Abstract
Evolutionary dynamics is well captured by the replicator equations when the population is infinite and well-mixed. However, the extinction dynamics is modified with finite and structured populations. Experiments on the non-transitive ecosystem containing three populations of bacteria found that the ecological stability sensitively depends on the spatial structure of the populations. Based on the Reference–Gamble–Birth algorithm, we use agent-based Monte Carlo simulations to investigate the extinction dynamics in the rock-paper-scissors ecosystem with finite and structured populations. On the fully-connected network, the extinction time in stable and unstable regimes falls into two universal functions when plotted with the rescaled variables. On the two dimensional grid, the spatial structure changes the transition boundary between stable and unstable regimes but doesn’t change its extinction trend. The finding of universal scaling in extinction dynamics is unexpected, and may provide a powerful method to classify different evolutionary dynamics into universal classes.
Similar content being viewed by others
Introduction
The paramount question in evolution is how to maintain robust biodiversity in an ecosystem hosting mutually competing species1,2,3,4,5. The opposite side of the same question is about extinction mechanism. Derived from the evolutionary game theory, replicator equations6,7,8,9,10,11 deliver a deterministic description for population dynamics with frequency-dependent selection. The competitions between individuals of different species are described by the payoff matrix12,13 in the game theory and the resultant replicative dynamics has been rather successful in many ecosystems11. Despite of the general applicability with success, it is important to keep in mind that the replicative dynamics is limited to the well-mixed configurations in the infinite population limit.
What happens if the populations are finite and/or competitions between individuals occur on a network with specific spatial structure? Many studies14,15,16,17,18,19,20,21,22,23,24,25,26,27,28,29,30,31 show that ecosystems with finite populations and/or on structured networks reveal significant deviations from the replicative dynamics. For instance, in a recent study32 on the non-transitive rock-paper-scissors ecosystem containing three populations of Escherichia coli. They found that diversity is rapidly lost in the ecosystem when dispersal and interaction occur over relatively large spatial scales, while all species coexist when ecological competitions are localized. That is to say, the ecological stability seems to be weakened in the well-mixed populations. Their findings inspire us to explore the role of spatial structure in evolutionary dynamics.
In addition, there has been controversy over the effect of finite population in the rock-paper-scissors ecosystem. Some study33 shows that a critical population size \(N_c\) exists: For \(N>N_c\), the infinite-population behavior is recovered, while for \(N<N_c\), the stochastic fluctuations arisen from finite population drives the ecosystem toward extinction. On the other hand, the other study34claims that the extinction time is always finite and the ecological stability needs to be defined by the trend of the extinction dynamics. The ecological stability of an ecosystem with finite population means that the extinction time grows exponentially with the population size, \(T_{ex} \sim \exp (N/N^*)\), where \(N^*\) is some characteristic constant depending on the model details. This important conflict together with the previous experimental observations encourages us to investigate these issues by numerical simulations.
To put the discussions on firm ground, we concentrate on the non-transitive rock-paper-scissors game35,36,37,37,38,39,40,41,42,43,44,45,46,47,48,49,50,51,52,53,54,55,56, known as a paradigm to illustrate the species biodiversity. Recent empirical investigations have identified several ecological systems as the rock-paper-scissors system including the coral reef invertebrates57, the mating strategies of lizards in California58, the bacterial game in vitro and in vivo32,59,60, the vertebrate community in the high-Arctic tundra in Greenland61 and the plant communities62,63. These findings in nature make rock-paper-scissors game not only a theoretically interesting model but also a practical system to study.
In this Report, we use agent-based Monte Carlo simulations to investigate the extinction dynamics in the rock-paper-scissors ecosystem with finite and structured populations. Making use of numerical simulations, one can extract the extinction time in different regimes. However, for the finite and structured populations, the usual Moran algorithm64,65 and its generalized versions24,66,67 may not be appropriate. One shall turn to the evolutionary theory on graphs21,68,69,70 to include the local mutual competitions. Following the same spirit, we introduce a local three-party Reference–Gamble–Birth (RGB) algorithm and perform the agent-based Monte Carlo simulations for the evolutionary dynamics. The RGB algorithm only requires local information in the stochastic processes and captures the network structure into account. And, it correctly reproduces the replicator dynamics when the populations are infinite and well-mixed.
