Coevolutionary dynamics via adaptive feedback in collectiverisk social dilemma game
Abstract
Human society and natural environment form a complex giant ecosystem, where human activities not only lead to the change in environmental states, but also react to them. By using collectiverisk social dilemma game, some studies have already revealed that individual contributions and the risk of future losses are inextricably linked. These works, however, often use an idealistic assumption that the risk is constant and not affected by individual behaviors. Here, we develop a coevolutionary game approach that captures the coupled dynamics of cooperation and risk. In particular, the level of contributions in a population affects the state of risk, while the risk in turn influences individuals’ behavioral decisionmaking. Importantly, we explore two representative feedback forms describing the possible effect of strategy on risk, namely, linear and exponential feedbacks. We find that cooperation can be maintained in the population by keeping at a certain fraction or forming an evolutionary oscillation with risk, independently of the feedback type. However, such evolutionary outcome depends on the initial state. Taken together, a twoway coupling between collective actions and risk is essential to avoid the tragedy of the commons. More importantly, a critical starting portion of cooperators and risk level is what we really need for guiding the evolution toward a desired direction.
Editor's evaluation
The paper provides a valuable, indepth mathematical analysis of the coevolutionary dynamics resulting from a coupling of players' strategies and (collective) risk, as well as illustrative numerical simulations of the system's trajectories for different starting conditions. It is therefore a solid contribution to our understanding of how cooperation can be sustained when there is feedback between individual decisions and the global risk of disaster. This paper will be of interest to scientists working on mathematical biology/ecology, and more generally various aspects of human decisionmaking, the interplay between human decisions and the environment, and public goods provision.
https://doi.org/10.7554/eLife.82954.sa0Introduction
Human activities constantly affect the natural environment and cause changes in its quality, which in turn affects our daily life and health conditions (Patz et al., 2005; Steffen et al., 2006; Perc et al., 2017; Obradovich et al., 2018; Hilbe et al., 2018; Su et al., 2019; Su et al., 2022). A wellknown example is climate change, which is one of the biggest contemporary challenges of our civilization (Parmesan and Yohe, 2003; Stone et al., 2013). A large number of carbon emissions caused by human activities will exacerbate the greenhouse effect, which risks raising global temperatures to dangerous levels. The direct consequences of global warming are the melting of glaciers and the rise in sea level, which will inevitably affect human activities (Schuur et al., 2015; Obradovich and Rahwan, 2019; Moore et al., 2022). Similarly, we can give more examples of coupled human and natural systems to continue this list, such as habitat destruction and the spread of infectious diseases (Liu et al., 2001; Liu et al., 2007; Chen and Fu, 2019; Tanimoto, 2021; Chen and Fu, 2022). At present, the importance of developing a new comprehensive framework to study the coupling between human behavior and the environment has been recognized by a number of interdisciplinary approaches (Weitz et al., 2016; Chen and Szolnoki, 2018; Tilman et al., 2020).
Evolutionary game theory provides a powerful theoretical framework for studying the coupled dynamics of human and natural systems (Maynard Smith, 1982; Weibull, 1997; Stewart and Plotkin, 2014; Radzvilavicius et al., 2019; Park et al., 2020; Niehus et al., 2021; Han et al., 2021; Cooper et al., 2021). Furthermore, coevolutionary game models have recognized the fact that individual payoff values are closely related to the state of the environment (Weitz et al., 2016; Szolnoki and Chen, 2018; Chen and Szolnoki, 2018; Hauert et al., 2019; Tilman et al., 2020; Wang and Fu, 2020; Yan et al., 2021). For example, Weitz et al., 2016 considered the dynamical changes of the environment, which modulates the payoffs of individuals. Their results show that individual strategies and the environmental state may form a sustained cycle where strategy swing between full cooperation and full defection, while the environment state oscillates between the replete state and the depleted state. Along this line, feedbackevolving game systems with intrinsic resource dynamics (Tilman et al., 2020), asymmetric interactions in heterogeneous environments (Hauert et al., 2019), and timedelay effect (Yan et al., 2021) have been also investigated where periodic oscillation of strategy and environment is observed. As a general conclusion, the feedback loop between individual strategies and related environment is a key element to maintain longterm cooperation and sustainable use of resources.
Despite the mentioned efforts, the research on possible consequences of the feedback between human activity and natural systems is still in early stage. Staying at the abovementioned example, potential feedback loops between human activities and climate change exist (Obradovich and Rahwan, 2019). However, most scholars study these two topics, that is, human contributions to climate change and social impacts of the changing climate on human behavior, in a separated way (Vitousek et al., 1997; Barfuss et al., 2020). On the one hand, some of them usually focus on how human behaviors (use of land, oceans, fossil fuel, freshwater, etc.) affect environment (Vitousek et al., 1997). On the other hand, researchers who are interested in society and biology frequently focus on how environmental change will affect human behaviors (Culler et al., 2015; Obradovich and Rahwan, 2019; Celik, 2020). Recently, these two approaches have been merged into a single framework, called collectiverisk social dilemma game, which serves as a general paradigm for studying climate change dilemmas (Milinski et al., 2008). Within it, a group of individuals decide whether to contribute to reach a collective goal. If the total contributions of all individuals exceed a certain threshold, then the disaster is averted and all individuals benefit from it. Otherwise, the disaster occurs with a probability (also known as the risk of collective failure), resulting in fatal economic losses for all participants. Both behavioral experiments and theoretical works show that the risk of future losses plays an important role in the evolution of cooperation (Milinski et al., 2008; Santos and Pacheco, 2011; Chen et al., 2012a; Vasconcelos et al., 2013; Hilbe et al., 2013; Vasconcelos et al., 2014; Barfuss et al., 2020; Domingos et al., 2020; Sun et al., 2021; Chica et al., 2022).
Previous studies based on the collectiverisk dilemma game revealed that the risk of collective failure could affect individuals’ motivation to cooperate when they face the problem of collective action, but ignored an important practical aspect. That is, human decisionmaking is not only affected by changes in the risk state, but also affects the level of risk (Chen et al., 2012a). Indeed, the risk of collective failure is lower in a highly cooperative society, but becomes significant in the opposite case. This fact is not only reflected in climate change (Moore et al., 2022), but also in the spread of infectious diseases (Chen and Fu, 2022) and vaccination (Nichol et al., 1998; Chen and Fu, 2019). Furthermore, although the risk level varies in a changing population, their relation is not necessarily straightforward. For example, a study revealed that the infectionfatality risk (IFR) of COVID19 in India decreased linearly from June 2020 to September 2020 due to improved healthcare or increased vaccination (Yang and Shaman, 2022). Throughout the whole process (from March 2020 to April 2021), the statistical curve of IFR is nonlinear, that is, when the epidemic broke out, the value of IFR remained at a high level, and then with the increase in vaccination or the improvement in healthcare, the IFR value gradually decreased, then flattened and remained at a low level (Yang and Shaman, 2022). On the other hand, the change in risk is bound to affect individuals’ decisionmaking, which has been confirmed in behavioral experiments and theoretical research (Milinski et al., 2008; Pacheco et al., 2014). Though potential feedback loops between strategy and risk of future losses are already recognized, a study focusing on their direct interaction is missing. Furthermore, it is still an open question whether the character of feedback mechanism plays an essential role in the final evolutionary outcome. Hence, how the impacts of risk on human systems might, in turn, alter the future trajectories of human decisionmaking remains largely unexplored.
