(abstract) — One of the effects of climate change is an increase in extreme weather and natural disasters. Unless CO₂ emissions are significantly reduced very soon, it is inevitable that the effects of disaster will exceed many (and ultimately all) societies’ mitigation capacity. Compounding unmitigated disaster effects will slowly but surely push a society towards collapse. Because no part of the planet is safe from the increase in natural disaster intensity and because some of the effects of disasters – such as refugees and economic decline – spill over boundaries, this will eventually lead to global societal collapse. Furthermore, just reducing CO₂ emissions is insufficient to prevent this, as disaster intensity is expected to exceed mitigation capacity in some global regions within one or two decades from now. To avoid a cascade of collapse it is necessary to reverse economic globalization and decrease long-distance trade, as well as to implement a global resettlement program for the increasing number of climate refugees.


Half a year ago I posted an article on this blog discussing certain “doomist” scenarios of the effects of climate change. In that article, On the Fragility of Civilization, I also presented the results of some simulations based on a fairly simple theoretical model on the effects of disasters that I had developed myself.1 Hoping to get some feedback on that model, I wrote a paper explaining the model and its implications, as well as its limitations and how it compares to related models and approaches. Versions of that paper I submitted to two different academic journals (but not simultaneously, of course). Given the inherently multidisciplinary character of the paper, finding appropriate journals was a bit more difficult than expected, but I ended up finding a few that seemed to include the paper’s topic and approach within their scope. (On a side note, although all my publications of the past decade are within philosophy or adjacent fields, I have a degree in geography and have published in that field as well. I always thought of geography as the science of the interaction between man and his physical/spatial environment, and considering that that is exactly what this paper is about, I see this paper as my first geography paper in over a decade. It is quite likely that it will also be my last.)

The first journal I submitted the paper to was a fairly traditional social science journal (albeit with an apparently appropriate scope that explicitly included environmental issues and their social effects). Because the default attitude within the social sciences with regards to climate change appears to be cautious optimism, and because pessimistic scenarios are more or less taboo, I didn’t really expect to get a useful response, but the reply from the editor I got was even more hostile and dismissive than I expected. So I decided to try a very different kind of journal next.

That second journal was less traditional, more focused on problems of sustainability, and much more sympathetic towards model-based approaches, so it seemed a good fit. The editor’s attitude towards my paper was sympathetic as well, and this time it was sent to two reviewers (which is standard procedure). Based on those reviewers’ comments the editor suggested to revise my paper and resubmit. I declined, however.

Both reviewers objected to the qualitative nature of the theoretical model in the paper. This is understandable given that more typical models of societal collapse are mathematical models with clearly defined and quantified variables (see also section 7 below), but as I explain in the paper (see section 5) this just isn’t possible in this context – there is insufficient data and too much uncertainty to estimate all the functions and to measure all the variables. The reviewers recommended to develop the model into a much more detailed mathematical model, but I also explain in the paper (see section 7) that I don’t think that this is particularly useful. Such a mathematical model would give much more exact predictions indeed, but those would be the result of combining advanced mathematical modeling and massive computing power with mere guesses about data and model relations because – again – there just isn’t enough data of the right kind. In other words, the apparent exactness of such models would be deceptive rather than helpful.

In a similar vein, the reviewers suggested to make the model more realistic by including more variables, more geographical variation, and more complex interactions and relations between variables. While I agree that this might improve the model (and actually did this in the simulation model based on the theoretical model), it would even further aggravate the data insufficiency problem. (See also section 5 below.) Furthermore, if you read the paper (see below) it should be fairly obvious that such “improvements” wouldn’t change the model’s implications – and thus, the paper’s conclusions – at all, and that given the stated purpose of the model and paper, there is absolutely no need for such added complexity.

The editor’s suggestion was to develop the theoretical basis of the model and improve its grounding in the literature on disaster mitigation and related topics, so others may develop the model into a more advanced simulation in the future. While I appreciate this suggestion – like I appreciate the comments by the reviewers – this is not something I expect to be able to do in a reasonable amount of time because it would require too much research in areas that I’m insufficiently familiar with.

One of the reviewers called the paper’s topic “of utmost importance” and I certainly agree with that judgment. Following the reviewers’ and editor’s suggestions may lead to a more advanced and (seemingly!) exact model and to a “publishable” paper in several years from now, but that would be utterly pointless for two reasons. Firstly, and as already explained above, the “exactness” achieved would be mere deception and would be very unlikely to change or add any important implications. And secondly, as I also point out in the paper’s conclusion, we don’t really need more exact predictions, we need to act. And most importantly, we need to act now, because if we wait much longer we’re going to see the paper’s predictions in real life.

So, because I don’t expect significantly different outcomes if I submit the paper elsewhere, and because the paper’s topic is indeed “of utmost importance” and of utmost urgency, I’ll just leave it as it as – that is as a “working paper”. It is available as such at my Academia.edu page and at my homepage. It can also be read here – that is, the remainder of this article is identical to the latest version of the paper. (Which differs in minor respects from versions that were submitted.)

Please, note that the theoretical model is presented in more or less mathematical form. If you have an aversion of formulas, I recommend to skip sections 2 and 3 and to go straight to section 4 after the introduction (i.e. section 1). That 4th section gives a plain language summary (without equations) of the preceding sections. (And no equations appear in later sections.)

1. introduction

There is broad consensus within climate science that CO₂-induced global warming will lead to more extreme weather and an increase in the frequency and severity of natural disasters such as storms (including hurricanes and typhoons), droughts, and floods (IPCC 2014; 2018; UNDRR 2019). We are already experiencing this increase, and the situation is expected to get much worse, and to continue getting worse for many decades to come.

In this paper, I will use a very simple model to show that this will inevitably lead to global societal collapse, unless appropriate preventive action is taken. The core of the model is the rather simple and obvious idea that if disasters continuously increase, while mitigation capacity does not (and cannot) keep up with that increase, then it is inevitable that there will be a point at which a society can no longer cope with disaster and starts to decline.

A formal description of the model is presented in section 2, and its main implications are discussed in section 3. It is argued that refugee flows and the economic interdependence resulting from globalized trade make collapse difficult to contain, and that because no part of the planet is safe from increasing natural disaster intensity, societal collapse will eventually become global. Section 4 explains the model and its implications in plain language for readers who are uncomfortable with equations. After that, section 5 looks into the difficulties in fitting the model to the real world in order to make reliable predictions, and section 6 discusses the model’s implications for the prevention of global societal collapse. What is required is a fast reduction of CO₂ emissions, de-globalization, and refugee resettlement. The final section compares the model with some other models and predictions of collapse and recapitulates this paper’s main findings.

