World3

How it works

World3 was built at MIT for the Club of Rome by a team led by Dennis Meadows, Donella Meadows and Jørgen Randers, and published in 1972 as The Limits to Growth. This page runs World3-03, the last version its authors released, from their 2004 book Limits to Growth: The 30-Year Update. It follows the world from 1900 to 2100 as one place, with no countries and no prices.

The engine is industrial capital. Factories make output, and part of that output is invested in more factories, so industry grows exponentially. The rest is spent on services, consumer goods, farming and, more and more as the easy deposits run out, on getting resources out of the ground. That last share is what ends growth in the standard run: when it rises, less is left over to invest.

Around this loop sit about a hundred and fifty more equations. Population is split into four age groups whose birth and death rates depend on food, health services, pollution, crowding and how many children families want. Farmland grows when there is capital to develop it and is lost to erosion and to building. Pollution reaches the environment twenty years after it is released and is absorbed more slowly the more of it there is. Many relationships are lookup tables, drawn from data where it existed and from judgement where it did not.

Each change on this page reruns the whole two hundred years in half-year steps, which takes a few hundredths of a second. The playhead only reveals a run that is already finished. The charts keep the book's scales, and widen only when your own mix of policies goes past them.

These are scenarios, not forecasts. The authors never claimed the model could say what will happen in a given year. It shows how a world with these feedbacks tends to behave: whether it overshoots, collapses or settles. Gaya Herrington compared four of the scenarios with fifty years of data in 2021 and found the world tracking closest to Scenario 2 and to a scenario with comprehensive technology.

How close is it to reality?

The hollow dots are what actually happened. Population and life expectancy are measured in the same units as the model, so they sit directly on its charts. The welfare chart shows the UN's Human Development Index, which the model's welfare index was built to imitate. Food, goods and pollution are measured differently in the model and in the world, so tick Index to 1970 to compare their shape instead: each line and each dot is divided by its own 1970 value.

The proxies are rough. Food is calories per person, not the model's grain-equivalent kilograms. Goods are household spending per person. Pollution is carbon dioxide above its pre-industrial level, one of many persistent pollutants. There is no good world series for services back to 1970 or for the stock of resources left, so those charts have no dots. Read the dots for direction, not for decimals, and compare more than one scenario: Scenarios 1 and 2 share their history until about 2010 and part ways after it.

Sources: population and life expectancy, UN World Population Prospects 2024; daily calorie supply, FAO, both via Our World in Data; Human Development Index, UNDP Human Development Report 2025; household final consumption per person in constant 2015 dollars, World Bank; atmospheric CO₂, NOAA Global Monitoring Laboratory, Mauna Loa annual means. Fetched November 2026.

Things to try

1. The standard run. Business as usual. Resources run short, industry peaks around 2015 and falls, and population peaks near 7.6 billion before following it down.Run
2. Twice the resources. Growth lasts twenty years longer, and pollution climbs to three times its standard peak. The collapse still comes, now from poisoned land and shorter lives.Run
3. Technology one fix at a time. Scenarios 3 to 6 add pollution control, higher yields, erosion protection and resource efficiency in turn. Watch each fix move the limit somewhere else: with higher yields but no erosion protection, farmland is nearly gone by 2100.Run 4
4. Fewer children is not enough. Scenario 7 adds two-child families to Scenario 2 and nothing else. Population peaks lower, but pollution still brings the decline.Run
5. A sustainable world. Scenario 9 combines two-child families, a cap on goods per person and all four technologies from 2002. Population levels off near 8 billion and welfare holds to 2100.Run
6. Twenty years earlier, or forty years later. Scenario 10 is the same policy started in 1982: fewer people, less pollution, more resources left. Then drag Policies start in to 2040 and watch the same policies come too late to stop the fall.Run

About this version

The model code is a port of PyWorld3-03 by Tim Schell, which updates Charles Vanwynsberghe's PyWorld3 to the 2004 model, and is shared under the CeCILL 2.1 licence like its source. At yearly steps it reproduces the PyWorld3-03 validation runs to fourteen significant digits, and it stays within a few percent of the authors' original STELLA model. The page runs at half-year steps, as STELLA did.

One switch is arranged differently. PyWorld3-03 ties the land-erosion table to its general policy year, which its standard runs leave switched off; here that table has its own switch, the erosion-protection policy, and the other 1975 settings stay on as in the book. PyWorld3-03 does not implement every part of the 2004 model, so scenarios 2 to 10 approximate the book's runs rather than reproduce them. Scenario 6 in particular stays high to 2100 here. A 2023 recalibration of World3 to recent data (Nebel and others) is not included.

References