Advertisement
Browse Subject Areas
?

Click through the PLOS taxonomy to find articles in your field.

For more information about PLOS Subject Areas, click here.

  • Loading metrics

Effect of Spatial Dispersion on Evolutionary Stability: A Two-Phenotype and Two-Patch Model

  • Qing Li,

    Affiliation School of Mathematical Sciences, Capital Normal University, Beijing, China

  • Jiahua Zhang,

    Affiliation Key Lab of Animal Ecology and Conservational Biology, Institute of Zoology, Chinese Academy of Sciences, Beijing, China

  • Boyu Zhang ,

    zhangby@bnu.edu.cn

    Affiliation Laboratory of Mathematics and Complex Systems, Ministry of Education, School of Mathematical Sciences, Beijing Normal University, Beijing, China

  • Ross Cressman,

    Affiliation Department of Mathematics, Wilfrid Laurier University, Waterloo, ON, Canada

  • Yi Tao

    Affiliation Key Lab of Animal Ecology and Conservational Biology, Institute of Zoology, Chinese Academy of Sciences, Beijing, China

Effect of Spatial Dispersion on Evolutionary Stability: A Two-Phenotype and Two-Patch Model

  • Qing Li, 
  • Jiahua Zhang, 
  • Boyu Zhang, 
  • Ross Cressman, 
  • Yi Tao
PLOS
x

Abstract

In this paper, we investigate a simple two-phenotype and two-patch model that incorporates both spatial dispersion and density effects in the evolutionary game dynamics. The migration rates from one patch to another are considered to be patch-dependent but independent of individual’s phenotype. Our main goal is to reveal the dynamical properties of the evolutionary game in a heterogeneous patchy environment. By analyzing the equilibria and their stabilities, we find that the dynamical behavior of the evolutionary game dynamics could be very complicated. Numerical analysis shows that the simple model can have twelve equilibria where four of them are stable. This implies that spatial dispersion can significantly complicate the evolutionary game, and the evolutionary outcome in a patchy environment should depend sensitively on the initial state of the patches.

Introduction

In order to explain the evolution of animal behavior, Maynard Smith and Price [1] developed the concept of evolutionarily stable strategy (ESS) (see also [25]). Prior et al. [6] investigated an evolutionary game model that incorporates both spatial dispersion and density effects in the evolutionary dynamics. In this model, the population is considered to be dispersed in a patchy environment, where the background fitness and payoff matrix in each patch can be different. Migration from region to region is considered as an incidental aspect of the population, i.e., the migration is a chance event unrelated to an individual’s phenotype (strategy) or the fitness of the patch. As pointed out by Prior et al. [6], their assumptions differ from that of Ludwig and Levin [7] who treat the tendency to migrate as an individual characteristic subject to selection (see also [813]), and also differ from that of Hines and Maynard Smith [14] who interpret the effect of spatial dispersion as an increased tendency to interact with opponents sharing one’s own characteristics (see also [15]). Recently, Cressman and Krivan [16] investigated the migration dynamics for the ideal free distribution (IFD) in a patchy environment. They showed that IFD is evolutionarily stable under the assumptions that individuals never migrate from patches with a higher payoff to patches with a lower payoff and some individuals always migrate to the best patch. But migration does not necessarily lead to IFD if migration rates are independent of the payoffs of the patches.

For the evolutionary game dynamics in a patchy environment, Prior et al. [6] mainly focused their analysis on the stability of the homogeneous states, where they assumed that all patches have the same payoff matrix and density-dependent background fitness. Their main results showed that a stable equilibrium (e.g. an evolutionarily stable strategy) of the non-dispersed frequency dynamics becomes a stable equilibrium of the large system if population density stabilizes at these fixed frequencies.