2. formal description of the model

The purpose of the theoretical model is to explain the relation between natural disasters and societal collapse. It is assumed that societal collapse corresponds to a (very) high value on a scale measuring civic unrest. Hence, the main endogenous variable is civic unrest u and the main exogenous variable is disaster intensity d. Important endogenous variables in addition to u are the state of the economy e, mitigation capacity c, displacement of people r, and physical and psychological health issues h.

With one exception (namely, expected economic growth g), all variables are relative to population size (of the society/country modeled), and specific to a given year. Variables without an index refer to the given year; the index “−1” refers to the previous year; and the index “n” is a placeholder that stands for some year. The delta symbol Δ means the difference in value between a year and the previous year; thus Δy = yy−1. Most equations include a variable x to represent external effects that are outside the scope of the model, as well as a variable i to represent interaction effects between endogenous variables.

All unspecified functions are assumed to be continuously increasing and close to linear. All inputs are assumed to be positively related to the output. Where necessary, a minus sign is added: if z is a function of y such that z and y are inversely related, then z = fyz(−y). As in this example, all functions are identified by a two-letter index – the first letter identifies the main input variable; the second letter the output variable.

Disaster intensity d, the main exogenous variable, is a measure of the frequency and severity of natural disasters in a given society or area. It can be roughly defined as the number of people in a society or country that were affected by natural disasters in the given year, multiplied by the length of their exposure to the effects of those disasters (or that disaster) and the severity of those effects, divided by population size. It is uncontroversial among climate scientists that climate change will lead to more extreme weather and an increase in the frequency and severity of natural disasters such as storms, droughts, and floods (IPCC 2014; 2018; UNDRR 2019). This increase in the disaster intensity will continue for decades, and possibly even centuries, although it is still possible to significantly reduce the rate and extent of that increase. Nevertheless, at least on decadal timescales, it is expected that the average yearly increase in disaster intensity is greater than zero:

$$\text{[D]}   \langle \Delta d_n \rangle > 0$$

The social effects of natural disasters are grouped into three different kinds: economic effects, displacement of people, and a variety of broadly health-related effects including mortality. The severity of these effects depends on disaster intensity, but also on a society’s mitigation capacity c – that is, its ability to cope with disaster: to provide food and shelter to displaced people, to maintain public order, to help the sick and wounded and prevent epidemics, to repair the damage, and so forth.

Mitigation capacity depends on a number of factors. For example, tightly knit communities may be much more resilient in the face of disaster, and other kinds of social capital also improve resilience (e.g. Aldrich & Meyer 2015). However, on national and larger scales, wealth is the most important determinant of a society’s ability to cope with or mitigate natural disasters. Consequently, a change in c is primarily dependent on economic growth Δe:

$$\text{[C]}   c = c_{-1} + f_{ec} ( \Delta e ) + x_c$$

Economic damage due to national disasters consists of two components: losses due to temporary halts and setbacks in production and distribution, and losses due to the destruction of infrastructure and economic facilities (i.e. the facilities involved in production and distribution of goods and services). Losses of the second kind can be compensated to some extent by rebuilding and repairs, and in favorable circumstances investments can even lead to a growth in economic infrastructure and facilities p. Hence,

$$\text{[P]}   \Delta p = \ – f_{dp} ( p_{-1} , d ) + f_{cp} ( p_{-1} , c ) + x_p $$

in which “−fdp(p−1,d)” determines the loss of economic infrastructure and facilities due to disaster, and “+fcp(p−1,c)” the gains due to reconstruction, recovery, and investment. Losses depend mostly on disaster intensity d, while reconstruction and repair of disaster damage primarily depends on mitigation capacity c.

The aggregate economic effect of disaster, reconstruction, and other relevant factors is modeled as follows:

$$\text{[E]}   e = ( g \times e_{-1} ) \ – \ f_{de} ( d , e_{-1} ) + f_{pe} ( \Delta p , e_{-1} ) \ – \ i_e + t + x_e$$

in which g stands for “expected economic growth”, “−fde(d,e−1)” is economic damage due to temporal halts and setbacks in production, “+fpep,e−1)” is economic losses and gains due to the change in economic infrastructure and facilities Δp (which itself depends on disasters and reconstruction; see [P] above), and t represents the contribution of trade to the state of the economy. (As mentioned above, i stands for interaction effects, which will be discussed below, while x represents external effects that are outside the scope of the model.) Δe, which plays a role in [C] as well some equations below, is e − e−1.

Expected economic growth g is the economic growth that the society modeled (or similar societies) could reasonably be expected to experience if it would not be hit by natural disasters in the given year, or at least not by more disasters than what used to be normal. For the past half century, the global average growth rate has been between 2% and 4% mostly (which would suggest a value for g between 1.02 and 1.04), but it dipped below zero in 2009, and there has been considerable difference between countries. For the model, this does not matter much, but it complicates application to the real world. (See section 5.)

Mitigation capacity c co-determines the extent of economic recovery from disasters, but plays an equally – if not even more – important role in the non-economic direct effects of natural disasters: displacement of people, and physical and psychological health-related problems. Displaced people include evacuees and refugees. (The difference between those two groups is merely one of the extent of assistance offered: evacuees are helped with food and shelter, while refugees are left to tend for themselves.) The share of displaced people in the population r depends on disaster intensity, mitigation, and migration m:

$$\text{[R]}   r = r_{-1} + f_{dr} ( d ) \ – \ f_{cr} ( r_{-1} , c ) + i_r + m + x_r$$

in which “+fdr(d)” determines the increase in the relative number of evacuees and refugees due to disaster, “−fcr(r−1,c)” the number of displaced people who are fully (re-)integrated into society and thereby lose their “displaced” status, and m is the number of immigrating displaced people (from societies/countries other than the one modeled) minus emigrating displaced people. The extent to which a society is able to provide appropriate housing and income to evacuees, refugees, and immigrants – that is, fcr(r−1,c) – is primarily dependent on the society’s mitigation capacity c.