Two major results are found in our numerical simulations. The first one is the universal scaling: the extinction time is always finite and falls onto the universal function when plotted with the rescaled variables. On the fully-connected network, two universal functions for stable and unstable regimes are found and the phase transition between them corresponds to neutral (critical) dynamics where the extinction time shows power-law behavior. The unexpected universal scaling may provide a powerful method to classify ecosystems with different dynamical behaviors into universal classes. The second major result is the notion of global payoff versus local payoff. On the two dimensional grid, the spatial structure changes the transition boundary between stable and unstable regimes but the universal function remains the same. This observation motivates us to introduce the notion of global payoff which turns out to be consistent with our numerical simulations. Our numerical simulations not only resolve the conflicting results in the previous studies33,34 but also suggest a potential tool to classify different evolutionary dynamics by universal scaling.
Methods
Rock-paper-scissors game
Here we consider the symmetric cyclic competitions between species A, B and C described by the payoff matrix
where the positive-definite parameter g denotes the gain in the cyclic competitions. The replicative dynamics of the ecosystem with populations \(N_A\), \(N_B\), \(N_C\) for the species A, B, C is captured by the following differential equations,
where \(a(t) \equiv N_A/N\) is the frequency (relative population) of species A in the total population \(N=N_A+N_B+N_C\). The other frequencies \(b(t) \equiv N_B/N\) and \(c(t) \equiv N_C/N\) are defined in similar fashion. The parameter \(\phi = (g-1)(ab+bc+ca)\) is the average fitness of the ecosystem.
The symmetric replicator equations11,34 host a non-trivial fixed point \((a^*, b^*, c^*) = (\frac{1}{3}, \frac{1}{3}, \frac{1}{3})\). If the gain is greater than unity, \(g>1\), the flows around the fixed point is stable. On the other hand, if the gain is less than unity, \(g<1\), the fixed point becomes unstable. At the critical gain \(g_c = 1\), it corresponds to the zero-sum game and the flow around the fixed point is neutral. Therefore, according to the replicator equations, the ecosystem goes through a phase transition from the stable regime to the unstable one when the gain g is tuned through the critical point \(g_c=1\).
RGB algorithm
The above scenario is well known for the rock-paper-scissors game when the population is infinite and well-mixed. But, what happens when the population is finite and the competitions among different species occur on the network with specific spatial structure? To investigate these important issues, we utilized the agent-based Monte Carlo simulations. To incorporate the locality of mutual competitions on the reasonable ground, we introduce a local three-party Reference–Gamble–Birth (RGB) algorithm. For each RGB update, it consists four steps as shown in Fig. 1. In the RGB algorithm, the third-party Reference provides local survival standard and determines the death probability of Gamble.
The death matrix D plays a central role in RGB algorithm. Similar to payoff matrix, its matrix elements \(D_{ij}\) defining the death probability for all possible competitions among species. The definition of death matrix indicates that the death matrix share the same matrix dimension with the payoff matrix and all its matrix elements must sit inside the range \(0\le D_{ij}\le 1\). In this section, we are going to show (i) the equivalent dynamics between replicator equations and the deterministic dynamics of stochastic updates in RGB algorithm in well-mixed and infinite populations; (ii) a simple scaling relation, i.e. \(D=P/s\), to construct the death matrix D for RGB algorithm from an arbitrary negative payoff matrix P; (iii) extracting the extinction time (i.e. a simple scaling \(T_{ex}=T_{RGB}/s\)) for a payoff matrix P by using the RGB algorithm.
In a single update, there are two possible ways to change the population size \(N_i\) of species i: (1) \(\Delta N_{i}=1\) with probability \(D_{kj}x_{i}x_{j}x_{k}\) coming from the process of the decease of a species-k Gamble when contacting with a Reference of species j and the random replacement of a Birth of species i, where \(k \ne i\). The death probability is \(D_{kj}\) in this case. (2) \(\Delta N_{i}=-1\) with probability \(D_{ij}x_{i}x_{j}x_{k}\) coming from the process that a species-i Gamble dies during the interaction between a species-j Reference and a Birth of species-k is randomly chosen to fill the empty site, where \(k \ne i\). The death probability is \(D_{ij}\) here. Weighting all possible competitions with the corresponding probabilities, the average change in populations after a single update takes the following form,
where \(\Delta \tau\) is the virtual time interval in the numerical simulation. However, within a realistic time interval \(\Delta t\), we expect the number of encountering events is proportional to the population size N24, i.e. \(\Delta \tau = N \Delta t\). Thus, the real time dynamics in the continuous limit when the population becomes infinite is
where \(x_{i}=N_{i}/N\) denotes the frequency of species i. Defining \(d_{i}=\sum _{j}D_{ij}x_{j}\) as the frequency-dependent death probability of species-i and \(\phi _{d}=\sum _{i}d_{i}x_{i}\) as the average death probability of the entire populations, the real time deterministic RGB dynamics becomes
This is the death version of replicator equation with \(b_{i}=0\) for all species. The RGB algorithm introduces selection among species by a death matrix, and therefore its deterministic dynamics is captured by the replicator equations without the birth terms.