In this paper, following Prior et al. [6], a simple two-patch and two-phenotype model is investigated. Three basic assumptions for this model are:

  1. (i). The environment consists of two patches, called patch 1 and patch 2, respectively. Individuals can move from one patch to the other at any time t. The migration rates are patch-dependent but independent of individual’s phenotype [6]. Let c1 denote the probability that an individual moves from patch 1 to patch 2, and, similarly, c2 the probability that an individual moves from patch 2 to patch 1.
  2. (ii). In each of two patches, individuals display two possible phenotypes (strategies), denoted by R1 and R2, and individuals interact in random pairwise contests. The payoff matrix is in patch 1 and in patch 2, where aij (or bij) is the payoff of an individual displaying phenotype Ri when it plays against an individual displaying phenotype Rj in patch 1 (or in patch 2) for all i, j = 1, 2. Without loss of generality, we also assume that aij ≥ 0 and bij ≥ 0 for all i, j = 1, 2.
  3. (iii). In each of the two patches, the background fitness is density-dependent [34], which is defined as α1β1 n1 in patch 1, and α2β2 n2 in patch 2, where n1 is the total population size in patch 1 and n2 the total population size in patch 2. We also assume that αi > ci for all i = 1, 2. That is, the migration rates are small enough to ensure that population size in a patch will increase when there are few individuals in the patch.

Let xi denote the number of individuals with phenotype Ri in patch 1, and yi the number of individuals with phenotype Ri in patch 2 (i = 1, 2). Clearly, n1 = x1 + x2 and n2 = y1 + y2. Similarly, let p denote the frequency of phenotype R1 in patch 1, and q the frequency of phenotype R1 in patch 2, i.e., p = x1/n1 and q = y1/n2. According to the basic assumption (ii), the expected payoff of an individual displaying phenotype Ri is fi = pai1 + (1 − p)ai2 in patch 1, and gi = qbi1 + (1 − q)bi2 in patch 2 for i = 1, 2. Similarly, according to the basic assumption (iii), the (total) fitness of an individual displaying phenotype Ri is defined as Fi = (α1β1 n1) + fi in patch 1, and Gi = (α2β2 n2) + gi in patch 2 for i = 1, 2. Thus, the time evolution of xi and yi can be given by (1) respectively, for i = 1, 2. Dynamics (1) also equivalent to the following system expressed in terms of phenotypic frequency and population size in each patch. (2) where and are the average payoffs in patch 1 and patch 2, respectively.

In this paper, the equilibria of dynamics (2) and their stabilities are analyzed. Different from Prior et al. [6] who focused on the homogeneous states, we are primarily interested in analyzing the heterogeneous states, where two patches have different ESSs. Our main goal is to reveal the dynamical properties of the evolutionary game in a heterogeneous patchy environment.

Results

Symmetric equilibria of dynamics (2)

For given p and q with 0 ≤ p, q ≤ 1, an equilibrium of dynamics (3) denoted by (n1(p, q), n2(p, q)), satisfies (4) It is clear that (n1, n2) = (0, 0) is always a solution of Eq (4) for any given p and q. Furthermore, (p, q, 0, 0) must be unstable under dynamics (2) since α1 > c1 and α2 > c2. Notice that n2 is a parabolic function of n1 and vice versa, Eq (4) also has a unique positive solution, denoted by with and (see Fig 1). When this solution corresponds to an equilibrium of dynamics (2), we call it a positive equilibrium. In the rest of this paper, we only focus on the number and stabilities of these positive equilibria.

thumbnail
Fig 1. The unique positive equilibrium of dynamics (3).

The red curves correspond to the second equation of Eq (4) and the green curves to the first equation of Eq (4). For any given p and q, the two curves have a unique positive intersection (see the black spots). Parameters are taken as β1 = β2 = 0.01, α1 = 0.75, , c1 = 0.5 and c2 = 0.5 in all two panels. Furthermore, α2 = 0.55 and in panel a, and α2 = 1 and in panel b.

https://doi.org/10.1371/journal.pone.0142929.g001

A positive equilibrium, , is called a symmetric equilibrium of dynamics (2) if p = q. That is, at a symmetric equilibrium, population compositions in the two patches are the same.

It is easy to see that dynamics (2) always have two symmetric boundary equilibria, and , where at these equilibria, all individuals in the two patches display the same phenotype. Stabilities of the boundary equilibria can be characterized by analyzing the Jacobian matrix of dynamics (2) (see Method section). The main result is that if R1 (or R2) is an ESS for both payoff matrices A and B, then the boundary equilibrium (or ) must be asymptotically stable. This result is consistent with that of Prior et al. [6], where the stable equilibrium of the non-dispersed system becomes a stable equilibrium of dynamics (2). However, under the influence of migration, a symmetric boundary equilibria could be stable even if the corresponding phenotype is not an ESS in either patch. For instance, the boundary equilibrium is asymptotically stable if a11 < a21 as long as a21a11 is small enough.