Health-related problems h are a rather broad category in the model, including mortality, injuries, and the effects of shortages of food and water, as well as physical and mental diseases and disorders – such as epidemics, PTSD, anxiety, depression, increased aggression, and so forth – either resulting directly from natural disasters, or arising in a disaster’s aftermath (Watts et al. 2017; Clayton et al. 2017; Evans 2019). Not all effects of disaster are immediate. For example, the main effects of drought only start to realize after the harvest fails and supplies run out; psychological effects can develop slowly and last for many years; and the health effects of famine often last a lifetime.

In the model, h is a measure for the sum of all of these physical and mental health problems in the population, and is determined as follows:

$$\text{[H]}   h = h_{-1} + f_{dh} ( d ) \ – \ f_{ch} ( h_{-1} , c ) + i_h + x_h$$

in which “+fdh(d)” stands for the increase of health-related problems due to disaster, and “−fch(h−1,c)” for a decrease due to mitigation.

The three main equations that model direct disaster effects – [E], [R], and [H] – all include interaction effects. Economic decline can lead to voluntary migration (i.e. displacement), for example, and to stress and other health effects; displacement has physical and mental health effects, and potential economic effects as well; and health-related problems also tend to have economic effects and may also cause displacement (especially in the case of epidemics). Furthermore, in such interaction effects both changes in and levels of variables play a role. For example, both economic decline (Δe<0) and low economic development (a low value on e) affect voluntary migration. And the other way around, both the relative number of refugees r and a change therein Δr affect the economy.

$$\text{[I]}   \begin{matrix} i_e = f_{ie} ( \Delta r , r , \Delta h , h , \Delta u , u ) \\ i_r = f_{ir} ( – \Delta e , -e , \Delta h , h , \Delta u , u ) \\ i_h = f_{ih} ( – \Delta e , -e , \Delta r , r , \Delta u , u ) \end{matrix}$$

What is important to realize about these interaction effects is that they are effectively multipliers of disaster effects. They do not change the nature and/or direction of effects, but merely speed them up and reinforce them. This multiplier effect should not be overestimated, however, except that at very high levels of civic unrest (i.e. in case of civil war or similar social breakdown), the effects thereof are likely to exceed the effects of natural disaster.

Civic unrest u itself is assumed to depend on economic decline and increases in displacement and health-related problems:

$$\text{[U]}   u = u_{-1} \ – \ f_{eu} ( \Delta e ) + f_{ru} ( \Delta r ) + f_{hu} ( \Delta h ) + x_u$$

Civic unrest – in the sense intended here – is the deterioration of faith in, and acceptance of the socio-political status quo. Dmitry Orlov (2013) argued that subsequent stages of collapse are tied to “the breaching of a specific level of trust, or faith, in the status quo” (p. 14). In Gramscian terms, such a deterioration of faith and/or acceptance is a breakdown of (cultural) hegemony. Gramsci (1971) introduced the term “hegemony” to describe the acceptance and/or consent that the socio-political status quo depends on (see also Brons 2017). Without hegemony, only brute force can keep a socio-political system or class in control, but brute force is costly and inefficient. An increase in civic unrest is the undermining of acceptance/consent (and trust or faith, as Orlov suggested, but those are very closely related to acceptance and consent). At low levels, civic unrest may only be expressed at the voting booth; at higher levels riots may occur; and at extreme levels acceptance of the socio-political status quo completely evaporates and a society collapses into chaos or civil war. Well before that highest level is reached certain parts of society may already have collapsed, however. Economic decline can lead to financial collapse independently from civic unrest, for example (but in most circumstances, financial collapse would significantly raise the civic unrest level).

For convenience, the main equations of the model, as well as assumption [D], are reprinted here:

$$\text{[D]}   \langle \Delta d_n \rangle > 0$$
$$\text{[C]}   c = c_{-1} + f_{ec} ( \Delta e ) + x_c$$
$$\text{[P]}   \Delta p = \ – f_{dp} ( p_{-1} , d ) + f_{cp} ( p_{-1} , c ) + x_p $$
$$\text{[E]}   e = ( g \times e_{-1} ) \ – \ f_{de} ( d , e_{-1} ) + f_{pe} ( \Delta p , e_{-1} ) \ – \ i_e + t + x_e$$
$$\text{[R]}   r = r_{-1} + f_{dr} ( d ) \ – \ f_{cr} ( r_{-1} , c ) + i_r + m + x_r$$
$$\text{[H]}   h = h_{-1} + f_{dh} ( d ) \ – \ f_{ch} ( h_{-1} , c ) + i_h + x_h$$
$$\text{[I]}   \begin{matrix} i_e = f_{ie} ( \Delta r , r , \Delta h , h , \Delta u , u ) \\ i_r = f_{ir} ( – \Delta e , -e , \Delta h , h , \Delta u , u ) \\ i_h = f_{ih} ( – \Delta e , -e , \Delta r , r , \Delta u , u ) \end{matrix}$$
$$\text{[U]}   u = u_{-1} \ – \ f_{eu} ( \Delta e ) + f_{ru} ( \Delta r ) + f_{hu} ( \Delta h ) + x_u$$

3. implications

An obvious implication of [E] is that – if trade and external effects are kept constant – actual economic growth Δe depends on the ratio of expected growth g to the economic effects of disaster (−fde(d,e−1)+fpep,e−1)−ie)/e−1. However, while g is either fixed or fluctuating within fairly narrow margins (see also section 5), [D] implies that the economic (and other) effects of disaster are increasing, and will continue to increase. Consequently, from [D] and [E] it follows that any economy will eventually start to decline. That is, if disaster intensity continues to increase – as climate scientists expect to be the case in “business as usual” scenarios (IPCC 2014; 2018) – then inevitably, disaster damage will at some point exceed economic growth, and thus cause decline.

Furthermore, a society’s mitigation capacity c is primarily dependent on the state of its economy, which implies that relative to the state of the economy, c is constant or even declining. Hence, what [D], [E], and [C] together make explicit is the basic fact that if – relative to the state and size of the economy – the frequency and severity of disasters (and therefore, disaster damage) continue to grow while mitigation capacity is more or less stable or even declining, then it is inevitable that at some point disasters will exceed the ability to cope, as illustrated in Figure 1.

Figure 1.

This point may be far in the future for rich countries in global regions that experience relatively few major disasters, but it is much closer by for poorer countries in disaster-prone regions. In 2017, the economy of Puerto Rico (and much more than just the economy) was almost completely destroyed by two hurricanes, for example.