To fill this gap, we propose a coupled coevolutionary game framework based on the collectiverisk dilemma to describe reciprocal interactions and feedbacks between decisionmaking procedure of individuals and risk. In particular, we assume that the increasing freeriding behaviors will slowly increase the risk of collective failure, and the resulting highrisk level will in turn stimulate individual contributions. However, the increase in contribution will gradually reduce the risk of collective failure, and the resulting lowrisk level will promote the prevalence of freeriding behaviors again. This general feedback loop is illustrated in Figure 1. Importantly, we respectively consider two conceptually different feedback protocols describing the effect of strategy on risk. Namely, both linear and highly nonlinear (exponential) feedback forms are checked. Our analysis identifies the conditions for the existence of stable interior equilibrium and stable limit cycle dynamics in both cases.
Materials and methods
Collectiverisk social dilemma game
Request a detailed protocolWe consider an infinite wellmixed population in which $N$ individuals are selected randomly to form a group for playing the collectiverisk social dilemma game. Each individual in the group has an initial endowment $b$ and can choose one of the two strategies, that is, cooperation and defection. Cooperators will contribute an amount $c$ to the common pool, whereas defectors contribute nothing. The remaining endowments of all individuals can be preserved if the overall number of cooperators exceeds a threshold value $M$, where $1<M<N$ (Milinski et al., 2008; Santos and Pacheco, 2011). Otherwise, individuals will lose all their endowments with a probability $r$, which characterizes the risk level of collective failure. Accordingly, the payoffs of cooperators and defectors in a group of size $N$ with ${j}_{C}$ cooperators and $N{j}_{C}$ defectors can be summarized as
where $\theta (x)$ is the Heaviside function, that is, $\theta (x)=0$ if $x<0$, being one otherwise. Here, we would like to note that the collectiverisk social dilemma is a kind of public goods games, which are a special and extended version of Donor & Recipient game by referring to the concept of universal dilemma strength (Wang et al., 2015; Ito and Tanimoto, 2018; Tanimoto, 2021). However, following previous work on collectiverisk social dilemma (Santos and Pacheco, 2011), we retain the parameters mentioned above for the sake of convenience, instead of replacing them with the universal dilemma strength.
To analyze the evolutionary dynamics of strategies in an infinite population, we use replicator equations to describe the time evolution of cooperation (Taylor and Jonker, 1978; Schuster and Sigmund, 1983). Accordingly, we have
where $x$ denotes the frequency of cooperators in the population, while ${f}_{C}$ and ${f}_{D}$ respectively denote the average payoffs of cooperators and defectors, which can be calculated as
where ${P}_{C}$ and ${P}_{D}$ are shown in Equations 1 and 2. After some calculations, the difference between the average payoffs of cooperators and defectors can be written as
In the above replicator equation, we describe a gametheoretic interaction involving the risk of collective failure, which is a positive constant in previous works (Santos and Pacheco, 2011; Chen et al., 2012a). Here, we are focusing on a dynamical system where there is feedback between strategic behaviors and risk. In particular, the impact of strategies on the risk level is channeled through a function $U(x,r)$, which depends on both key variables. Then by using the general form of the feedback, the coevolutionary dynamics can be written as
where $\epsilon $ denotes the relative speed of strategy update dynamics (Weitz et al., 2016), such that when $0<\epsilon \ll 1$ the strategies evolve significantly faster than the change in the risk level. In the following, we consider both linear and nonlinear forms of feedback describing the effect of strategy distribution on the evolution of risk.
Linear effect of strategy on risk
Request a detailed protocolIn the first case, we assume that the effect of strategies on the risk level takes a linear form, which is the most common form that can be used to describe the characteristic attributes between key variables. Just to illustrate it by a specific example, the probability of influenza infection among individuals who have not been vaccinated decreases linearly with the increase in vaccine coverage (Vardavas et al., 2007). Here, we consider that the value of risk decreases linearly with the increase in cooperation level. Furthermore, by following the work of Weitz et al., 2016, we can write the dynamical equation of risk as
where $u(1x)x$ denotes the increase in risk by the defection level at rate $u$ and the decrease by the fraction of cooperators at relative rate one. Then the dynamical system is described by the following equation:
Exponential effect of strategy on risk
Request a detailed protocolTo complete our study, we also apply a nonlinear form of feedback function. The most plausible choice is when the risk level depends exponentially on the population state. To be more specific, we consider that the risk will decrease when the frequency of cooperators in the population exceeds a certain threshold value $T$. Otherwise, the risk level will increase. Such a scenario is suitable for describing climate change and the spread of infectious diseases, in which the risk can increase sharply, such as the occurrence of extreme weather (Eckstein et al., 2021) or a sudden outbreak of an epidemic in a region (Yang and Shaman, 2022). Here, we use the sigmoid function to describe the effect of strategy on the risk state (Boza and Számadó, 2010; Chen et al., 2012b; Couto et al., 2020), which can be written as
where $\beta $ characterizes the steepness of the function and $r(1r)$ ensures that the risk state remains in the $[0,1]$ domain. For convenience, we introduce the variable $\xi =xT$ and the function $B(\xi )=\frac{1}{1+{e}^{\beta \xi}}\frac{1}{1+{e}^{\beta \xi}}$. Thus we have $\dot{r}=r(1r)B(\xi )$. When $\beta =0$, we know that $B(\xi )=0$. In this situation, strategies have no effect on the risk level. For $\beta =+\mathrm{\infty}$, the function $B(\xi )$ becomes steplike so that the risk will decrease only if the frequency of cooperators in the group exceeds the threshold $T$. Otherwise, the risk level remains high. To study the consequence of a proper feedback effect, we apply a finite $\beta >0$ value. In Figure 2, we illustrate how $B(\xi )$ varies with $\xi $ for four different values of $\beta $.
Accordingly, the feedbackevolving dynamical system where the effect of strategies on the risk state is expressed by the exponential form can be written as
Here, in order to help readers to overview easily all the parameters and variables introduced in our work, we present them in Table 1. In the following section, we respectively investigate the coevolutionary dynamics of strategy and risk when considering linear and exponential feedback forms. We note that the details of theoretical analysis can be found in Appendix 1.
Results
System I: Coevolutionary dynamics with linear feedback
