grows with size. Nobody sat down and designed the dividend. It is sitting in the math, in the way that N-squared grows faster than N. The city is the universe’s way of taking advantage of a particular shape of arithmetic, in the same way the slime mold and the mycelium and the lung are. We didn’t invent cities. We built the conditions under which the math made cities inevitable, and the cities organized themselves around the math.
This is also, by the way, why nobody can plan a city. The thing the city is doing is fundamentally a thing that happens at scales no individual planner can hold in their head. The best you can do is design infrastructure that lets the reactor run, and then get out of its way.