When mitigation capacity can no longer keep up with disasters, the social effects of (ever-increasing) natural disasters start compounding fast. Damaged infrastructure can no longer be repaired, the growing number of displaced people can no longer be meaningfully assisted or (re)integrated into society, health-related problems such as PTSD, anxiety, and epidemics explode, and the economy plummets (see [P], [R], [H], and [E], respectively). And according to [U], all of these effects will lead to an increase in dissatisfaction and civic unrest.

If trade, migration, and external effects that are outside the scope of the model have no significant effects, then the model implies that with a continuous increase in disaster damage, there will be a continuous increase in civic unrest, which will eventually result in societal collapse. Unless there are major oversights in this model, this outcome is as inevitable as it is obvious. Furthermore, even without the assumption of continuously increasing disaster intensity [D], eventual collapse is inevitable if average disaster intensity reaches such a level that the effects of natural disasters exceed the mitigation capacity of a country or society. (This should be obvious from figure 1 above. The dotted line does not need to continue increasing, if it flattens out after crossing the continuous line then the effect will be the same.)

Trade, migration, and external effects may both hasten or slow down collapse, but cannot prevent it in a world where no society is safe from increasing disasters. The most important external effects in a single, isolated society are social and cultural circumstances that lead to a larger mitigation capacity (e.g. Aldrich & Meyer 2015) – that is xc in [C] – and cultural circumstances and socio-political events that decrease unrest – that is xu in [U]. Both could slow down the rise of civic unrest considerably.

If the model is applied to multiple, interacting societies (rather than a single, isolated society) there is another important external effect with regards to mitigation. That is, international aid can substitute for (and/or complement) autonomous mitigation. Even if a society’s own mitigation capacity is low, some of the effects of natural disasters may be relatively small if mitigation is paid for or provided by other societies. This is only possible if those other countries have surplus mitigation capacity, of course. However, because no societies will be spared from the increase in natural disasters and other negative effects of climate change, it is expected that mitigation capacity will decline globally – although not at a universal speed – and therefore, that cross-border assistance will gradually dwindle. (See also sections 5 and 6.)

Much more important than these external effects are migration m in [R] and trade t in [E]. Displaced people tend to migrate when their host society – for whatever reason – becomes too inhospitable. Or to put this another way, growing refugee populations tend to spill over national borders, adding to the populations of displaced people in those adjacent countries. And by migrating, refugees somewhat release the pressure in the countries or societies they leave, but increase the pressure in their new host society. (That is, if they leave society A for B, then they reduce Δr in A and increase Δr in B, which according to [U], if everything else is equal, leads to a decrease in civic unrest u in A and an increase in B.) Especially if numbers of immigrating refugees are much larger than the host society can handle (i.e. well beyond its mitigation capacity), then immigration can lead to a significant rise in civic unrest.

Trade makes economies partially dependent on the economies of their trade partners. If two adjacent societies trade a lot of what they produce with each other (which is the case for many pairs of adjacent societies), then a significant economic decline in one of the two will in almost all cases cause an economic decline in the other. (Except, of course, if A trades with B and C, and although B declines, this decline is compensated by a growth in C. In a world of universal decline, this can only provide temporary relief, however.)

Trade and migration can both spread and mitigate the effects of disaster – they are similar in that respect – but they differ in their spatial scales. Most migration takes place over relatively short distance, while trade is far less spatially restricted. There is a vast global trade network that connects places and societies all over the planet. If such long-distance links are ignored, societies approaching collapse would only affect their neighbors (through declining trade and increasing migration), pushing those closer towards collapse as well, and societal collapse would spread slowly like an oil-stain. But the global trade network ties societies together over much larger distances, and a collapse of that network (due to societal collapse of major players in that network) would probably be a tipping point in the spread of societal collapse, sending a shock-wave of economic decline and deteriorating mitigation capacity all around the globe. What started as a regional security issue then suddenly becomes a global problem with potentially fatal consequences.

4. plain language summary of the model and its implications

One of the most important effects of climate change is an increase in natural disasters and extreme weather such as storms, droughts, and floods. Such natural disasters have (at least) three kinds of social effects: economic damage, displacement of people (that is, refugees and evacuees), and a range of health-related problems ranging from injury or even death to PTSD and anxiety.

If the economic damage of disasters exceeds economic growth, then the economy starts to decline, which affects a society’s mitigation capacity (that is, its ability to cope with the effects of disaster). The lower the mitigation capacity of a society, the less it can do to (re)integrate or otherwise help displaced people, and the less it can do to counter health problems caused directly or indirectly by natural disasters. Then, the population of displaced people (which then mainly consists of refugees) as well as health-related problems start to increase fast. Furthermore, all three direct effects of disasters – economic decline, displaced people, and health-related problems – lead to an increase of dissatisfaction and civic unrest.

Figure 2 is a graphical representation of these relations and effects. Disasters and the three other circles with continuous outlines continue to increase, while the two circles with dotted outlines continue to decrease. All of these changes are ultimately dependent on the predicted increase of the frequency and severity of natural disasters. If those continue to grow, then it is inevitable that a level will be reached at which mitigation capacity approaches zero. And unmitigated effects of disasters lead to civic unrest, which then continues to rise until the society collapses into chaos or civil war.

Figure 2: A graphical representation of the model.

The figure also shows two arrows coming from the outside. The arrow leading to displaced people represents immigration of refugees from adjacent societies (and/or emigration to them). The arrow leading to the economy represents the effects of trade. If the economy of important trade partners declines, then so does the economy of the society modeled. In other words, what these two arrows mean is that if a society is situated in a part of the world that is approaching collapse, then that society will be negatively affected thereby, pulling that society closer to collapse as well. Because ultimately all countries are adjacent and all countries are part of the same global trade network, this implies that societal collapse will gradually spread throughout this network (and thus, the whole world).

This effect only accelerates global societal collapse, however, and does not cause it. What causes it is the mathematical certainty that when disaster damage keeps increasing, it will at some point surpass a society’s ability to cope with and mitigate disaster, because that ability depends on economic growth which cannot continuously increase. In other words, if natural disasters continue to occur more often and continue to cause more damage, then societal collapse will inevitably follow. It may be far away for some countries (and much closer by for others), but no country can outrun continuously increasing disasters forever.

5. the model as a tool for prediction

The model and its implications presented in the preceding sections raise two obvious questions: What is the most likely time-frame for societal collapse? And can it be prevented? The second question will be discussed in section 6. The first – as well as some closely related questions – will be addressed here.