We first consider the case of linear feedback. More precisely, we assume that the risk value of collective failure will decrease linearly with the increase in cooperation and increase linearly with the increase in defection level. The resulting dynamical system is presented in Equation 5. After some calculations, we find that this equation system has at most seven fixed points, which are $(0,0)$, $(0,1)$, $(1,0)$, $(1,1)$, $(\frac{u}{1+u},{r}^{*})$, $({x}_{1}^{*},1)$, and $({x}_{2}^{*},1)$, where ${r}^{*}=\frac{c}{\left(\genfrac{}{}{0pt}{}{N1}{M1}\right){(\frac{u}{1+u})}^{M1}{(\frac{1}{1+u})}^{NM}b}$, ${x}_{1}^{*}$ and ${x}_{2}^{*}$ are the real roots of the equation $\left(\genfrac{}{}{0pt}{}{N1}{M1}\right){x}^{M1}{(1x)}^{NM}b=c$. We further perform theoretical analysis for these equilibrium points, as provided in Appendix 1. In order to describe the stable states of System I for the complete parameter regions, we present a schematic plot in the parameter space $(u,\frac{c}{b})$, as shown in Figure 3. We use different colors to distinguish the evolutionary outcomes for specific pairs of key parameters. In the following, we discuss the representative results in detail.
System I has an interior equilibrium point
When $(\genfrac{}{}{0pt}{}{N1}{M1})(\frac{u}{1+u}{)}^{M1}(\frac{1}{1+u}{)}^{NM}b>c$, we know that our coevolutionary system has an interior fixed point. According to its stability, we can distinguish three subcases here. Namely, when $u>\frac{M1}{NM}$, then the existing interior fixed point is stable. Besides, since $(\genfrac{}{}{0pt}{}{N1}{M1})(\frac{M1}{N1}{)}^{M1}(1\frac{M1}{N1}{)}^{NM}b>c$, there exist seven fixed points in the system, namely, $(0,0),(0,1),(1,0),(1,1),(\frac{u}{1+u},{r}^{*}),({x}_{1}^{*},1)$, and $({x}_{2}^{*},1)$. Here, only $(0,1)$ and $(\frac{u}{1+u},{r}^{*})$ are stable (marked by the yellow area in Figure 3). Besides, we provide numerical examples to illustrate the above theoretical analysis (see the top row of Figure 4). We find that bistable dynamics can appear, that is, depending on the initial conditions the system will evolve to one of two stable equilibria: here, $(0,1)$, which is the undesirable full defection equilibrium, or the interior fixed point suggests that cooperation can be maintained at a high level when the value of risk exceeds an intermediate value. Furthermore, we note that the results are not affected qualitatively by the feedback speed in any of the cases (see Appendix 1—figure 1 in Appendix 1).
If the enhancement rate of risk caused by defection drops to a certain threshold, namely, $u=\frac{M1}{NM}$, a Hopf bifurcation takes place, which is supercritical (marked by the blue line in Figure 3). In this situation, System I has all seven fixed points. As analyzed in Appendix 1, only $(0,1)$ is stable. Furthermore, we provide numerical examples to illustrate our theoretical analysis (see the second row of Figure 4). We find that the system is bistable: depending on the initial fractions of cooperators and risk, the system can evolve either to a highrisk state without cooperation or to a limit cycle where the frequencies of cooperation and risk show periodic oscillations.
When the enhancement rate of risk caused by defection is weak and meets $u<\frac{M1}{NM}$ condition, then the interior fixed point is unstable. Besides, since $(\genfrac{}{}{0pt}{}{N1}{M1})(\frac{M1}{N1}{)}^{M1}(1\frac{M1}{N1}{)}^{NM}b>c$, there exist all seven fixed points. According to the theoretical analysis presented in Appendix 1, only $(0,1)$ fixed point is stable. In the third row of Figure 4, we present some representative numerical examples. They show that all trajectories in the state space terminate at the fixed point $(0,1)$, which is consistent with our theoretical results. This means that no individual chooses to contribute to the common pool, leading to the failure of collective action, and finally, all individuals inevitably lose all their endowments.
System I has no interior equilibrium point
The alternative case is when there is no interior fixed point, namely, $(\genfrac{}{}{0pt}{}{N1}{M1})(\frac{u}{1+u}{)}^{M1}(\frac{1}{1+u}{)}^{NM}b\le c$. In this situation, when $(\genfrac{}{}{0pt}{}{N1}{M1})(\frac{M1}{N1}{)}^{M1}(1\frac{M1}{N1}{)}^{NM}b>c$, System I has six fixed points, which are $(0,0),(0,1),(1,0),(1,1),({x}_{1}^{*},1),$ and $({x}_{2}^{*},1)$, respectively. The theoretical analysis, presented in Appendix 1 , shows that $(0,0),(1,0),(1,1),$ and $({x}_{1}^{*},1)$ are unstable, $(0,1)$ is stable, and $({x}_{2}^{*},1)$ is stable for $x}_{2}^{\ast}<\frac{u}{1+u$ (shown by the green area in Figure 3). In the bottom row of Figure 4, we provide some numerical examples to illustrate our theoretical results. The phase plane dynamics show that most trajectories in phase space converge to the stable equilibrium point $({x}_{2}^{*},1)$, which suggests that driven by the high risk of future loss, most individuals will contribute to the common pool. Besides, the remaining trajectories in the phase space will converge to the fixed point $(0,1)$, which means a complete failure when all individuals lose all remaining endowments.
Furthermore, we prove that the fixed point $({x}_{2}^{*},1)$ is unstable when $x}_{2}^{\ast}>\frac{u}{1+u$ in Appendix 1. For the special case of ${x}_{2}^{*}=\frac{u}{1+u}$, we find that one eigenvalue of the Jacobian matrix at $({x}_{2}^{*},1)$ is zero and the other one is negative. We provide the stability analysis of this fixed point by using the center manifold theorem (Khalil, 1996). When $(\genfrac{}{}{0pt}{}{N1}{M1})(\frac{M1}{N1}{)}^{M1}(1\frac{M1}{N1}{)}^{NM}b\le c$, $(0,1)$ is the only stable equilibrium point of the System I.
System II: Coevolutionary dynamics with exponential feedback
In this section, we consider the case of exponential feedback. Here, there are at most seven equilibrium points of the replicator Equation 7. Namely, $(x,y)=(0,0)$, $(0,1)$, $(1,0)$, $(1,1)$, $({x}_{1}^{*},1)$, $({x}_{2}^{*},1)$, and $(T,\frac{c}{\left(\genfrac{}{}{0pt}{}{N1}{M1}\right){T}^{M1}{(1T)}^{NM}b})$, in which ${x}_{1}^{*}$ and ${x}_{2}^{*}$ satisfy the equation $\left(\genfrac{}{}{0pt}{}{N1}{M1}\right){x}^{M1}{(1x)}^{NM}b=c$ and $x}_{1}^{\ast}<\frac{M1}{N1}<{x}_{2}^{\ast$ (Santos and Pacheco, 2011). For convenience, we set $\overline{r}=\frac{c}{\left(\genfrac{}{}{0pt}{}{N1}{M1}\right){T}^{M1}{(1T)}^{NM}b}$. Here, the first six equilibria are boundary fixed points, and the last one is an interior fixed point. In Appendix 2, we analyze the stability of these equilibria under four different parameter ranges by evaluating the sign of the eigenvalues of the Jacobian (Khalil, 1996). The basins of each solution in parameter space $(T,\frac{c}{b})$ are shown in Figure 5. In the following, we will discuss the evolutionary outcomes depending on whether System II has an interior equilibrium point.
System II has an interior equilibrium point
In this case $c<(\genfrac{}{}{0pt}{}{N1}{M1}){T}^{M1}(1T{)}^{NM}b$, there are three typical dynamical behaviors for the evolution of cooperation and risk according to the stability conditions of the interior equilibrium point (for details, see Appendix 2).
When $T>\frac{M1}{N1}$, the interior fixed point is stable. Besides, since $(\genfrac{}{}{0pt}{}{N1}{M1})(\frac{M1}{N1}{)}^{M1}(1\frac{M1}{N1}{)}^{NM}bc>0$, there exist two boundary fixed points, which are $({x}_{1}^{*},1)$ and $({x}_{2}^{*},1)$. Thus the system has seven fixed points, which are $(0,0),(0,1),(1,0),(1,1),({x}_{1}^{*},1),({x}_{2}^{*},1)$, and $(T,\overline{r})$. From the Jacobian matrices, we can conclude that the fixed points $(0,0),(0,1),(1,0),(1,1),({x}_{1}^{*},1),$ and $({x}_{2}^{*},1)$ are unstable, while $(0,1)$ and $(T,\overline{r})$ are stable. The latter case is shown in the top row of Figure 6, where we plot the phase plane and temporal dynamics of the system. It suggests that there is a stable interior fixed point, and most trajectories in phase space converge to this nontrivial solution, which means that the system can evolve into a state where the risk is kept at a low level and almost half of the individuals contribute to the common pool. The remaining trajectories in the phase space will converge to the alternative destination in which the risk level becomes particularly high and cooperators disappear.