Now, we can derive the relation between the death matrix in RGB algorithm and the payoff matrix in replicator equations when frequency-dependent selection is taken into account. The gauge redundancy of the payoff matrix indicates the replicator equations remain invariant when adding a arbitrary constant to all elements in the payoff matrix11. Based on the gauge redundancy, we consider a negative payoff matrix P with negative values for all matrix elements (i.e. \(P_{ij}\le 0\) for all i and j) without loss of generality. Because \(0\le D_{ij}\le 1\), a proper rescaling factor is required to establish the equivalence between the payoff matrix \(P_{ij}\) and the death matrix \(D_{ij}\). Introducing a proper rescaling factor s, which has a lower bound \(s_m = \text {max}(|P_{ij}|)\) defined by the payoff matrix, we define the death matrix from the negative payoff matrix as
In the case of frequency-dependent selection, \(f_{i}=\sum _{j}P_{ij}x_{j}\), we can rewrite the frequency-dependent death probability as \(d_{i}=-\sum _{j}P_{ij}x_{j}/s=-f_{i}/s\) and the average death probability of the entire populations as \(\phi _{d}=-\sum _{i}f_{i}x_{i}/s=-\phi /s\). Thus, the deterministic dynamics of RGB algorithm then is related to the replicator equations by the rescaling factor s,
As shown in the above equation, the RGB algorithm generates a rescaled population dynamics of replicator equations with the rescaled fitness \(f_{i}/s\). The rescaled fitness in RGB algorithm reduces the strength of competition and therefore slows down the dynamics of replicator equations by a factor of s. Thus, the extinction time obtained by the numerical simulations should also be rescaled by the same factor, i.e. \(T_{ex}=T_{RGB}/s\), accordingly to yield the extinction time of replicator equations.
The RGB algorithm has several advantages over the Moran process. First of all, only local information is involved in the stochastic process, more realistic than the need of global information for the Moran process. Secondly, the RGB algorithm reproduces the dynamics of frequency-dependent selection described by the replicator equations in infinite and well-mixed populations when the rescaling of time is restored properly. Finally, the the RGB algorithm takes the network structure into account, capturing the sensitive dependence of evolutionary dynamics in structured populations.
Results
Universal scaling in well-mixed populations
We perform agent-based Monte Carlo simulations to extract the mean extinction time. For each run, when one of the species goes extinct, we stop the simulated evolution and mark the time of extinction. The mean extinction time \(T_{ex}(N,g)\), depending on the population size N and the gain g defined in the payoff matrix, is computed by taking ensemble average of 10,000 runs in numerical simulations. We first focus on the well-mixed population, i.e. the fully connected network where every individual is connected to each other. It is rather remarkable that all numerical data can be collapsed onto the universal functions,
where \(\Delta g \equiv |g-g_c|\) denotes the deviation from the critical gain \(g_c=1\). As shown in Fig. 2, the numerical data for \(g>g_c\) (stable regime) and \(g<g_c\) (unstable regime) collapse onto the universal scaling functions \(F_{\pm }(x)\) respectively. The exponent \(\alpha =1\) can be extracted from extinction time of different population sizes at the critical point \(\Delta g =0\) and the other exponent \(\beta =1\) is the optimal fitting to collapse all data with different N and g. In the following paragraphs, we would like to show that these scaling variables can be understood within the framework of the renormalization group (RG)71. But, it is worth mentioning that the scaling variable \(x = N(g-1)\) agrees with the our expectation from population genetics72 that the extinction time in finite populations will scale as the product of population size and selection strength.
Let us derive the scaling form of the extinction time71 by RG analysis. Assuming the scaling dimensions for \(T_{ex}\), N and \(\Delta g\) are \(d_T\), \(d_N\) and \(d_g\) respectively,
Thus, the extinction time takes the following scaling form,
Choose the rescaling factor properly so that \(b^{d_N} N=N_0\), where \(N_0\) is some large number. The extinction time can be rewritten in terms of the rescaled variables,
Introducing two exponents \(\alpha \equiv d_T/d_N\) and \(\beta \equiv -d_g/d_N\), the above scaling relation can be cast in the elegant result in Eq. (10). Note that the scaling functions \(F_{\pm }(x)\) for stable and unstable regimes are different but the exponents \(\alpha\) and \(\beta\) are universal in the vicinity of the critical point.