Answering the question about likely time-frames for societal collapse would require a much more sophisticated simulation model than the simple theoretical model presented above. And while some of the extensions and adaptations needed are fairly easy to implement, some others would be very hard and/or controversial. For example, a simulation model would need a demographic module to model births, deaths, and migration for residential and displaced populations as well as transfers between residential and displaced populations. This is relatively easy.

It would also need a distinction between kinds of disasters because the model as presented in section 2 implies that the mix of effects is the same for all disasters, and while that is not a problematic assumption in a theoretical model (as long as economic damage and non-economic disaster effects are continuously increasing with increasing disaster intensity, then the relative sizes of these effects does not matter), it would be a serious flaw in a more realistic simulation. Distinguishing kinds of disasters is not a problem in itself, but it aggravates another obstacles in developing the model into a realistic simulation.

Estimating all the functions in the equations of the model requires a lot of reliable data of the right kind, but much of the data needed is unavailable or insufficiently reliable. It is hard to find detailed and reliable information about disaster damage (especially in less developed countries), for example, and the more categories of disasters a model would distinguish, the more data it would need to estimate the functions determining the effects of all those different kinds of disaster.

Furthermore, there are no single, unambiguous, and uncontroversial ways to measure most of the main variables in the model, and different choices with regards to operationalization and measurement will lead to different outcomes. Health-related problems h and civic unrest u are, perhaps, the most obvious examples of this problem, but measuring the state and size of the economy e is not much less problematic. GDP per capita may seem an obvious choice, but there is a sizable literature arguing against GDP (or GDP/capita) as a good measure of economic health (e.g. Vaury 2007). There may be good reasons to exclude the financial sector, for example. If the economic role of the financial sector is to extract wealth from the “real economy” without contributing to it, as Michael Hudson (2015) and many others have argued, then an increase in GDP due to growth of the financial sector is more likely to decrease mitigation capacity than increase it. This is controversial, of course, but that is exactly the point – almost everything in economics is controversial, even if many (orthodox) economists pretend it is not.

For the same reason there is no uncontroversial way to model expected economic growth g either. Many economists would suggest to fix it at approximately 2%, because that has been more or less the target or standard level for developed countries in the past decades. But this would ignore economic crises and other fluctuations (because those do not exist according to mainstream economic models). Furthermore, economists who emphasize the role of debt, such as Steve Keen (2017) or Michael Hudson (2015), argue that many countries have become (or will soon become) “debt zombies” with growth rates that are close to zero. And according to advocates of “peak oil” or other resource peak theories, economic growth will start to decline soon under the influence of increasing resource scarcity (Heinberg 2007; Hall & Klitgaard 2018). Choosing any of these (or other) alternatives will offend adherents of the competing options.

Perhaps, the hardest problems of all, however, have to do with civil unrest u. As mentioned above, there may not be an unambiguous and uncontroversial measure of u, but even if there would be one, the data necessary for reliable estimates of the economic and other effects of unrest (i.e. the role of u in the functions in [I]) and of the thresholds in civic unrest that lead to riots, civil war, and societal collapse probably does not exist. For the purely theoretical model this is irrelevant because all that matters is that infinitely increasing civic unrest will eventually lead to societal collapse, but for a simulation that is supposed to produce predictions this is a very serious problem.

Technically, creating a simulation model based on the foregoing is not difficult, on the other hand. It can be done in a standard spreadsheet in one or two days. Probably unsurprisingly, I did. I created a simulation with an 8×8 grid of countries, two kinds of randomly occurring disasters (frequent and small, and infrequent and large, with different mixes of effects), and an additional module to simulate demographics including migration. Health effects were reduced to mortality, so the simulation did not take other kinds of health-related problems and their effects into account. To increase realism, the 64 countries in the simulation could start at different economic levels and with different propensities for the two kinds of disasters. The effects of disasters were based on a – undoubtedly non-representative – sample of recent disasters. And to simulate model-external decreases in civic unrest another random effect was added.

Depending on model settings (i.e. extent of disaster damage, growth of disaster intensity, expected economic growth g, and so forth), simulations predicted global societal collapse in between 20 and 40 years from now (with a few simulations with very extreme settings outside that range), with most of them in the 25 to 30 years range. However, because of the problems explained in the preceding paragraphs, these predictions should probably be taken with a very large grain of salt. (And it is also for this reason that I do not give a detailed description of the simulation model and its parameters here.)2

Consequently, as a tool for prediction the model (either in its original abstract form or in the form of the derived simulation) is pretty much useless. Predicting the time until global societal collapse was never the intended purpose of the model presented here, however. Rather, its purpose was and is twofold: firstly, to show that if disaster intensity continues to increase as climate scientists predict, then eventual global societal collapse is a mathematical certainty; and secondly, to gain insight on how to prevent or mitigate that collapse (see next section). Nevertheless, if there are other reasons to believe that global societal collapse is likely within the time frame that the simulation model suggests, that would at least provide some additional support for the model.

Given the inertia of social systems, the lower bound of the 20 to 40 years window of probable collapse does not seem implausible. Except in case of nuclear war, widespread financial collapse, or some other extreme event, rich Western nations are not going to collapse soon – even if the rest of the world is collapsing around them. What seems more questionable than this lower bound is the upper bound, however.

Right now, the average global temperature is approximately 1°C higher than the 1950 baseline (and more than 1.5°C higher than before the industrial revolution). We will reach 1.5°C by 2030 and 2°C soon after 2040 (Xu, Ramanathan, & Victor 2018). These may seem like small numbers, but they have big effects. Many food crops become less productive at higher temperatures. Rice yields drop by about 10% for every 1°C above the ideal average temperature of 25°C, for example, and heat waves above 35°C at the wrong time of year will lead to crop failure for a wide range of essential crops (Hatfield et al. 2011). Much more important than the direct effect of heat is the effect of drought, however. The small difference between 1.5°C of average global warming in 2030 and 2°C soon after 2040 corresponds to a vast difference in exposure to the effects of drought. At 1.5°C less than 10% of the land surface and world population will be affected by aridification (i.e. severe drying), but at 2°C these percentages rise to between 24% and 32% of land surface and between 18% and 24% of world population (Park et al. 2018). That is a two- or even three-fold increase of the number of people suffering the effects of severe drought in one decade. Another recent study suggests that two thirds of the world population will experience an increase of drought with warming, and that the small difference between 1.5°C and 2°C will more than double the global average drought duration (Naumann et al. 2018).