When $T=\frac{M1}{N1}$, the eigenvalues of Jacobian matrix at the interior fixed point are a purely imaginary conjugate pair. Then, according to the Hopf bifurcation theorem (Kuznetsov, 1998; Guckenheimer and Holmes, 2013), the system undergoes a Hopf bifurcation at $T=\frac{M1}{N1}$ and a limit cycle encircling around interior equilibrium emerges. By calculating the first Lyapunov coefficient, we can evaluate that the limit cycle is stable (see Appendix 2). Besides, there exist two boundary fixed points, $({x}_{1}^{*},1)$ and $({x}_{2}^{*},1)$, because $(\genfrac{}{}{0pt}{}{N1}{M1})(\frac{M1}{N1}{)}^{M1}(1\frac{M1}{N1}{)}^{NM}bc>0$. Thus the system has all seven fixed points. As we discuss in Appendix 2, only the fixed point $(0,1)$ is stable. A representative numerical example is shown in the second row of Figure 6, which is conceptually similar to those we observed for System I. More precisely, the population either converges toward a limit cycle in the interior space or arrives to the undesired $(0,1)$ point where there are no cooperators, but just high risk.
The interior fixed point is unstable when $T<\frac{M1}{N1}$. Besides, there are two boundary fixed points, $({x}_{1}^{*},1)$ and $({x}_{2}^{*},1)$, because $(\genfrac{}{}{0pt}{}{N1}{M1})(\frac{M1}{N1}{)}^{M1}(1\frac{M1}{N1}{)}^{NM}bc>0$. In this situation, the system has all seven fixed points. Theoretical analysis, presented in Appendix 2, confirms that only $(0,1)$ is stable. This is illustrated in the third row of Figure 6 where all trajectories terminate in the mentioned point, signaling that the tragedy of the commons state is inevitable.
System II has no interior equilibrium point
When $c\ge (\genfrac{}{}{0pt}{}{N1}{M1}){T}^{M1}(1T{)}^{NM}b$, there is no interior fixed point in System II. In this case, when $(\genfrac{}{}{0pt}{}{N1}{M1})(\frac{M1}{N1}{)}^{M1}(1\frac{M1}{N1}{)}^{NM}bc<0$, there are four equilibrium points, namely, $(0,0),(0,1),(1,0),(1,1)$, where $(0,1)$ is stable. When $(\genfrac{}{}{0pt}{}{N1}{M1})(\frac{M1}{N1}{)}^{M1}(1\frac{M1}{N1}{)}^{NM}bc>0$, there exist two boundary fixed points, $({x}_{1}^{*},1)$ and $({x}_{2}^{*},1)$. Altogether, the system has six fixed points, which are $(0,0)$, $(0,1)$, $(1,0)$, $(1,1)$, $({x}_{1}^{*},1),$ and $({x}_{2}^{*},1)$. As we discuss in Appendix 2, the fixed points $(0,0),(1,0),(1,1),({x}_{1}^{*},1)$ are unstable, while $(0,1)$ is stable. In the special case of ${x}_{2}^{\ast}<T$, the fixed point $({x}_{2}^{*},1)$ becomes stable, which suggests that there is a significant cooperation at a high risk level. A representative numerical illustration is shown in the bottom row of Figure 6, signaling the importance of the initial conditions because the trajectories converge either to the fixed point $(0,1)$ or to $({x}_{2}^{*},1)$.
Discussion
Human behavior and the natural environment are inextricably linked. Motivated by this fact, rapidly growing research efforts have recognized the importance of developing a new comprehensive framework to study the coupled human–environment ecosystem (Stern, 1993; Liu et al., 2007; Farahbakhsh et al., 2022). Starting from the powerful concept of coevolutionary game theory, several works focus on depicting the reciprocal interactions and feedback between human behaviors and natural environment – both the impact of human behaviors on nature and the effects of environment on human behaviors (Weitz et al., 2016; Chen and Szolnoki, 2018; Tilman et al., 2020). Along this research line, we have developed a feedbackevolving game framework to study the coevolutionary dynamics of strategies and environment based on collectiverisk dilemmas. Here, the environmental state is no longer a symbol of resource abundance, but depicts the risk level of collective failure. More precisely, we assume that the frequencies of strategies directly affect the risk level and reversely, the change in risk state stimulates individual behavioral decisionmaking. Importantly, we have explored both linear and highly nonlinear feedback mechanisms that characterize the link between the main system variables.
In particular, we have incorporated the strategiesrisk feedback mechanism into replicator dynamics and explored the possible consequences of coevolutionary dynamics. We have shown that sustainable cooperation level can be reached in the population in two different ways. First, the coevolutionary dynamics can converge to a fixed point. This fixed point can be in the interior, indicating that the frequency of cooperators and the level of risk can be respectively stabilized at a certain level, or at the boundary, indicating that highlevel cooperation can be maintained even at a significantly highrisk environment. Second, the system has a stable limit cycle where persistent oscillations in strategy and risk state can appear. In addition, we have found that the abovedescribed evolutionary outcomes do not depend significantly on the character of feedback mechanism of how strategy change affects on risk level. No matter it is linear or nonlinear, what really counts is the existence of the proper feedback. Importantly, we have theoretically identified those conditions that are responsible for the final dynamical outcomes.
In this work, we introduce a twoway coupling between strategy and environment. Indeed, the effect of a twoway interplay between environments and strategy has been involved in previous works. For example, Hilbe et al. considered that the public resource is changeable and depends on strategic choices of individuals (Hilbe et al., 2018). By analyzing the stochastic dynamics of the system, they found that the interplay between reciprocity and payoff feedback can be crucial for cooperation. And they considered this twoway interplay between environments and strategy in repeated stochastic game with discrete time steps. Differently, in our work we focus on oneshot collectiverisk social dilemma with such twoway interplay. We find that a twoway coupling between collective actions and risk is essential to avoid the tragedy of the commons.
Previous theoretical studies have revealed that the coevolutionary game models describing the complex interactions between collective actions and environment can produce periodic oscillation dynamics (Weitz et al., 2016; Tilman et al., 2020). Although our feedbackevolving game model can also produce persistent oscillations, there are some differences. In particular, we have theoretically proved that Hopf bifurcation can take place and a stable limit cycle can appear in the system, which is different from the heteroclinic cycle dynamics reported by Weitz et al., 2016. Besides, we have found that the existence of a limit cycle does not depend on the speed of coupling (see Appendix 1—figure 1, Appendix 2—figure 1), whereas Tilman et al., 2020 reported the opposite conclusion. Furthermore, we observe that a small amplitude oscillation is more conducive to maintaining the stability of the system than a large magnitude oscillation because a higher risk will make it easier for all individuals to lose all their endowments.