The emergent universal scaling is highly non-trivial. For instance, extinction time of different population sizes at the critical point all collapse onto just one point (green dot in Fig. 2). Our numerical simulations show that the extinction time is always finite but the ecological stability can still be defined through the trend of extinction time. Plotted with the rescaled variables, stable and unstable regimes fall onto two universal functions \(F_{\pm }(x)\) and can be differentiated without ambiguity. In general, the exact forms of the scaling functions \(F_{\pm }(x)\) are difficult to obtain. However, its asymptotic form can be understood by the following arguments. Consider the stable regime first. In the asymptotic limit (\(|x|= N |\Delta g| \gg 1\)), the extinction of the ecosystem arises from a chain of successive rare event against the flow back to the stable equilibrium. In the stable regime, the clusters within the correlation length are locally stable with a small probability p to go extinct. Suppose the typical size of the cluster is \(N_c\) so that there are roughly \(N/N_c\) clusters. The probability for the whole ecosystem to go extinct is thus \(p^{N/N_c} = q^N\), where \(q = p^{1/N_c}\). In consequence, the extinction time grows exponentially with the population size N and the asymptotic form of \(F_{+}(x)\) should be
Here A is some positive constant and \(N_{+} = 1/(A \Delta g)\) is the characteristic constant for the exponential growth. In the unstable regime, the population decays exponentially, \(N(t)=Ne^{-\gamma t}\). The extinction time is usually estimated by the criterion \(N(T_{ex} ) \approx {{\mathcal {O}}}(1)\), leading to the logarithmic trend \(T_{ex} \sim \ln N\). The asymptotic form of the scaling function \(F_{-}(x)\) thus takes the form,
The asymptotic forms in Eqs. 16, 17 has been derived in the previous studies34 and agree with the numerical simulations presented in Fig. 2 rather well. Even though the asymptotic forms of the scaling functions can be understood rather well, the exact forms of the scaling functions \(F_{\pm }(x)\) are difficult to obtain.
Global payoff matrix in structured populations
Now we turn to the extinction dynamics on the 2D grid. As explained in the previous paragraphs, only local neighbors participate in each RGB updates. It is striking that the biodiversity is enhanced due to the spatial structure of the network: the extinction time always follows the exponential trend as shown in Fig. 3. Even in the extreme case \(g=0\), the stability of the cyclic-competing ecosystem is still protected by the grid structure. Is the extinction time described by the same scaling function as in the well-mixed case? The answer, after finding appropriate rescaled variable, turns out to be positive.
Even though the phase boundary between stable and unstable regimes are modified by the spatial structure of the network, the universal functions remain the same. We are thus inspired to introduce the notion of global payoff matrix as explained below. Local population dynamics generated by local interactions depends on the payoff matrix parametrized by the local gain g. But, the extinction time is a global property: even with the same local gain g, the 2D grid and the fully-connected network (well-mixed population) give different results. Because the fully-connected network looks the same either locally or globally, it is invariant under renormalization-group transformation and serves as a good reference. In RG analysis, the partition function is kept invariant to extract the renormalized coupling constants71. Here we adopt the same method and introduce the global gain G on an arbitrary network via the relation
where the superscript FC stands for “fully-connected”. That is to say, the extinction time for the local gain g on the considered network is set equal to that for the global gain G on the fully-connected network. According to the definition, \(g=G\) on the fully-connected network, reflecting the fact that local and global properties are the same in the well-mixed population. In our numerical simulations, the asymptotic limit is approximated by the largest population size in available simulations. To accurate the mapping, the characteristic constant in universal scaling function (\(F_{+}\)) is used to defined the global payoff matrix on the 2D grid. The mapping between G and g on the 2D grid is carried out numerically (shown in Fig. 4). The red line (\(G=g\)) represents the trivial relation for the fully-connected network. All data sit in the stable regime (blue background) and confirm that the dynamics on the 2D grid is always ecologically stable with \(G>1\). The mapping between the local and the global gains in the 2D grid reveals strong renormalization due to the spatial structure of the network. The global gain provides a convenient method to compare the effects of population structures on the extinction time.