If these predictions are right, then the number of people that is exposed to more or less compromised food and water security will jump from a few 100s of millions in 2030 to well over a billion by approximately 2040. Some of those people will perish in famines or in conflicts over increasingly scarce resources, but a substantial number will try to flee. Numbers of climate refugees are often estimated to reach several 100s of millions by 2050 (e.g. Myers 2002; ESF 2017), but that might be an underestimate in two ways: the real number is likely to be much higher (especially if other effects of climate change and indirect effects such as societal collapse and civil war are taken into account), and the refugee explosion is likely to start (at least) a decade sooner.

Consequently, by 2040 about a quarter of the planet (and possibly even more) will already have difficulty to continue functioning “normally”. And that is just the beginning. Unless preventive action is taken soon, temperatures, natural disasters, and numbers of refugees will continue to rise sharply after that. It seems doubtful that many societies will be able withstand that onslaught for long. Hence, if anything, the 40-year upper bound of collapse “predicted” by the simulation model seems optimistic.

6. preventing collapse

Much of the projected average temperature increase for the coming two decades is already locked in. Earth’s climate system does not respond immediately to changes in CO₂ levels, but lags behind a bit. Hence, the warming we experience now is largely due to the CO₂ we have emitted in past decades. It is this warming that is the main driver of the increase in extreme weather and natural disasters (IPCC 2014; 2018). Warmer oceans produce more frequent and stronger storms, for example. Warmer poles weaken the jet-streams leading to more extreme winter weather in the moderate zones. Changing warming patterns lead to changing wind systems and precipitation patterns, causing droughts and floods. And so forth.

Effects of climate change like these are distributed unevenly over the planet’s surface, however. The tropics and subtropics are hit hardest by drought (Central America and the Middle East are drying out fast, for example), while much of the temperate zones might only have to deal with the relatively “minor” nuisances of erratic weather (such as heatwaves and exceptionally cold winters) and the occasional flood – at least, for now. Hence, the parts of the planet that are likely to suffer the worst effects of climate change in the coming decades are the parts that already have lower mitigation capacities. Given the projections of temperatures, drought, and numbers of refugees mentioned in the last paragraphs of the preceding section, this will almost certainly mean that a substantial part of the planet will be pushed beyond its ability to mitigate and will approach or even cross the threshold of societal collapse before 2040.

Consequently, the question of how to prevent collapse is not a useful question. We are already too late to prevent collapse, and should ask instead whether and how we can prevent it from enveloping the whole planet, or in other words, whether and how we can stop it in its track. The model presented here suggests a twofold answer to that question. Firstly, we must ensure that a sufficiently large part of the planet remains at a disaster level that is below its mitigation capacity, and secondly we must prevent that regional collapse sets off an unstoppable cascade.

The 2°C average global warming projected for the early 2040s (Xu, Ramanathan, & Victor 2018) is more or less unavoidable (barring extreme measures), but we can still prevent continuing warming after that. If we fail to do so, then we will soon approach or even pass several tipping points in the Earth system, many of which are expected to be located somewhere between 1.5°C and 3°C (Lenton et al. 2008; Drijfhout et al. 2015; Steffen et al. 2018). Such tipping points tip large parts (or even the whole) of the Earth’s climate and biosphere from one more or less stable situation into another – possibly unstable – situation. Many tipping points lead to increased warming, and consequently, passing some of them may set off a cascade of catastrophic changes and accelerated global warming. If we do pass such tipping points – or in other words, if we do pass 2°C of average global warming – then disaster intensity will start to increase so rapidly that global societal collapse becomes inevitable.

Consequently, part of what needs to be done to prevent global collapse is to reduce CO₂ emissions to zero now, or as soon as possible. The Intergovernmental Panel on Climate Change recently suggested that we have until 2030 to do this (IPCC 2018), but if warming projections and expectations about tipping points mentioned in the previous paragraph are right, then the window may actually be considerably narrower than that. Reducing CO₂ emissions is only part of the answer, however. They will help keep a larger part of the planet at a disaster level below mitigation capacity, but will do little to prevent regional collapse from spreading beyond control.

There are two main ways in which societies (or countries) in the model interact: through trade (t in [E]) and migration (m in [R]). Dependency on trade makes societies sensitive to economic decline of their trade partners. Globalization has tied the whole world together in an ever denser trade network (especially since the 1990s), and consequently, regional or even local economic problems can have global implications. (The impact of the 2011 earthquake and tsunami in Japan on worldwide car production is a well-known example.) To prevent economic collapse from spreading, economies need to become more resilient and less dependent on long-distance trade. In other words, we need a de-globalization. (But de-globalization is also necessary to reduce CO₂ emissions, as long-distance trade has been an important source of increasing emissions since the 1990s.)

Even more important than economic resilience is refugee management, however. If substantial parts of the tropics and subtropics experience societal collapse (as suggested in the previous section) there will be many 100s of millions of climate refugees seeking food, water, and safety elsewhere. Many will die in famines, droughts, and wars over scarce resources, but there also will be many that make it to other global regions. Unfortunately, those will already have too many problems of their own to be able to handle such an influx. Trying to keep them out is not an option, however. In Six Degrees, Mark Lynas (2007) wrote that “In a situation of serious conflict, invaders do not take kindly to residents denying them food: if a stockpile is discovered, the householder and his family – history suggests – may be tortured and killed, both for revenge and as a lesson to others” (p. 213). Something similar will apply to the national or regional level: walls and armed guards cannot keep out refugees if they number in the millions, and how much trouble they will cause for the host society will largely depend on how that host society treats them. “Invaders” will not take kindly to host societies that deny them food and shelter.

This is, of course, what the model predicts: refugees tend to spill over borders and add to the population of displaced people already present there. The extent to which this leads to an increase in civic unrest depends on mitigation – that is, it depends on whether and how displaced people are (re)integrated into the host society. If a society is unwilling or unable to mitigate the refugee problem (i.e. housing, feeding, and otherwise helping refugees, rather than rejecting them), then the rising share of displaced people in the society’s population will increase its civic unrest level, which – given that the host society will be facing other problems and natural disasters as well – sooner or later will lead to collapse.