The reciprocal feedback process, though many types have not been well characterized, occurs at all levels of our life (Liu et al., 2007; Ezenwa et al., 2016; Obradovich and Rahwan, 2019). Consequently, they may play an indispensable role in maintaining the stability of human society and the ecosystem. Mathematical modeling based on evolutionary game theory is a powerful tool for addressing social–ecological and human–environment interactions and analyzing the evolutionary dynamics of these coupled systems. The mathematical framework proposed in this article considers two characteristic forms to describe the effect of strategy on risk, namely, linear and nonlinear (exponential) forms of feedback. Although these two forms can be equivalent under some limit conditions, there are essential differences. On the one hand, linear relationship is a relatively simple way to describe the correlation mode of two factors, which is common in real society. For example, with the increase in protection awareness and vaccination proportion, the mortality rate of the epidemic decreased gradually (Yang and Shaman, 2022). Furthermore, linear feedback has been used to describe the interactions between actions of the population and environmental state (Weitz et al., 2016; Tilman et al., 2020). However, linear link cannot fully describe the relationship between variables in real societies. For example, in recent years, extreme weather phenomena have occurred more frequently, with greater intensity and wider impact areas. Thus the feedback between human behaviors and environment may take on a more complex nonlinear form. In this work, we consider that the strategy of the population has an exponential effect on risk level, and such form can describe the phenomenon that risk will rise and fall sharply with the change in strategy frequency (Figure 2). It is worth emphasizing that although we use different forms of feedback to describe the impact of strategies on risk, the evolutionary dynamics have not changed substantially, which highlights the prime importance of the feedback mechanism independently of its actual form.
Our feedbackevolving game model reveals that the coupled strategy and environment system will produce a variety of representative dynamical behaviors. We find that the undesired equilibrium point $(0,1)$ in our feedback system is always evolutionarily stable, which does not depend on whether the effect of strategy on risk is linear or exponential. Such evolutionary outcome means that all individuals are unwilling to contribute to achieving the collective goal, which leads to the failure of collective action, and all individuals inevitably lose their remaining endowments. In realworld scenarios, such as climate change (Milinski et al., 2008) and the spread of infectious diseases (Cronk and Aktipis, 2021; Chen and Fu, 2022), once the whole society is in such a state, it is undoubtedly disastrous for the public. Therefore, how to adjust and control the system to deviate from this state is particularly important for policymakers.
Finally, it is worth emphasizing that the feedback loop operates over time. In this situation, the change in risk state or strategy frequency may lead to the change in other factors, such as collective target, which provides an opportunity for the emergence of new feedback loops. Thus, multiple types of feedback loops are possible in a single coupled system. Such multiple feedback loops have been confirmed in the coupling system of animal behavior and disease ecology (Ezenwa et al., 2016). Therefore, a promising expansion of our current model could be to consider the multiple feedback loops.
Code availability
Appendix 1
We first study the case where the strategy of the population has a linear effect on the risk level. Then the dynamical system can be written as
This equation system has at most seven fixed points, which are $(0,0)$, $(0,1)$, $(1,0)$, $(1,1)$, $(\frac{u}{1+u},\text{}\frac{c}{(\genfrac{}{}{0pt}{}{N1}{M1})(\frac{u}{1+u}{)}^{M1}(\frac{1}{1+u}{)}^{NM}b})$, $({x}_{1}^{\ast},1)$ and $({x}_{2}^{*},1)$, where ${x}_{1}^{*}$ and ${x}_{2}^{*}$ are the real roots of the equation $\left(\genfrac{}{}{0pt}{}{N1}{M1}\right){x}^{M1}{(1x)}^{NM}b=c$. For convenience, we introduce the abbreviation ${r}^{*}=\frac{c}{\left(\genfrac{}{}{0pt}{}{N1}{M1}\right){(\frac{u}{1+u})}^{M1}{(\frac{1}{1+u})}^{NM}b}$ and $\mathrm{\Gamma}(x)=\left(\genfrac{}{}{0pt}{}{N1}{M1}\right){x}^{M1}{(1x)}^{NM}$. In the following, we analyze the stability of these equilibrium points.
(1) When $0<{r}^{\ast}<1$, namely, $\mathrm{\Gamma}(\frac{u}{1+u})>\frac{c}{b}$, the system has an interior fixed point. Accordingly, the Jacobian for the interior fixed point is
where ${\overline{a}}_{11}=\frac{c}{\epsilon}[M1\frac{u(N1)}{u+1}]$, ${\overline{a}}_{12}=\frac{1}{\epsilon}(\genfrac{}{}{0pt}{}{N1}{M1})(\frac{u}{1+u}{)}^{M}(\frac{1}{1+u}{)}^{NM+1}b$, and ${\overline{a}}_{21}={r}^{*}(1{r}^{*})(1+u)$.
(i) When ${\overline{a}}_{11}>0$, namely, $u<\frac{M1}{NM}$, the existing interior fixed point is unstable. Since $\mathrm{\Gamma}(\frac{M1}{N1})>\frac{c}{b}$, we can know that the two boundary fixed points $({x}_{1}^{*},1)$ and $({x}_{2}^{*},1)$ exist. Thus, the system has seven fixed points in the parameter space, namely, $(0,0)$, $(0,1)$, $(1,0)$, $(1,1)$, $(\frac{u}{1+u},{r}^{*})$, $({x}_{1}^{*},1),$ and $({x}_{2}^{*},1)$. The Jacobian matrices of these equilibrium points are respectively given as follows.
For $(x,r)=(0,0)$, the Jacobian is
thus the fixed equilibrium is unstable.
For $(x,r)=(0,1)$, the Jacobian is
thus the fixed equilibrium is stable.
For $(x,r)=(1,0)$, the Jacobian is
thus the fixed equilibrium is unstable.
For $(x,r)=(1,1)$, the Jacobian is
thus the fixed equilibrium is unstable.
For $(x,r)=({x}_{1}^{*},1)$, the Jacobian is
thus the fixed equilibrium is unstable since $x}_{1}^{\ast}<\frac{M1}{N1$.
For $(x,r)=({x}_{2}^{*},1)$, the Jacobian is
because $u<\frac{M1}{NM}$ and $x}_{2}^{\ast}>\frac{M1}{N1$, then $1\frac{1}{1+u}<\frac{M1}{N1}<{x}_{2}^{\ast}$. Thus this fixed equilibrium is unstable.
(ii) When ${\overline{a}}_{11}=0$, namely, $u=\frac{M1}{NM}$, the trace and determinant of the Jacobian matrix at the interior equilibrium point are respectively given by
The eigenvalues of the Jacobian matrix can be calculated
where $\overline{\mu}=\frac{{\overline{a}}_{11}}{2}=0$ and ${\overline{w}}^{2}={\overline{a}}_{12}{\overline{a}}_{21}$.
Accordingly, we know that the eigenvalues satisfy the following conditions:
The first two conditions imply that the eigenvalues of Jacobian matrix at $(\frac{u}{1+u},{r}^{*})$ has a pair of pure imaginary roots. The third condition means that the pair of complexconjugate eigenvalues crosses the imaginary axis with nonzero speed. According to Hopf bifurcation theorem (Kuznetsov, 1998), we know that a Hopf bifurcation takes place at $u=\frac{M1}{NM}$. In order to determine the stability of the existing limit cycle from Hopf bifurcation, we need to calculate the first Lyapunov coefficient. We denote that ${F}_{1}(x,r)=\frac{x(1x)}{\epsilon}[\left(\genfrac{}{}{0pt}{}{N1}{M1}\right){x}^{M1}{(1x)}^{NM}rbc]$ and ${F}_{2}(x,r)=r(1r)[u(1x)x]$.
Let $q,p\in {\u2102}^{2}$ respectively denote the eigenvectors of the Jacobian matrix $J(T,{r}^{*})$ and its transpose. We then have
which satisfy
To achieve the necessary normalization $<p,q>=\overline{{p}_{1}}{q}_{1}+\overline{{p}_{2}}{q}_{2}=1$, we can take
According to Kuznetsov, 1998, we construct the complexvalued function
where $p,q$ are given above, to evaluate its formal partial derivatives with respect to $x,r$ at $(T,{r}^{*})$, obtaining ${g}_{20}={G}_{xx},{g}_{11}={G}_{xr}$, and ${g}_{21}={G}_{xxr}$. After some calculations, we can get the first Lyapunov coefficient
Specifically, when ${l}_{1}<0$, a unique and stable limit cycle bifurcates from the equilibrium appears, while when ${l}_{1}>0$, the Hopf bifurcation is subcritical such that an unstable limit cycle will be generated. Due to the complexity of the system, it is difficult to conduct bifurcation analysis collectively. Here, we conduct a numerical analysis to investigate the stability of the existing limit cycle when the model parameters are consistent with Figure 4d. By using the algorithm in Kuznetsov, 1998, we can get ${l}_{1}=1.407166124\times {10}^{8}<0$, which implies that the Hopf bifurcation is supercritical.
Besides, since $\mathrm{\Gamma}(\frac{M1}{N1})>\frac{c}{b}$, we can state that the two boundary fixed points $({x}_{1}^{*},1)$ and $({x}_{2}^{*},1)$ exist. Thus the system has seven equilibrium points, which are $(0,0)$, $(0,1)$, $(1,0)$, $(1,1)$, $(\frac{u}{1+u},{r}^{*})$, $({x}_{1}^{*},1),$ and $({x}_{2}^{*},1)$, respectively. Accordingly to the sign of the eigenvalues of the Jacobian matrices, we know that only $(0,1)$ is stable.
(iii) When ${\overline{a}}_{11}<0$, namely, $u>\frac{M1}{NM}$, the trace and determinant of the Jacobian matrix at the interior equilibrium point are respectively given by