On the other hand, the symmetric interior equilibrium exists only for a very special case p* = q* ∈ (0, 1), where and . In this case, is an symmetric interior equilibrium of dynamics (2), where is the solution of Eq (4) for p = p* and q = q*. Furthermore, it is asymptotically stable if p* (= q*) is an ESS for both A and B (see Method section).

General cases

For more general situations (i.e., p* ≠ q*), it is very tedious to determine the equilibria of dynamics (2). In fact, numerical simulations show that dynamics (2) may have twelve equilibria. To investigate the properties of the equilibria of dynamics (2), two cases are considered below. The first case is special in that there is no migration in one direction (i.e., one of c1 and c2 is 0) and we analyze the number of stable equilibria for all possible payoff structures. The second case is more general (i.e., c1 > 0 and c2 > 0) and we show the equilibria of dynamics (2) and their stabilities for 0 < p*, q* < 1.

Case 1. c1 > 0 and c2 = 0. Without loss of generality, we here assume that c1 > 0 but c2 = 0, i.e. individuals can only move from patch 1 to patch 2 but not from patch 2 to patch 1 (the case of c1 = 0 and c2 > 0 can be analyzed analogously). Then, dynamics (2) can be rewritten as (5) and (6)

Notice that dynamics (5) is independent of dynamics (6). Thus, as an equilibrium of dynamics (2), , is locally asymptotically stable if and only if is locally asymptotically stable under dynamics (5) and is locally asymptotically stable under dynamics (6), where p and n1 in dynamics (6) correspond to the stable equilibrium, , of dynamics (5).

We first look at the stability of dynamics (5). It is easy to see that: (i) the boundary equilibrium (or ) is locally asymptotically stable if and only if p = 1 (or p = 0) is an ESS for the payoff matrix A, i.e. a11 > a12 (or a22 > a12), where for p = 1 and for p = 0; and (ii) if the unique interior equilibrium exists, then it is globally asymptotically stable if and only if p* is an ESS for the payoff matrix A, where and .

From dynamics (6), the frequency in a stable equilibrium of dynamics (2), , should obey the equation (7) where corresponds to the stable equilibrium, , of dynamics (5). In the Method section, we analyze the solutions of Eq (7) and the stabilities of the corresponding equilibria under dynamics (6). According to the stability conditions of dynamics (5) and (6), the equilibria of dynamics (2) and their properties can be summarized as follows:

  1. If R1 (or R2) is the only ESS for A and B, then the symmetric boundary equilibrium (or ) is unstable and the other (or ) is stable. Furthermore, Eq (7) has no interior solution. This implies that (or ) is also globally asymptotically stable, i.e., all individuals in the two patches will eventually display R1 (or R2) under evolutionary dynamics (2) (see Fig 2A and 2B).
  2. If both R1 and R2 are ESSs for A but R1 (or R2) is the only ESS for B, then the symmetric boundary equilibrium (or ) is unstable and the other (or ) is stable. Furthermore, Eq (7) has a unique (interior) solution , which corresponds to an asymptotically stable equilibrium (or ) of dynamics (2). In this situation, either all individuals in the system display R1 (or R2), or individuals in patch 1 display R2 (or R1) and two phenotypes coexist in patch 2 (see Fig 2C and 2D).
  3. If p* is the only ESS for A (0 < p* < 1) and R1 (or R2) is the only ESS for B, then both the symmetric boundary equilibria and are unstable. Furthermore, Eq (7) has a unique (interior) solution , which corresponds to an asymptotically stable equilibrium of dynamics (2). Numerical simulation shows that this equilibrium is also globally stable. This implies that the two phenotypes will stably coexist in the system (see Fig 2E and 2F).
  4. If R2 (or R1) is the only ESS for A but R1 (or R2) is the only ESS for B, then both the symmetric boundary equilibria and are unstable. Furthermore, Eq (7) has a unique (interior) solution , which corresponds to an asymptotically stable equilibrium (or ) of dynamics (2). Similarly as (3), this equilibrium is also globally stable, i.e., R2 (or R1) can invade patchy 2 under the influence of migration (see Fig 2G and 2H).
  5. If both R1 and R2 are ESSs for A and B, then both the symmetric boundary equilibria and are asymptotically stable. Furthermore, Eq (7) has at most four (interior) solutions, where two are stable equilibria of dynamics (2) and the other two are unstable. This implies that the evolutionary outcome in this situation is very difficult to predict since the system can have four stable states (see Fig 2I).
  6. If both R1 and R2 are ESSs for A and q* is an ESS for B, then both the symmetric boundary equilibria and are unstable. Furthermore, Eq (7) has at most two (interior) solutions, where both of them are stable equilibria of dynamics (2) (see Fig 2J).
  7. If p* is an ESS for A and q* is an ESS for B, then both the symmetric boundary equilibria and are unstable. Furthermore, Eq (7) has a unique (interior) solution , which corresponds to an asymptotically stable equilibrium of dynamics (2). Similarly as (3), this equilibrium is also globally stable and two phenotypes will stably coexist in the system (see Fig 2K).
  8. If p* is an ESS for A and both R1 and R2 are ESSs for B, then both the symmetric boundary equilibria and are unstable. Furthermore, Eq (7) has at most three (interior) solutions, which are denoted by , and with . The two interior equilibria corresponding to and are stable under dynamics (2) and the interior equilibrium corresponding to is unstable (see Fig 2L).
  9. If R2 (or R1) is the only ESS for A and q* is an ESS for B, then both the symmetric boundary equilibria and are unstable. Furthermore, Eq (7) has a unique (interior) solution , which corresponds to an asymptotically stable equilibrium (or ) of dynamics (2). Similarly as (4), this equilibrium is also globally stable and two phenotypes will stably coexist in patch 2 (see Fig 2M and 2N).
  10. If R2 (or R1) is the only ESS for A and both R1 and R2 are ESSs for B, then the symmetric boundary equilibrium (or ) is stable and the other (or ) is unstable. Furthermore, Eq (7) has at most two (interior) solutions, where one corresponds to a stable equilibria of dynamics (2) and the other is unstable. Similarly as (2), either all individuals in the system display R2 (or R1), or individuals in patch 1 display R2 (or R1) and two phenotypes coexist in patch 2 (see Fig 2O and 2P).

thumbnail
Fig 2. Solutions of Eq (7) and their stabilities under dynamics (2) when c1 > 0 and c2 = 0.

Intersections of h1(q) and h2(q) (i.e., the solutions of Eq (7)) are shown for all sixteen possible situations. The red curves denote h1(q) and the green curves h2(q). The intersections denoted by black spots correspond to stable equilibria of dynamics (2), and the intersections denoted by black circles correspond to unstable equilibria. Parameters are taken as β1 = β2 = 0.01, α1 = 1, α2 = 2 and c1 = 0.5 in all sixteen panels. Payoff matrixes in the panels are: In panel a, A = [1.5, 1.5; 0, 0], B = [5, 1; 0, 0]. In panel b, A = [0, 0; 1, 1], B = [0, 0; 5, 1]. In panel c, A = [1, 0; 0, 1], B = [3, 1; 0, 0]. In panel d, A = [1, 0; 0, 1], B = [0, 0; 3, 1]. In panel e, A = [0, 1.25; 5, 0], B = [3, 1; 0, 0]. In panel f, A = [0, 1.25; 5, 0], B = [0, 0; 3, 1]. In panel g, A = [0, 0; 1, 1], B = [3, 1; 0, 0]. In panel h, A = [1, 1; 0, 0], B = [0, 0; 3, 1]. In panel i, A = [1, 0; 0, 1], B = [2, 0; 0, 1]. In panel j, A = [1, 0; 0, 1], B = [0, 1; 4, 0]. In panel k, A = [0, 2; 2, 0], B = [0, 1; 5, 0]. In panel l, A = [0, 2; 2, 0], B = [1, 0; 0, 1]. In panel m, A = [0, 0; 1, 1], B = [0, 1; 4, 0]. In panel n, A = [1.5, 1.5; 0, 0], B = [0, 1; 4, 0]. In panel o, A = [0, 0; 1, 1], B = [2, 0; 0, 1]. In panel p, A = [1, 1; 0, 0], B = [2, 0; 0, 1]. The positions of the interior ESSs p* and q* (0 < p*, q* < 1) are marked by dashed lines.