Consequently, an unmitigated refugee crisis of the scale expected for the coming decades will lead to societal collapse spreading like an oil-stain until it covers the whole world. And therefore, preventing collapse requires a very different approach to the refugee problem than what is common now. Rather than building walls and fences to try to keep refugees out, the world needs a massive resettlement program for climate refugees.

Unfortunately, it seems that none of the policies that are necessary to prevent global societal collapse will be implemented in the near future. Instead of rapidly reducing CO₂ emissions, the fossil fuel industry keeps drilling and the world keeps burning. Most countries are becoming more rather than less dependent on (long-distance) trade. And instead of resettling displaced people, most of the rich countries that have sufficient mitigation capacity choose to hide behind ever higher walls and barriers (e.g. Parenti 2011; Wainwright & Mann 2018). It is important to realize, however, that we still can avoid the apocalyptic scenario of global societal collapse. If we do not, that is because we made a choice not to avoid it, not because we cannot avoid it. (Or actually, most of us will have no part in that choice, and the consequences of the choice made by some others are just forced upon us. Perhaps, one of the greatest sources of civic unrest in the future will be the realization of this fact, and the resulting – and entirely justified – anger towards those who are most to blame for our situation.)

7. discussion

The model presented in this paper shows that a continuous increase of the frequency and/or severity of natural disasters will gradually deteriorate societies’ ability to cope with those disasters’ effects, leading to rising civic unrest, and ultimately societal collapse. Many other models predicting collapse have been developed in the past half century, but the present model differs from the “typical” collapse model in two important ways.

Firstly, ever since Jay Forrester’s influential World1 (Forrester 1971) and World3 (Meadows et al. 1972) models, the focus has typically been on resource depletion and pollution as causes of collapse. Recent examples include the HANDY model (Motesharrei, Rivas, & Kalnay 2014), and models developed by Bardi (2017) and Nitzbon, Heitzig & Parlitz (2017). Notwithstanding the common focus on resource depletion and pollution, there are important differences between these models, of course. The HANDY model, for example, includes economic stratification (i.e. inequality) as a key explanatory variable of collapse. Nevertheless, none of these models awards an important role to natural disasters or mitigation. (It is possible that the lack of policy influence of these “typical” collapse models is partially due to their focus on resource depletion and pollution. According to mainstream, “neo-classical” economics, which has a virtual monopoly on policy advice, resource depletion does not exist because thanks to technological process and the magic of the market alternative (re)sources will always be found, and pollution is an externality that is usually ignored.)

Secondly, “typical” collapse models are simulation models of complex systems with non-linear dynamics, and consequently, the implications of these models tend to become visible only after running many simulations. Joseph Tainter’s (1988) model even suggests complexity itself as a cause of collapse. The present model is not a simulation model, however, even if its extension into a simple simulation model was discussed in section 5. Moreover, it does not involve complexity in the technical sense – it lacks the kind of non-linear feedback loops that produce the unpredictability that is a hallmark of complexity, for example. Instead, the model merely formalizes the rather simple idea that if mitigation capacity is more or less fixed (or declining) relative to the size of the economy while disaster intensity continues to grow, then it is a mathematical necessity that sooner or later disaster damage will exceed mitigation capacity.

In other words, in the model presented here collapse is a mathematical necessity, rather than the outcome of simulation. And because of that, the model may have more in common with, for example, Malthus’s theory on the divergence between population growth and food production or Marx’s theory on the tendency of the rate of profit to fall, than with the “typical” collapse models mentioned above. There is, however, a fundamental difference between those two theories on the mathematical necessity of some kind(s) of catastrophic events and the model presented here. This model only shows that societal collapse becomes inevitable when aggregate disaster damage exceeds a society’s mitigation capacity. This will eventually happen everywhere if climate change remains unchecked, but the current kind of climate change is a historical contingency and could have been avoided. (In fact, it almost was avoided. See Rich 2019.) And by turning the tides, the most catastrophic outcome can (probably) still be avoided (see previous section). Hence, while Malthus and Marx’s (presumed) “catastrophes” are much like natural facts, according to the model presented here, eventual catastrophe will be our own doing – there is nothing natural about it.

The model presented here should also be distinguished from scenarios of global societal collapse based on predictions of climate change that have lower probability but more serious consequences (often called “fat tail” events). In contrast to such scenarios (see Spratt & Dunlop 2019 for a recent, representative example), the model presented here does not assume any kind of low-probability events or low-probability climate change effects. Rather, it takes at its starting point the uncontroversial notion that climate change is already leading to, and will continue to lead to an increase in the frequency and severity of natural disasters.

Furthermore, while such scenarios tend to involve dates and time-lines, as explained above (see section 5), the model presented here was not intended to give exact predictions about the time-frame of collapse, nor about where it will hit first and how it will spread. Rather, its purpose was to make explicit what should be obvious – namely, that if disaster intensity continues to increase as climate scientists predict, then eventual global societal collapse is a mathematical certainty – and to gain at least some insight on whether and how global societal collapse can (still) be prevented (see section 6). Nevertheless, the model probably can be adapted and extended into a much more detailed and realistic simulation model that might be able to produce more accurate predictions about possible futures. This raises the question of whether it is actually worth doing that. It is not immediately obvious that we need detailed predictions of how and when civilization as we know it is going to end. What we need is to prevent that catastrophe, and that requires action rather than further research.


Aldrich, D., Michelle, M. 2015. Social Capital and Community Resilience. American Behavioral Scientist, 59 (2), 254-269.

Bardi, U. 2017. The Seneca Effect: Why Growth is Slow but Collapse is Rapid. Berlin: Springer.

Brons, L. 2017. The Hegemony of Psychopathy. Santa Barbara: Brainstorm Books.

Clayton, S., Manning, C., Krygsman, K., Speiser, M. 2017. Mental Health and Our Changing Climate: Impact, Implications, and Guidance. Washington: American Psychological Association and ecoAmerica.

Drijfhout, S., Bathiany, S., Beaulieu, C., Brovkin, V., Claussen, M., Huntingford, C., Scheffer, M., Sgubin, G., Swingedouw, D. 2015. Catalogue of Abrupt Shifts in Intergovernmental Panel on Climate Change Climate Models. Proceedings of the National Academy of Sciences, E5777-E5786.

ESF. 2017. Beyond Borders: Our Changing Climate – Its Role in Conflict and Displacement. London: Environmental Justice Foundation.

Evans, G. 2019. Projected Behavioral Impacts of Global Climate Change. Annual Review of Psychology, 70, 6.1-6.26.