Thus the interior fixed point is stable. Besides, since $\mathrm{\Gamma}(\frac{M1}{N1})>\frac{c}{b}$, two boundary fixed points, $({x}_{1}^{*},1)$ and $({x}_{2}^{*},1)$, exist. Thus there are seven fixed points in the system, which are $(0,0)$, $(0,1)$, $(1,0)$, $(1,1)$, $(\frac{u}{1+u},{r}^{*})$, $({x}_{1}^{*},1),$ and $({x}_{2}^{*},1)$, respectively. Here, the fixed points $(0,1)$ and $(\frac{u}{1+u},{r}^{*})$ are stable, while others are unstable.
(2) When ${r}^{\ast}\ge 1$, namely, $\mathrm{\Gamma}(\frac{u}{1+u})\le \frac{c}{b}$, the system has no interior equilibrium point. In this case, when $\mathrm{\Gamma}(\frac{M1}{N1})>\frac{c}{b}$, the system has six fixed points, which are $(0,0),(0,1),(1,0),(1,1),({x}_{1}^{*},1),$ and $({x}_{2}^{*},1)$, respectively. According to the sign of the largest eigenvalues of the Jacobian matrices, we can say that $(0,0),(1,0),(1,1),({x}_{1}^{*},1)$ are unstable, while $(0,1)$ is stable. Particularly, when $x}_{2}^{\ast}<\frac{u}{1+u$, the fixed point $({x}_{2}^{*},1)$ is stable, and it is unstable when $x}_{2}^{\ast}>\frac{u}{1+u$. When ${x}_{2}^{*}=\frac{u}{1+u}$, we know that one eigenvalue of the Jacobian matrix is zero and the other eigenvalue is negative. Then we study its stability by using the center manifold theorem (Khalil, 1996). For the fixed point $({x}_{2}^{*},1)$, the Jacobian matrix can be written as
where ${\gamma}_{11}=\frac{c}{\epsilon}(M1{x}_{2}^{*}(N1))$ and ${\gamma}_{12}=\frac{c}{\epsilon}{x}_{2}^{*}(1{x}_{2}^{*})$. To do that, we take ${z}_{1}=x{x}_{2}^{*}$ and ${z}_{2}=r1$, then the system can be rewritten as
Let $Q$ be a matrix whose columns are the eigenvectors of $J({x}_{2}^{*},1)$, which can be written as
Then we have
We further take ${[{\eta}_{1}\mathit{\hspace{1em}}{\eta}_{2}]}^{T}={Q}^{1}[{z}_{1}\mathit{\hspace{1em}}{z}_{2}]$, and then we have ${\eta}_{1}={z}_{1}+\frac{{\gamma}_{12}}{{\gamma}_{11}}{z}_{2}$ and ${\eta}_{2}={z}_{2}$. We then have
According to the center manifold theorem, we know that ${\eta}_{1}=h({\eta}_{2})$ is a center manifold. Then we start to try $h({\eta}_{2})=O({{\eta}_{2}}^{2})$, which yields the reduced system
Since $(1+u)\frac{{\gamma}_{12}}{{\gamma}_{11}}\ne 0$, the fixed point ${\eta}_{2}=0$ of the reduced system is unstable. Accordingly, the fixed point $({x}_{2}^{*},1)$ of the original system is unstable.
When $\mathrm{\Gamma}(\frac{M1}{N1})=\frac{c}{b}$, the system has five fixed points, which are $(0,0),(0,1),(1,0),(1,1),$ and $(\frac{M1}{N1},1)$, respectively. According to the sign of the eigenvalues in the Jacobian matrices, we can state that only $(0,1)$ is stable. When $\mathrm{\Gamma}(\frac{M1}{N1})<\frac{c}{b}$, the system has four fixed points, namely, $(0,0),(0,1),(1,0)$, and $(1,1)$. Here, only $(0,1)$ is stable.
Appendix 2
System II with exponential feedback is described by
where the parameter $\beta >0$ represents the steepness of the function.
This equation system has at most seven fixed points, which are $(0,0)$, $(0,1)$, $(1,0)$, $(1,1)$, $(T,\frac{c}{(\genfrac{}{}{0pt}{}{N1}{M1}){T}^{M1}(1T{)}^{NM}b})$, $({x}_{1}^{\ast},1)$ and $({x}_{2}^{*},1)$, where ${x}_{1}^{*}$ and ${x}_{2}^{*}$ are the real roots of the equation $\left(\genfrac{}{}{0pt}{}{N1}{M1}\right){x}^{M1}{(1x)}^{NM}b=c$ and $x}_{1}^{\ast}<\frac{M1}{N1}<{x}_{2}^{\ast$. For simplicity, we introduce the abbreviation $\overline{r}=\frac{c}{\left(\genfrac{}{}{0pt}{}{N1}{M1}\right){T}^{M1}{(1T)}^{NM}b}$ and $\mathrm{\Gamma}(x)=\left(\genfrac{}{}{0pt}{}{N1}{M1}\right){x}^{M1}{(1x)}^{NM}$. In the following, we study the stabilities of equilibria based on whether the system has an interior equilibrium point.
(1) When $0<\overline{r}<1$, namely, $\mathrm{\Gamma}(T)>\frac{c}{b}$, System II has an interior equilibrium point.
The Jacobian matrix evaluated at this equilibrium is
where ${a}_{11}=\frac{c}{\epsilon}(M1T(N1))$, ${a}_{12}=\frac{1}{\epsilon}(\genfrac{}{}{0pt}{}{N1}{M1}){T}^{M}(1T{)}^{NM+1}b$, and ${a}_{21}=\frac{\overline{r}(1\overline{r})\beta}{2}$. Notice that $\frac{1}{\epsilon}(\genfrac{}{}{0pt}{}{N1}{M1}){T}^{M}(1T{)}^{NM+1}b>0$ and $\frac{\overline{r}(1\overline{r})\beta}{2}<0$, then the trace and determinant of the Jacobian matrix are respectively given by
The eigenvalues of the Jacobian matrix can be calculated as
Here, we set that $\mu (T)=\frac{{a}_{11}}{2},{w}^{2}(T)=\frac{{a}_{11}^{2}+4{a}_{12}{a}_{21}}{4},$ and ${T}_{0}=\frac{M1}{N1}.$
(i) When ${a}_{11}>0$, namely, $T<{T}_{0}$, the interior equilibrium point is unstable. Since $\mathrm{\Gamma}({T}_{0})>\frac{c}{b}$, we can know that the two boundary fixed points $({x}_{1}^{*},1)$ and $({x}_{2}^{*},1)$ exist. Thus, the system has seven fixed points in the parameter space, namely, $(0,0),(0,1)$, $(1,0),(1,1),(T,\overline{r}),({x}_{1}^{*},1),$ and $({x}_{2}^{*},1)$.
For $(x,r)=(0,0)$, the Jacobian is
thus the fixed equilibrium is unstable.
For $(x,r)=(0,1)$, the Jacobian is
thus the equilibrium point is stable.
For $(x,r)=(1,0)$, the Jacobian is
thus the fixed point is unstable.
For $(x,r)=(1,1)$, the Jacobian is
thus the fixed equilibrium is unstable.
For $(x,r)=({x}_{1}^{*},1)$, the Jacobian is
thus the fixed equilibrium is unstable since $x}_{1}^{\ast}<{T}_{0$.
For $(x,r)=({x}_{2}^{*},1)$, the Jacobian is
thus the fixed equilibrium is unstable since $T<{T}_{0}<{x}_{2}^{\ast}$.
(ii) When ${a}_{11}=0$, namely, $T={T}_{0}=\frac{M1}{N1}$, we have $\mu ({T}_{0})=0$. Moreover, ${w}^{2}(T)=\frac{{a}_{11}^{2}+4{a}_{12}{a}_{21}}{4}=$ $\frac{1}{\epsilon}(\genfrac{}{}{0pt}{}{N1}{M1}){T}^{M}(1T{)}^{NM+1}b\frac{\overline{r}(1\overline{r})\beta}{2}>0$. Therefore, the eigenvalues of the Jacobian matrix are a purely imaginary conjugate pair ${\lambda}_{1,2}({T}_{0})=\pm iw({T}_{0})$. Considering that $\frac{\mathrm{\partial}\mu (T)}{\mathrm{\partial}T}{}_{{T}_{0}}=\frac{c(N1)}{2\epsilon}<0$, then we know that the system undergoes a Hopf bifurcation at $T={T}_{0}$ and there exists a limit cycle around the interior equilibrium. Accordingly, we can evaluate the direction of the limit cycle bifurcation by computing the first Lyapunov coefficient l_{1} of the system. Here, we also conduct numerical calculations to investigate the stability of the existing limit cycle when the model parameters are consistent with Figure 6d. By using the algorithm in Kuznetsov, 1998, we can get ${l}_{1}=1.876221498\times {10}^{8}$, which implies that the Hopf bifurcation is supercritical.
Besides, since $\mathrm{\Gamma}({T}_{0})>\frac{c}{b}$, we know that there are seven equilibrium points in System II. They are $(0,0)$, $(0,1)$, $(1,0)$, $(1,1)$, $(T,\overline{r})$, $({x}_{1}^{*},1),$ and $({x}_{2}^{*},1)$. According to the sign of the eigenvalues of the Jacobian matrices, only $(0,1)$ is stable.
(iii) When ${a}_{11}<0$, namely, $T>{T}_{0}$, the interior equilibrium point is stable. Besides, since $\mathrm{\Gamma}({T}_{0})>\frac{c}{b}$, we find that there are seven fixed points in the system, which are $(0,0)$, $(0,1)$, $(1,0)$, $(1,1)$, $(T,\overline{r})$, $({x}_{1}^{*},1),$ and $({x}_{2}^{*},1)$, respectively. Here, the fixed points $(0,1)$ and $(T,\overline{r})$ are stable, while others are unstable.
(2) When $\overline{r}\ge 1$, namely, $\mathrm{\Gamma}(T)\le \frac{c}{b}$, System II has no interior equilibrium point. In this case, when $\mathrm{\Gamma}({T}_{0})>\frac{c}{b}$, the system has six fixed points, which are $(0,0),(0,1),(1,0),(1,1),({x}_{1}^{*},1),$ and $({x}_{2}^{*},1)$, respectively. According to the sign of the largest eigenvalues of the Jacobian matrices, we can say that $(0,0),(1,0),(1,1),({x}_{1}^{*},1)$ are unstable, while $(0,1)$ is stable. Particularly, when ${x}_{2}^{\ast}<T$, the fixed point $({x}_{2}^{*},1)$ is stable, and it is unstable when ${x}_{2}^{\ast}>T$. When $\mathrm{\Gamma}({T}_{0})=\frac{c}{b}$, the system has five fixed points, which are $(0,0),(0,1),(1,0),(1,1),$ and $({T}_{0},1)$, respectively. According to the sign of the eigenvalues in the Jacobian matrices, we can see that only $(0,1)$ is stable. When $\mathrm{\Gamma}({T}_{0})<\frac{c}{b}$, the system has four fixed points, namely $(0,0),(0,1),(1,0)$, and $(1,1)$. Here, only $(0,1)$ is stable.
Data availability
The Mathematica (Wolfram Mathematica 11.1) and Matlab (Matlab R2014a) source codes used to generate Figures 4 and 6 are available on Dryad.