The global gain G turns out to be the appropriate rescaled variable. When plotted with the global gain, the extinction time on the 2D grid collapse onto the same scaling function \(F_+(x)\) as shown in Fig. 5. Because the instability of the ecosystem is wiped out by the grid structure, the other branch \(F_-(x)\) does not show up. The reappearance of universal scaling indicates that the population structure only renormalizes the phase boundary but the underlying extinction behavior remains the same.
Spatial and temporal correlations on the grid
To explore the different dynamics arisen from the network structure, it is helpful to study the spatial and temporal correlation functions,
where the summation is over all species, \(X = A, B, C\). The random variable \(X(r,t)=1\) when the species X is spotted at the distance r and in the later time t. Otherwise, its value is zero. The ensemble average \(\langle X(r,t)X(0,0) \rangle\) measures the probability of finding the same species with the space-time separation (r, t) and equals \(\langle X(r,t) \rangle \langle X(0,0) \rangle\) if no correlation is present. The cross-species correlation functions are neglected here because they reflect the same trend as observed in C(r, t).
The spatial and temporal dependences of the correlation function is shown in Fig. 6. The spatial correlation is short-ranged with correlation length \(\xi \approx 3.3\) (lattice unit) for \(g=1\). The data fits the usual exponential decay rather well and implies no long-ranged pattern formation in the 2D grid. The temporal correlation is more subtle – it shows non-trivial oscillations \(\omega \approx 0.18\) with a relatively long correlation time \(\tau \approx 37.8\). There exists clear oscillatory signature before the temporal correlation damps out. This oscillation in the steady state is not predicted in the replicative dynamics. Because the initial configuration is randomly sampled, the population size of each species is the same at the beginning. According to the replicator equations, the initial condition sits right on the fixed point and the dynamics is trivially static – no oscillation at all. One may cautiously argue that, due to statistical fluctuations, the initial condition is not exactly on the fixed point. Straightforward analysis leads to an intrinsic oscillation frequency \(\omega _{0}=(g+1)/2\sqrt{3}\) near the fixed point. The sensitive dependence on the local gain g is not found in the numerics. Thus, the temporal oscillation must come from a different origin.
The situation is close to the nematic phase73 where the correlation in one direction (temporal) is stronger than the other (spatial). One can picture the whole system is roughly divided into many spiraling units of linear size \(\xi\). Each unit generates local spiral waves but never synchronizes into the global spiral due to interactions with the wavefronts of other neighboring units. Because the spiral unit is local, its oscillation cannot be captured by the usual replicative dynamics which works for uniformly mixed populations. This picture purveys a simple explanation why the biodiversity is greatly enhanced in the 2D grid. Suppose the probability for a spiral unit to go out is p. The interactions among the units ignite or suppress the neighboring ones. Thus, the probability for the complete extinction \(P_{ex}\) requires all units to go out at the same time, \(P_{ex} \approx p^{N/\xi ^2}\), where \(N/\xi ^2\) is roughly the number of spiraling units. The extinction is proportional to the reciprocal of \(P_{ex}\),
where \(N_+ = \xi ^2 / |\ln p|\). The exponential trend of the extinction time arises from the concurrent decease of the spiraling units. Although complete understanding of the evolutionary dynamics on the 2D grid is not yet achieved, the snap shots in Fig. 6 supports the qualitative picture proposed here.
Discussion
Two major results are found in the agent-based Monte Carlo simulations. The first one is the emergence of universal scaling. The second one is the renormalized global payoff due to spatial structure of the network. While our numerical results clearly support the above conclusions, a rigorous proof remain open at the point of writing. In fact, some other numerical study74 on the cooperation in social games also found that the rescaled extinction time follows universal scaling. However, their emphasis is in the neutral regime (critical phenomena) where the power law reigns. But, in the rock-paper-scissors game we studied here, we find the phase transition can be collapsed onto two universal curves when plotted in terms of the rescaled variables. This is quite unexpected and shows that the universal scaling works not just in the neutral regimes (or equivalently, the criticality). Thus, our conclusions are stronger and can serve as an important method to classify different dynamical trends for ecosystems with finite populations.
One can try to understand the extinction trends qualitatively by the dynamics of the biodiversity indicator \(\chi (t) \equiv a(t) b(t) c(t)\). When the population is infinite and well-mixed, its dynamics can be derived from the replicator equation,
where \(\Gamma (a,b,c)=1-3(ab+bc+ca) >0\) due to the constraint \(a+b+c=1\) for the positive-definite frequencies. It is clear that the drift flows toward extinction (\(\chi =0\)) for \(g<1\) while the ecosystem is stable for \(g>1\). At the critical value \(g=1\), the drift disappears and the flow is neutral.