Forrester, J. 1971. World Dynamics. Cambridge: Wright-Allen.

Gramsci, A. 1971. Selections from the Prison Notebooks. New York: International Publishers.

Hall, C., Klitgaard, K. 2018. Energy and the Wealth of Nations: An Introduction to Biophysical Economics. Second Edition. Berlin: Springer.

Hatfield, J., Boote, K., Kimball, B., Ziska, L., Izaurralde, R., Ort, D., Thomson, A., Wolfe, D. 2011. Climate Impacts on Agriculture: Implications for Crop Production. Agronomy Journal, 103 (2), 351-370.

Heinberg, R. 2017. Peak Everything: Waking Up to the Century of Declines. Gabriola Island: New Society.

Hudson, M. 2015. Killing the Host: How Financial Parasites and Debt Bondage Destroy the Global Economy. Petrolia: Counterpunch Books.
IPCC. 2014. AR5 – Climate Change 2014: Impacts, Adaptation, and Vulnerability. Intergovernmental Panel on Climate Change.

IPCC. 2018. SR15 – Global Warming of 1.5°C. Intergovernmental Panel on Climate Change.

Keen, S. 2017. Can We Avoid Another Financial Crisis? Cambridge: Polity.

Lenton, T., Held, H., Kriegler, E., Hall, J., Lucht, W., Rahmstorf, S., Schellnhuber, H.J. 2008. Tipping Elements in the Earth’s Climate System. Proceedings of the National Academy of Sciences, 105 (6), 1786-1793.

Lynas, M. 2007. Six Degrees: Our Future on a Hotter Planet. New York: HarperCollins.

Meadows, D., Meadows, D.L., Randers, J., Behrens, W. 1972. The Limits to Growth; A Report for the Club of Rome’s Project on the Predicament of Mankind. New York: Universe Books.

Motesharrei, S., Rivas, J., Kalnay, E. 2014. Human and Nature Dynamics (HANDY): Modeling Inequality and Use of Resources in the Collapse or Sustainability of Societies. Ecological Economics, 101, 90-102.

Myers, N. 2002. Environmental Refugees: A Growing Phenomenon of the 21st Century. Philosophical Transactions of the Royal Society of London B, 357, 609-613.

Naumann, G., Alfieri, L., Wyser, K., Mentaschi, L., Betts, R.A., Carrao, H., Spinoni, J., Vogt, J., Feyen, L. 2018. Global Changes in Drought Conditions under Different Levels of Warming, Geophysical Research Letters, 45, 3285-3296.

Nitzbon, J., Heitzig, J., Parlitz, U. 2017. Sustainability, Collapse and Oscillations in a Simple World-Earth Model. Environmental Research Letters, 12, 074020.

Orlov, D. 2013. The Five Stages of Collapse: A Survivor’s Toolkit. Gabriola Island: New Society.

Parenti, C. 2011. Tropic of Chaos: Climate Change and the New Geography of Violence. New York: Nation Books.

Park, C.-E., Jeong, S.-J., Joshi, M., Osborn, T., Ho, C.-H., Piao, S., Chen, D., Liu, J., Yang, H., Park, H., Kim, B.-M., Feng, S. 2018. Keeping Global Warming within 1.5°C Constrains Emergence of Aridification. Nature Climate Change, 8, 70–74.

Rich, N. 2019. Losing Earth: The Decade We Could Have Stopped Climate Change. London: Picador.
Spratt, D., Dunlop, I. 2019. Existential Climate-Related Security Risk: a Scenario Approach. Melbourne: Breakthrough.

Steffen, W., Rockström, J., Richardson, K., Lenton, T., Folke, C., Liverman, D., Summerhayes, C., Barnosky, A., Cornell, S., Crucifix, M., Donges, J., Fetzer, I., Lade, S., Scheffer, M., Winkelmann, R., Schellnhuber, H.J. 2018. Trajectories of the Earth System in the Anthropocene. Proceedings of the National Academy of Sciences, 115 (33), 8252-8259.

Tainter, J. 1988. The collapse of complex societies. Cambridge: Cambridge University Press.

UNDRR. 2019. Global Assessment Report on Disaster Risk Reduction. Geneva: United Nations Office for Disaster Risk Reduction.

Vaury, O. 2007. Is GDP a Good Measure of Economic Progress?. In: E. Fullbrook (ed.), Real World Economics: A Post-Autistic Economics Reader (London: Anthem), pp. 215-219.

Wainwright, J., Mann, G. 2018. Climate Leviathan: A Political Theory of Our Planetary Future. London: Verso.

Watts N., Amann, M., Ayeb-Karlsson, S., Belesova, K., Bouley, T., Boykoff, M., Byass, P., Cai, W., Campbell-Lendrum, D., Chambers, J., Cox, P., Daly, M., Dasandi, N., Davies, M., Depledge, M., Depoux, A., Dominguez-Salas, P., Drummond, P., Ekins, P., Flahault, A., Frumkin, H., Georgeson, L., Ghanei, M., Grace, D., Graham, H., Grojsman, R., Haines, A., Hamilton, I., Hartinger, S., Johnson, A., Kelman, I., Kiesewetter, G., Kniveton, D., Liang, L., Lott, M., Lowe, R., Mace, G., Sewe, M.O., Maslin, M., Mikhaylov, S., Milner, J., Latifi, A.M., Moradi-Lakeh, M., Morrissey, K., Murray, K., Neville, T., Nilsson, M., Oreszczyn, T., Owfi, F., Pencheon, D., Pye, S., Rabbaniha, M., Robinson, E., Rocklöv, J., Schütte, S., Shumake-Guillemot, J., Steinbach, R., Tabatabaei, M., Wheeler, N., Wilkinson, P., Gong, P., Montgomery, H., Costello, A. 2017. The Lancet Countdown on Health and Climate Change: from 25 Years of Inaction to a Global Transformation for Public Health. The Lancet, 391 (10120), 581-630.

Xu W.Y., Ramanathan V., Victor D. 2018. Global Warming Will Happen Faster than we Think. Nature, 564, 30-32.

If you found this article and/or other articles in this blog useful or valuable, please consider making a small financial contribution to support this blog 𝐹=𝑚𝑎 and its author. You can find 𝐹=𝑚𝑎’s Patreon page here.


  1. See also Michael Mann versus the “Doomists”.
  2. For a bit more information about some of these simulation runs, see On the Fragility of Civilization.