Dryad Digital RepositoryCode for: Coevolutionary dynamics via adaptive feedback in collectiverisk social dilemma game.https://doi.org/10.5061/dryad.wdbrv15rz
References

BookThe effects of climate change on human behaviorsIn: Fahad S, Hasanuzzaman M, Alam M, Ullah H, Saeed M, Khan IA, Adnan M, editors. Environment, Climate, Plant and Aegetation Growth. Springer. pp. 577–589.https://doi.org/10.1007/9783030497323

Riskdriven migration and the collectiverisk social dilemmaPhysical Review E 86:036101.https://doi.org/10.1103/PhysRevE.86.036101

Punishment and inspection for governing the commons in a feedbackevolving gamePLOS Computational Biology 14:e1006347.https://doi.org/10.1371/journal.pcbi.1006347

Imperfect vaccine and hysteresisProceedings of the Royal Society BBiological Sciences 286:20182406.https://doi.org/10.1098/rspb.2018.2406

Highly coordinated nationwide massive travel restrictions are central to effective mitigation and control of covid19 outbreaks in ChinaProceedings of the Royal Society AMathematical Physical and Engineering Sciences 478:20220040.https://doi.org/10.1098/rspa.2022.0040

Cooperation dynamics under pandemic risks and heterogeneous economic interdependenceChaos, Solitons, and Fractals 155:111655.https://doi.org/10.1016/j.chaos.2021.111655

Governance of risky public goods under graduated punishmentJournal of Theoretical Biology 505:110423.https://doi.org/10.1016/j.jtbi.2020.110423

Design principles for riskpooling systemsNature Human Behaviour 5:825–833.https://doi.org/10.1038/s41562021011219

In a warmer Arctic, mosquitoes avoid increased mortality from predators by growing fasterProceedings of the Royal Society BBiological Sciences 282:20151549.https://doi.org/10.1098/rspb.2015.1549

Host behaviourparasite feedback: an essential link between animal behaviour and disease ecologyProceedings of the Royal Society BBiological Sciences 283:1828.https://doi.org/10.1098/rspb.2015.3078

Modelling coupled humanenvironment complexity for the future of the biosphere: strengths, gaps and promising directionsPhilosophical Transactions of the Royal Society BBiological Sciences 377:20210382.https://doi.org/10.1098/rstb.2021.0382

BookNonlinear Oscillations, Dynamical Systems, and Bifurcations of Vector FieldsSpringer Science & Business Media.https://doi.org/10.1007/9781461211402

Asymmetric evolutionary games with environmental feedbackJournal of Theoretical Biology 462:347–360.https://doi.org/10.1016/j.jtbi.2018.11.019

BookEvolution and the Theory of GamesCambridge University Press.https://doi.org/10.1017/CBO9780511806292

Benefits of influenza vaccination for low, intermediate, and highrisk senior citizensArchives of Internal Medicine 158:1769–1776.https://doi.org/10.1001/archinte.158.16.1769

Risk of a feedback loop between climatic warming and human mobilityJournal of the Royal Society Interface 16:20190058.https://doi.org/10.1098/rsif.2019.0058

Climate change governance, cooperation and selforganizationPhysics of Life Reviews 11:573–586.https://doi.org/10.1016/j.plrev.2014.02.003

Statistical physics of human cooperationPhysics Reports 687:1–51.https://doi.org/10.1016/j.physrep.2017.05.004

Replicator dynamicsJournal of Theoretical Biology 100:533–538.https://doi.org/10.1016/00225193(83)904459

BookGlobal Change and the Earth System: A Planet under PressureSpringer Berlin, Heidelberg.https://doi.org/10.1007/b137870

Collapse of cooperation in evolving gamesPNAS 111:17558–17563.https://doi.org/10.1073/pnas.1408618111

Evolution of prosocial behaviours in multilayer populationsNature Human Behaviour 6:338–348.https://doi.org/10.1038/s41562021012412

BookSociophysics Approach to EpidemicsSingapore: Springer.https://doi.org/10.1007/9789813364813

Evolutionary stable strategies and game dynamicsMathematical Biosciences 40:145–156.https://doi.org/10.1016/00255564(78)900779

Evolutionary games with environmental feedbacksNature Communications 11:915.https://doi.org/10.1038/s41467020145316

Can influenza epidemics be prevented by voluntary vaccination?PLOS Computational Biology 3:e85.https://doi.org/10.1371/journal.pcbi.0030085

A bottomup institutional approach to cooperative governance of risky commonsNature Climate Change 3:797–801.https://doi.org/10.1038/nclimate1927

Universal scaling for the dilemma strength in evolutionary gamesPhysics of Life Reviews 14:1–30.https://doi.org/10.1016/j.plrev.2015.04.033

Cooperator driven oscillation in a timedelayed feedbackevolving gameNew Journal of Physics 23:053017.https://doi.org/10.1088/13672630/abf205

COVID19 pandemic dynamics in India, the sarscov2 delta variant and implications for vaccinationJournal of the Royal Society Interface 19:191.https://doi.org/10.1098/rsif.2021.0900
Decision letter