Now consider the effect of finite population. At the critical point \(g=1\), the biodiversity indicator is no longer a constant of motion but suffers an \(\mathcal {O}(1/N)\) drift toward extinction34. The exponential trend for \(g>1\) can be understood as a chain of successive rare events against the flow while the logarithmic trend for \(g<1\) is the direct consequence of the exponential decay due to the drift toward extinction. Right at the critical point \(g=1\), the neutral drift implies the absence of any characteristic time scale and thus hints for the emergence of the universal scaling. Detail calculations for the fluctuation-induced drift leads to the extinction time \(T_{ex} \sim N\) and explains the exponent \(\alpha =1\). However, it remains an open question why the classification of the dynamics can be readily described by the scaling arguments developed for phase transitions in thermostatistics.
The 2D grid posts even more challenging puzzles. The success of applying global gain G to collapse all data onto the same scaling function is encouraging for further studies along the route of renormalization-group techniques. This shall provides an independent check on the relation between G and g, which is obtained numerically in our study here. The plausible coarse-graining in the time domain is further supported from the prolonged correlation time in the temporal correlation. The renormalization group may also provide more information about the exponents and the scaling functions.
Data availability
The datasets used and/or analysed during the current study available from the corresponding author on reasonable request.
References
May, R. Stability and Complexity in Model Ecosystems (Princeton University Press, 2001).
Maynard-Smith, J. Models in Ecology (Cambridge University Press, 1978).
Michod, R. Darwinian Dynamics: Evolutionary Transitions in Fitness and Individuality (Princeton University Press, 2000).
Neal, D. Introduction to Population Biology (Cambridge University Press, 2004).
Solé, R. & Bascompte, J. Self-organization in Complex Ecosystems (Princeton University Press, 2012).
Taylor, P. D. & Jonker, L. B. Evolutionary stable strategies and game dynamics. Math. Biosci. 40, 145–156 (1978).
Zeeman, E. C. Population dynamics from game theory. in Global Theory of Dynamical Systems, vol. 819, 471–497 (Springer, 1980).
Smith, J. Evolution and the Theory of Games (Cambridge University Press, 1982).
Weibull, J. W. Evolutionary Game Theory (MIT Press, 1997).
Hofbauer, J. & Sigmund, K. Evolutionary Games and Population Dynamics (Cambridge University Press, 1998).
Nowak, M. A. Evolutionary Dynamics (Harvard University Press, 2006).
Von Neumann, J. & Morgenstern, O. Theory of Games and Economic Behavior (Princeton University Press, 1947).
Smith, J. M. & Price, G. R. Logic of animal conflict. Nature 246, 15–18 (1973).
Foster, D. & Young, P. Stochastic evolutionary game dynamics. Theor. Popul. Biol. 38, 219–232 (1990).
Durrett, R. & Levin, S. The importance of being discrete (and spatial). Theor. Popul. Biol. 46, 363–394 (1994).
Cressman, R. & Vickers, G. T. Spatial and density effects in evolutionary game theory. J. Theor. Biol. 184, 359–369 (1997).
Cabrales, A. Stochastic replicator dynamics. Int. Econ. Rev. 41, 451–481 (2000).
Corradi, V. & Sarin, R. Continuous approximations of stochastic evolutionary game dynamics. J. Econ. Theor. 94, 163–191 (2000).
Taylor, C., Fudenberg, D., Sasaki, A. & Nowak, M. A. Evolutionary game dynamics in finite populations. Bull. Math. Biol. 66, 1621–1644 (2004).
Imhof, L. A. The long-run behavior of the stochastic replicator dynamics. Ann. Appl. Probab. 15, 1019–1045 (2005).
Lieberman, E., Hauert, C. & Nowak, M. A. Evolutionary dynamics on graphs. Nature 433, 312–316 (2005).
Traulsen, A., Pacheco, J. M. & Imhof, L. A. Stochasticity and evolutionary stability. Phys. Rev. E 74, 021905 (2006).
May, R. M. Network structure and the biology of populations. Trends Ecol. Evol. 21, 394–399 (2006).
Ohtsuki, H. & Nowak, M. A. The replicator equation on graphs. J. Theor. Biol. 243, 86–97 (2006).
Szabó, G. & Fáth, G. Evolutionary games on graphs. Phys. Rep. 446, 97–216 (2007).
Altrock, P. M. & Traulsen, A. Deterministic evolutionary game dynamics in finite populations. Phys. Rev. E 80, 011909 (2009).