https://doi.org/10.1371/journal.pone.0142929.g002

Case 2. c1 > 0 and c2 > 0. We now consider the case with c1 > 0 and c2 > 0. It is easy to check that dynamics (2) only have two boundary equilibria, and , and the existence of asymmetric boundary equilibrium is impossible, for instance, if p = 0 and q ≠ 0, then . We then focus on the number and stability of interior equilibria. Notice that an equilibrium of dynamics (2) should be the solution of equation (8) So if both n1 and n2 are positive, then from the first two equations of Eq (8), an interior equilibrium of dynamics (2) should obey the equations (9) Furthermore, from the third and the forth equations of Eq (8) (10) where we assume that both p* and q* are in the interval 0 < p*, q* < 1 (i.e., we consider the most complicated payoff structures).

From Eq (9), it is easy to see that for the situation with p* ≠ q*, if the interior equilibrium exists, then it should be in the region (0, p*) × (q*, 1), or (p*, 1) × (0, q*) if Δ1Δ2 > 0, and in the region (0, p*) × (0, q*), or (p*, 1) × (q*, 1) if Δ1Δ2 < 0. Of course, it is very difficult to get the exactly analytic solutions of Eqs (9) and (10) in general. The numerical analysis suggests that ten interior equilibria can exist (see Fig 3H). To show this, some examples are plotted in Fig 3. All of these examples show clearly that the equilibrium structure of dynamics (2) could be very complicated.

thumbnail
Fig 3. Equilibria of Eq (2) on the pq plane and their stabilities when c1 > 0 and c2 > 0.

The red curves correspond to Eq (10) and the green curves Eq (9). The intersections denoted by black spot correspond to stable equilibria of dynamics (2), and the intersections denoted by black circle correspond to unstable equilibria. Parameters are taken as: In panel a, A = [0, 1; 1, 0], B = [0, 5; 5, 0], β1 = β2 = 0.01, α1 = 1.25, α2 = 1.8, c1 = 0.25 and c2 = 0.8. In panel b, A = [0, 1; 1, 0], B = [0, 5; 5, 0], β1 = β2 = 0.01, α1 = 1.125, α2 = 0.1, c1 = 0.125 and c2 = 0.8. In panel c, A = [1, 0; 0, 1], B = [5, 0; 0, 5], β1 = β2 = 0.001, α1 = 1.4, α2 = 1.5, c1 = 0.4 and c2 = 0.5. In panel d, A = [0, 1; 1, 0], B = [5, 0; 0, 5] β1 = β2 = 0.001, α1 = 1.5, α2 = 1.2, c1 = 0.5 and c2 = 0.2. In panel e, A = [0, 5; 1, 0], B = [1, 0; 0, 5], β1 = β2 = 0.001, α1 = 1.5, α2 = 1.2, c1 = 0.5 and c2 = 0.2. In panel f, A = [0, 5; 1, 0], B = [0, 1; 5, 0], β1 = β2 = 0.001, α1 = 1.5, α2 = 1.2, c1 = 0.5 and c2 = 0.2. In panel g, A = [1, 0; 0, 5], B = [5, 0; 0, 1], β1 = β2 = 0.001, α1 = 1.4, α2 = 1.5, c1 = 0.4 and c2 = 0.5. In panel h, A = [1, 0; 0, 1], B = [2, 0; 0, 1], β1 = 0.05, β2 = 0.01, α1 = 1.75, α2 = 2, c1 = 0.25 and c2 = 0.01. The positions of p* and q* are marked by dashed lines (0 < p*, q* < 1, and note that p* and q* may not be ESS).