Raymond E GoldsteinReviewing Editor; University of Cambridge, United Kingdom

Christian RutzSenior Editor; University of St Andrews, United Kingdom
Our editorial process produces two outputs: (i) public reviews designed to be posted alongside the preprint for the benefit of readers; (ii) feedback on the manuscript for the authors, including requests for revisions, shown below. We also include an acceptance summary that explains what the editors found interesting or important about the work.
Decision letter after peer review:
Thank you for submitting your article "Coevolutionary dynamics via adaptive feedback in collectiverisk social dilemma game" for consideration by eLife. Your article has been reviewed by 2 peer reviewers, and the evaluation has been overseen by a Reviewing Editor and Christian Rutz as the Senior Editor. The reviewers have opted to remain anonymous.
The reviewers have discussed their reviews with one another, and the Reviewing Editor has drafted this decision letter to help you prepare a revised submission.
Essential revisions:
1) A technical suggestion to improve the universality of their model.
As mentioned in the review, the authors regarded c/b as a parameter representing 'dilemma strength', which has been commonly adopted by many previous works. This is fine. Although we do not know if the authors have recognized this or not, there is an established idea; 'the concept of universal dilemma strength parameter for 2 by 2 games'. If they leave a PGG and come to a 2by2 game version, it is socalled Donor & Recipient (D & R) game, where T := b, R := b – c, P := , and S := – c are imposed. Based on the concept, we quantify the Chickentype dilemma; Dr' := (T – R) / (R – P) = c/(b – c) and Stag Hunttype dilemma; Dr' := (P – S) / (R – P) = c/(b – c). Thus, a D & R game; 2 by 2 version of PGG, happens to have the same quantities of Dg' and Dr', which leads to a specific situation that the extent of social dilemma can be evaluated by a single model parameter; c / (b – c). Although we would not go so far as to say that they should replace 'c/b' by 'c/(b – c)', they should supplement in either Introduction or Model depiction part with the assertion that a PGG is a special and extended version of D & R game by referring to the concept of universal dilemma strength with citations to several relevant literatures; (i) Sociophysics Approach to Epidemics, Springer, 2021, (ii) Universal scaling for the dilemma strength in evolutionary games, Physics of Life Reviews 14, 130, 2015, (iii) Scaling the phase planes of social dilemma strengths shows gameclass changes in the five rules governing the evolution of cooperation, Royal Society Open Science, 181085, 2018.
2) While this is a solid paper, the presentation and flow could be improved. For one, it would be helpful to have a table with an overview of all the parameters and variables that are introduced over the course of the paper. Also, in the Results section, we believe it would help the flow and reading experience if there was more intuition about the actual interpretation of the equilibrium points. This means mentioning already earlier that (0,1) is the undesirable full defection equilibrium instead of only noting this in line 202, and explaining the meaning behind the other fixed points as well.
3) In terms of flow, it might also be useful to tighten up the Results and Discussion section considerably. In the former, some more technical details can be left out, given the very comprehensive appendix; furthermore, there are some redundant sentences (e.g., it would be enough to mention once that feedback speed does not influence the dynamics in any of the cases). This holds even more strongly for the Discussion: it is quite longwinded and would benefit from tightening and shortening, including a more concise summary of the results and less similarity with the Introduction, as well as removing redundant information where similar sentences appear in multiple paragraphs (e.g. about the robustness of the system to different feedback forms).
4) Finally, we think it would be interesting to compare the results of this paper in more depth with papers like (Hilbe, C., Šimsa, Š., Chatterjee, K., and Nowak, M. A. (2018). Evolution of cooperation in stochastic games. Nature, 446 559(7713):246249), which was already cited in the manuscript but not expanded upon. The authors of that work analyze a repeated stochastic game with discrete time steps, which is quite different from the model considered in the present manuscript but has a similar aim of describing how cooperation can evolve when there is a twoway interplay between environment and strategy. Such a comparison also makes sense with regard to the sentence that " a twoway coupling…is essential to avoid the tragedy of the commons" (Abstract, line 22).
https://doi.org/10.7554/eLife.82954.sa1Author response
Essential revisions:
1) A technical suggestion to improve the universality of their model.
As mentioned in the review, the authors regarded c/b as a parameter representing 'dilemma strength', which has been commonly adopted by many previous works. This is fine. Although we do not know if the authors have recognized this or not, there is an established idea; 'the concept of universal dilemma strength parameter for 2 by 2 games'. If they leave a PGG and come to a 2by2 game version, it is socalled Donor & Recipient (D & R) game, where T := b, R := b – c, P := , and S := – c are imposed. Based on the concept, we quantify the Chickentype dilemma; Dr' := (T – R) / (R – P) = c/(b – c) and Stag Hunttype dilemma; Dr' := (P – S) / (R – P) = c/(b – c). Thus, a D & R game; 2 by 2 version of PGG, happens to have the same quantities of Dg' and Dr', which leads to a specific situation that the extent of social dilemma can be evaluated by a single model parameter; c / (b – c). Although we would not go so far as to say that they should replace 'c/b' by 'c/(b – c)', they should supplement in either Introduction or Model depiction part with the assertion that a PGG is a special and extended version of D & R game by referring to the concept of universal dilemma strength with citations to several relevant literatures; (i) Sociophysics Approach to Epidemics, Springer, 2021, (ii) Universal scaling for the dilemma strength in evolutionary games, Physics of Life Reviews 14, 130, 2015, (iii) Scaling the phase planes of social dilemma strengths shows gameclass changes in the five rules governing the evolution of cooperation, Royal Society Open Science, 181085, 2018.
Thanks for the constructive suggestion. We have read them with great interest and found that all these mentioned works are conducive to reveal the role of dilemma strength in cooperation, and accordingly cited these relevant literatures in the Model section of the revised version.
2) While this is a solid paper, the presentation and flow could be improved. For one, it would be helpful to have a table with an overview of all the parameters and variables that are introduced over the course of the paper. Also, in the Results section, we believe it would help the flow and reading experience if there was more intuition about the actual interpretation of the equilibrium points. This means mentioning already earlier that (0,1) is the undesirable full defection equilibrium instead of only noting this in line 202, and explaining the meaning behind the other fixed points as well.
We thank the Reviewer for the valuable suggestions. We agree upon the Reviewer's point and we have added a table with an overview of all the parameters and variables that are introduced over the course of the paper in the revised version. In addition, we have added the actual interpretation of the equilibrium points in Result section to help the flow and reading experience.
3) In terms of flow, it might also be useful to tighten up the Results and Discussion section considerably. In the former, some more technical details can be left out, given the very comprehensive appendix; furthermore, there are some redundant sentences (e.g., it would be enough to mention once that feedback speed does not influence the dynamics in any of the cases). This holds even more strongly for the Discussion: it is quite longwinded and would benefit from tightening and shortening, including a more concise summary of the results and less similarity with the Introduction, as well as removing redundant information where similar sentences appear in multiple paragraphs (e.g. about the robustness of the system to different feedback forms).
We thank the Reviewer for his/her comments. We have removed some technical details and redundant sentences to tighten up the Results and Discussion section considerably. Concretely, we have removed the robustness of the system to different feedback forms and reduced the description of the impact of feedback speed on the evolutionary dynamics.
4) Finally, we think it would be interesting to compare the results of this paper in more depth with papers like (Hilbe, C., Šimsa, Š., Chatterjee, K., and Nowak, M. A. (2018). Evolution of cooperation in stochastic games. Nature, 446 559(7713):246249), which was already cited in the manuscript but not expanded upon. The authors of that work analyze a repeated stochastic game with discrete time steps, which is quite different from the model considered in the present manuscript but has a similar aim of describing how cooperation can evolve when there is a twoway interplay between environment and strategy. Such a comparison also makes sense with regard to the sentence that " a twoway coupling…is essential to avoid the tragedy of the commons" (Abstract, line 22).
This is a great comment. We have added a paragraph in the Discussion section to compare the results of this paper in more depth with papers like (Hilbe, C., Šimsa, Š., Chatterjee, K., and Nowak, M. A. (2018). Evolution of cooperation in stochastic games. Nature, 446 559(7713):246249). Indeed, we construct a quite different model, but have a similar aim of describing how cooperation can evolve when there is a twoway interplay between environment and strategy.
https://doi.org/10.7554/eLife.82954.sa2Article and author information
Author details
Funding
National Natural Science Foundation of China (61976048)
 Xiaojie Chen
National Natural Science Foundation of China (62036002)
 Xiaojie Chen
Natural Science Foundation of Shaanxi Province (JCQN0791)
 Linjie Liu
National Research, Development and Innovation Office (K142948)
 Attila Szolnoki
The funders had no role in study design, data collection and interpretation, or the decision to submit the work for publication.
Acknowledgements
This research was supported by the National Natural Science Foundation of China (grant nos. 61976048 and 62036002) and the Fundamental Research Funds of the Central Universities of China. LL acknowledges the support from Special Project of Scientific and Technological Innovation (grant no. 2452022012) and the Natural Science Foundation of Shaanxi Province (grant no. JCQN0791). AS was supported by the National Research, Development and Innovation Office (NKFIH), Hungary under grant no. K142948.
Senior Editor
 Christian Rutz, University of St Andrews, United Kingdom
Reviewing Editor
 Raymond E Goldstein, University of Cambridge, United Kingdom
Version history
 Received: August 24, 2022
 Preprint posted: December 20, 2022 (view preprint)
 Accepted: April 26, 2023
 Version of Record published: May 19, 2023 (version 1)
Copyright
© 2023, Liu et al.
This article is distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use and redistribution provided that the original author and source are credited.
Metrics

 422
 Page views

 85
 Downloads

 14
 Citations
Article citation count generated by polling the highest count across the following sources: Crossref, PubMed Central, Scopus.
Download links
Downloads (link to download the article as PDF)
Open citations (links to open the citations from this article in various online reference manager services)
Cite this article (links to download the citations from this article in formats compatible with various reference manager tools)
Further reading

 Computational and Systems Biology
The microbial community composition in the human gut has a profound effect on human health. This observation has lead to extensive use of microbiome therapies, including overthecounter 'probiotic' treatments intended to alter the composition of the microbiome. Despite so much promise and commercial interest, the factors that contribute to the success or failure of microbiometargeted treatments remain unclear. We investigate the biotic interactions that lead to successful engraftment of a novel bacterial strain introduced to the microbiome as in probiotic treatments. We use pairwise genomescale metabolic modeling with a generalized resource allocation constraint to build a network of interactions between taxa that appear in an experimental engraftment study. We create induced subgraphs using the taxa present in individual samples and assess the likelihood of invader engraftment based on network structure. To do so, we use a generalized LotkaVolterra model, which we show has strong ability to predict if a particular invader or probiotic will successfully engraft into an individual's microbiome. Furthermore, we show that the mechanistic nature of the model is useful for revealing which microbemicrobe interactions potentially drive engraftment.

 Computational and Systems Biology
Angiogenesis is a morphogenic process resulting in the formation of new blood vessels from preexisting ones, usually in hypoxic microenvironments. The initial steps of angiogenesis depend on robust differentiation of oligopotent endothelial cells into the Tip and Stalk phenotypic cell fates, controlled by NOTCHdependent cell–cell communication. The dynamics of spatial patterning of this cell fate specification are only partially understood. Here, by combining a controlled experimental angiogenesis model with mathematical and computational analyses, we find that the regular spatial Tip–Stalk cell patterning can undergo an order–disorder transition at a relatively high input level of a proangiogenic factor VEGF. The resulting differentiation is robust but temporally unstable for most cells, with only a subset of presumptive Tip cells leading sprout extensions. We further find that sprouts form in a manner maximizing their mutual distance, consistent with a Turinglike model that may depend on local enrichment and depletion of fibronectin. Together, our data suggest that NOTCH signaling mediates a robust way of cell differentiation enabling but not instructing subsequent steps in angiogenic morphogenesis, which may require additional cues and selforganization mechanisms. This analysis can assist in further understanding of cell plasticity underlying angiogenesis and other complex morphogenic processes.