Nathanson, C. G., Tarnita, C. E. & Nowak, M. A. Calculating evolutionary dynamics in structured populations. PLoS Comput. Biol. 5, e1000615 (2009).
Roca, C. P., Cuesta, J. A. & Sánchez, A. Evolutionary game theory: Temporal and spatial effects beyond replicator dynamics. Phys. Life Rev. 6, 208–249 (2009).
Tarnita, C. E., Antal, T., Ohtsuki, H. & Nowak, M. A. Evolutionary dynamics in set structured populations. Proc. Natl. Acad. Sci. U.S.A. 106, 8601–8604 (2009).
So, K. H. Z., Ohtsuki, H. & Kato, T. Spatial effect on stochastic dynamics of bistable evolutionary games. J. Stat. Mech. Theor. Exp. 2014, P10020 (2014).
Fu, F., Kocher, S. D. & Nowak, M. A. The risk-return trade-off between solitary and eusocial reproduction. Ecol. Lett. 18, 74–84 (2015).
Kerr, B., Riley, M., Feldman, M. & Bohannan, B. Local dispersal promotes biodiversity in a real-life game of rock-paper-scissors. Nature 418, 171–174 (2002).
Claussen, J. C. & Traulsen, A. Cyclic dominance and biodiversity in well-mixed populations. Phys. Rev. Lett. 100, 058104 (2008).
Frey, E. Evolutionary game theory: Theoretical concepts and applications to microbial communities. Physica A 389, 4265–4298 (2010).
Drossel, B. Biological evolution and statistical physics. Adv. Phys. 50, 209–295 (2001).
Frean, M. & Abraham, E. Rock-scissors-paper and the survival of the weakest. Proc. R. Soc. Lond. B 268, 1323–1327 (2001).
Czárán, T. L., Hoekstra, R. F. & Pagie, L. Chemical warfare between microbes promotes biodiversity. Proc. Natl. Acad. Sci. U.S.A. 99, 786–790 (2002).
Ifti, M. & Bergersen, B. Survival and extinction in cyclic and neutral three-species systems. Eur. Phys. J. E 10, 241–248 (2003).
Nowak, M. A. & Sigmund, K. Evolutionary dynamics of biological games. Science 303, 793–799 (2004).
West, S. A., Griffin, A. S., Gardner, A. & Diggle, S. P. Social evolution theory for microorganisms. Nat. Rev. Microbiol. 4, 597–607 (2006).
Reichenbach, T., Mobilia, M. & Frey, E. Coexistence versus extinction in the stochastic cyclic Lotka–Volterra model. Phys. Rev. E 74, 051907 (2006).
Reichenbach, T., Mobilia, M. & Frey, E. Noise and correlations in a spatial population model with cyclic competition. Phys. Rev. Lett. 99, 238105 (2007).
Reichenbach, T., Mobilia, M. & Frey, E. Mobility promotes and jeopardizes biodiversity in rock-paper-scissors games. Nature 448, 1046–1049 (2007).
Reichenbach, T., Mobilia, M. & Frey, E. Self-organization of mobile populations in cyclic competition. J. Theor. Biol. 254, 368–383 (2008).
Peltomäki, M. & Alava, M. Three- and four-state rock-paper-scissors games with diffusion. Phys. Rev. E 78, 031906 (2008).
Claussen, J. C. & Traulsen, A. Cyclic dominance and biodiversity in well-mixed populations. Phys. Rev. Lett. 100, 058104 (2008).
He, Q., Mobilia, M. & Täuber, U. C. Spatial rock-paper-scissors models with inhomogeneous reaction rates. Phys. Rev. E 82, 051909 (2010).
Shi, H., Wang, W.-X., Yang, R. & Lai, Y.-C. Basins of attraction for species extinction and coexistence in spatial rock-paper-scissors games. Phys. Rev. E 81, 030901 (2010).
Venkat, S. & Pleimling, M. Mobility and asymmetry effects in one-dimensional rock-paper-scissors games. Phys. Rev. E 81, 021917 (2010).
Jiang, L.-L., Zhou, T., Perc, M. & Wang, B.-H. Effects of competition on pattern formation in the rock-paper-scissors game. Phys. Rev. E 84, 021912 (2011).
Adamson, M. W. & Morozov, A. Y. Revising the role of species mobility in maintaining biodiversity in communities with cyclic competition. Bull. Math. Biol. 74, 2004–2031 (2012).
Lamouroux, D., Eule, S., Geisel, T. & Nagler, J. Discriminating the effects of spatial extent and population size in cyclic competition among species. Phys. Rev. E 86, 021911 (2012).