https://doi.org/10.1371/journal.pone.0142929.g003

We further look at the bifurcation behaviors of system (2) for the case that both R1 and R2 are ESSs for A and B (i.e., the most complicated case), and assume equal migration rates between regions, i.e., c1 = c2 = c. In this case, both the symmetric boundary equilibria and are asymptotically stable, and the system can have ten interior equilibria. When c = 0, it is easy to see that the system has nine equilibria in total, including four stable (boundary) equilibria and five unstable equilibria. The number of equilibria jumps from nine to twelve as soon as the migration rates becomes positive although the number of stable equilibria keeps unchanged (see Fig 4A and 4B). Furthermore, numerical simulation shows that both the numbers of stable equilibria and unstable equilibria decrease as c increases. In particular, when c > 0.019, the system has only two stable equilibria (i.e., the two symmetric boundary equilibria), and all interior equilibria are unstable (see Fig 4). These results suggest that small migration rates make the dynamical behavior of the system more complex.

thumbnail
Fig 4. Bifurcation behaviors of Eq (2) when c1 > 0 and c2 > 0.

In panel a, the number of equilibria and the number of stable equilibria are denoted by blue line and yellow line, respectively. In panels b-e, the red curves correspond to Eq (10) and the green curves Eq (9). The intersections denoted by black spot correspond to stable equilibria of dynamics (2), and the intersections denoted by black circle correspond to unstable equilibria. The positions of p* and q* are marked by dashed lines. Parameters are taken as: A = [1, 0; 0, 1], B = [2, 0; 0, 1], β1 = 0.05, β2 = 0.01, α1 = 1.75, α2 = 2. In panel b, c1 = c2 = c = 0.014; in panel c, c1 = c2 = c = 0.015; in panel d, c1 = c2 = c = 0.02; and in panel e, c1 = c2 = c = 0.05.

https://doi.org/10.1371/journal.pone.0142929.g004

Discussion

A vast amount of research has been devoted to analyze the influence of spatial diffusion on the evolutionary stability of ecology systems. One well known mathematical approach is the reaction-diffusion equation, where in this framework, individuals are dispersed in a continues space [1011, 1721]. For instance, Hofbauer et al. [11, 21] considered a population of two types of individuals distributed in an one-dimensional space, and assumed that the migration (or diffusion) rate is both individual-independent and location-independent. They showed that in two-strategy coordination games, if the reaction term of the reaction-diffusion equation is taken as replicator dynamics, then one strategy will drive out the other strategy in form of a traveling wave front, although there is no simple rule to decide which strategy can survive.

In this paper, we assume that individuals are distributed in a (discrete) patchy environment. Following Prior et al. [6], we investigate a simple two-phenotype and two-patch model, where individuals compete only with their immediate neighbors and the migration rates between patches are individual-independent but patch-dependent. Different from Prior et al. [6] who focused on the homogeneous patchy environment, we here are more interested in the dynamical stability in heterogeneous environment. Our main results show that: (i) if the pure strategy R1 (or R2) is an ESS for both two patches, then the boundary equilibrium corresponding to p = 1 and q = 1 (or p = 0 and q = 0) must be asymptotically stable; (ii) if the payoff matrices A and B satisfy , then the interior equilibrium corresponding to (p*, q*) is asymptotically stable if p* is an ESS for A and q* an ESS for B; (iii) as a special case with c1 > 0 and c2 = 0 (or c1 = 0 and c2 > 0), i.e. individuals can only move from patch 1 to patch 2 (or from patch 2 to patch 1), all possible situations for the existence and stability of boundary and interior equilibria are considered, and we find that dynamics (2) can have six equilibria where four of them are stable; (iv) for c1 > 0 and c2 > 0, the numerical analysis shows that the equilibrium structure and dynamical behavior of the system could be very complicated in general. In particular, dynamics (2) can have twelve equilibria where four of them are stable.

Our analysis provides an insight for understanding the effect of spatial dispersion on the evolutionary stability of patchy environment. Both the analytical analysis and the numerical simulation indicate that the original ESS formulations which ignore the dispersion process cannot be applied to predict the evolutionary outcome of the dispersion system even for small migration rates. For instance, in the case that both patches have multiple ESS’s and no dispersal between patches, the system has four (boundary) stable equilibria and five unstable equilibria. However, if one of c1 and c2 becomes positive, the system can have two to four stable equilibria and four to eight unstable equilibria. Furthermore, we found that both the numbers of stable equilibria and unstable equilibria decrease in the migration rates. This observation has an intuitive biological interpretation [6]. In a heterogenous patchy environment, the effect of selection is to make the overall population more heterogeneous in the sense of different patches have different population compositions, while the effect of migration is to move the population composition in each patch towards the mean of the overall population, i.e., migration promotes homogeneity. Thus, when the migration rates are small (i.e., the effect of selection is strong), similarly as the case of no dispersal, the system has two stable symmetric boundary equilibria and two stable asymmetric equilibria; and when the migration rates are large (i.e., the effect of migration is strong), the existence of stable asymmetric equilibrium is impossible, and the system has only two stable symmetric boundary equilibria, where at these equilibria all individuals display the same phenotype.