Rulands, S., Zielinski, A. & Frey, E. Global attractors and extinction dynamics of cyclically competing species. Phys. Rev. E 87, 052710 (2013).
Schreiber, S. J. & Killingback, T. P. Spatial heterogeneity promotes coexistence of rock-paper-scissors metacommunities. Theor. Popul. Biol. 86, 1–11 (2013).
Laird, R. A. Population interaction structure and the coexistence of bacterial strains playing ‘rock-paper-scissors’. Oikos 123, 472–480 (2014).
Szczesny, B., Mobilia, M. & Rucklidge, A. M. Characterization of spiraling patterns in spatial rock-paper-scissors games. Phys. Rev. E 90, 032704 (2014).
Jackson, J. B. C. & Buss, L. Alleopathy and spatial competition among coral reef invertebrates. Proc. Natl. Acad. Sci. U.S.A. 72, 5160–5163 (1975).
Sinervo, B. & Lively, C. The rock-paper-scissors game and the evolution of alternative male strategies. Nature 380, 240–243 (1996).
Kirkup, B. & Riley, M. Antibiotic-mediated antagonism leads to a bacterial game of rock-paper-scissors in vivo. Nature 428, 412–414 (2004).
Weber, M. F., Poxleitner, G., Hebisch, E., Frey, E. & Opitz, M. Chemical warfare and survival strategies in bacterial range expansions. J. R. Soc. Interface 11, 20140172 (2014).
Gilg, O., Hanski, I. & Sittler, B. Cyclic dynamics in a simple vertebrate predator-prey community. Science 302, 866–868 (2003).
Lankau, R. A. & Strauss, S. Y. Mutual feedbacks maintain both genetic and species diversity in a plant community. Science 317, 1561–1563 (2007).
Cameron, D. D., White, A. & Antonovics, J. Parasite-grass-forb interactions and rock-paper-scissor dynamics: Predicting the effects of the parasitic plant Rhinanthus minor on host plant communities. J. Ecol. 97, 1311–1319 (2009).
Moran, P. A. P. Random processes in genetics. Math. Proc. Camb. Phil. Soc. 54, 60–71 (1958).
Moran, P. The Statistical Processes of Evolutionary Theory (Clarendon Press, 1962).
Hofbauer, J. & Sigmund, K. Evolutionary game dynamics. Bull. Am. Math. Soc. 40, 479–519 (2003).
Traulsen, A., Pacheco, J. M. & Nowak, M. A. Pairwise comparison and selection temperature in evolutionary game dynamics. J. Theor. Biol. 246, 522–529 (2007).
Ohtsuki, H. & Nowak, M. A. Evolutionary stability on graphs. J. Theor. Biol. 251, 698–707 (2008).
Hindersin, L. & Traulsen, A. Most undirected random graphs are amplifiers of selection for birth-death dynamics, but suppressors of selection for death-birth dynamics. PLoS Comput. Biol. 11, e1004437 (2015).
Allen, B. et al. Evolutionary dynamics on any population structure. Nature 544, 227–230 (2017).
Nishimori, H. & Ortiz, G. Elements of Phase Transitions and Critical Phenomena (Oxford University Press, 2011).
Ewens, W. J. Mathematical Population Genetics (Springer, 2000).
Chaikin, P. M. & Lubensky, T. C. Principles of Condensed Matter Physics (Cambridge University Press, Cambridge, 2000).
Cremer, J., Reichenbach, T. & Frey, E. The edge of neutral evolution in social dilemmas. New J. Phys. 11, 093029 (2009).
Acknowledgements
We acknowledge support from Ministry of Science and Technology in Taiwan through Grant No. MOST 109-2112-M-007-026-MY3
Author information
Authors and Affiliations
Contributions
H.H.L. and C.C.C. propose the research topic and supervise the project. C.I.H. perform the numerical simulations. C.I.H., H.H.L. and C.C.C. discuss the results and prepare the manuscript together.
Corresponding authors
Ethics declarations
Competing interests
The authors declare no competing interests.
Additional information
Publisher's note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Rights and permissions
Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article's Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article's Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by/4.0/.
About this article
Cite this article
Huang, CI., Chen, CC. & Lin, HH. Universal scaling of extinction time in stochastic evolutionary dynamics. Sci Rep 12, 22403 (2022). https://doi.org/10.1038/s41598-022-27102-0
Received:
Accepted:
Published:
DOI: https://doi.org/10.1038/s41598-022-27102-0
- Springer Nature Limited