In this paper, we focus on the effect of spatial dispersion on two-patch system only. A natural extension would be to consider the three-patch system. However, analyzing the dynamical behavior of the three-patch system may be an even more difficult issue because the equilibrium structure of the two-patch system is already very complex. Another possible development would be to compare the evolutionary stability of the patchy environment under different migration rules. One commonly used migration rule is that individuals know perfectly the payoff in all patches and they always move to the patch with the highest payoff (i.e., ideal animals) [15]. In contrast, a more realistic model is that individuals do not migrate to patches with lower payoff [22]. Recent studies have shown that these two migration rules can lead to the IFD [16, 23]. Since that the IFD corresponds to a stable equilibrium of the non-dispersed evolutionary dynamics, we can then expect that these migration rules may also lead to the ESS of the non-dispersed evolutionary dynamics [23].

Methods

Stability of the symmetric equilibria

The Jacobian matrix of the dynamics (2) about the symmetric boundary equilibrium , denoted by J(1,1), is and similarly, the Jacobian matrix about , denoted by J(0,0), is For the matrix J(1,1), notice that the eigenvalues of the matrix have negative real parts if a11a21 > 0 and b11b21 > 0, and that the real parts of the eigenvalues of the matrix must be negative. So, if the pure strategy R1 is an ESS for both payoff matrices A and B, then the eigenvalues of J(1,1) must have negative real parts [6]. Similar to the matrix J(0,0), if the pure strategy R2 is an ESS for both payoff matrices A and B, then the eigenvalues of J(0,0) have negative real parts.

The Jacobian matrix about the symmetric interior equilibrium , denoted by J(p*, q*), is where Δ1 = a11a12a21 + a22 and Δ2 = b11b12b21 + b22. Also similar to the matrix J(1,1) (or the matrix J(0,0)), the eigenvalues of J(p*, q*) have the negative real parts if p* (= q*) is an ESS for both payoff matrices A and B, i.e., the equilibrium is asymptotically stable if p* (= q*) is an ESS for both A and B.

Stability analysis of dynamics (6) when c1 > 0 and c2 = 0

We first analyze the solutions of Eq (7). For convenience, let It is easy to see that Eq (7) has a boundary solution (or q = 1) if and only if (or ). Furthermore, the interior solutions of Eq (7) should correspond to the intersections of the functions h1(q) and h2(q) in the interval 0 < q < 1. Notice that h1(0) = h1(1) = h1(q*) = 0, Δ2 h1(q) > 0 for 0 < q < q* and Δ2 h1(q) < 0 for q* < q < 1 (if 0 < q* < 1), and that h2(0) ≥ 0, h2(1) ≤ 0, , h2(q) > 0 for and h2(q) < 0 for (if ). Then, for the existence of intersections in the interval 0 < q < 1, we have that:

  1. If R1 is the only ESS for both payoff matrices A and B, then, no intersection can exist (see Fig 2A); and, similarly, if R2 is the only ESS for both A and B, then no intersection can exist (see Fig 2B).
  2. If both R1 and R2 are ESSs for A but R1 is the only ESS for B, then only one intersection exists (see Fig 2C); and, similarly, if both R1 and R2 are ESSs for A but R2 is the only ESS for B, then only one intersection exists (see Fig 2D).
  3. If p* ∈ (0, 1) is an ESS for A and R1 is the only ESS for B, only one intersection exists (see Fig 2E); and, similarly, if p* is an ESS for A and R2 is the only ESS for B, then only one intersection exists (Fig 2F).
  4. If R2 is the only ESS for A and R1 is the only ESS for B, then only one intersection exists (see Fig 2G); and, similarly, if R1 is the only ESS for A and R2 is the only ESS for B, then only one intersection exists (see Fig 2H).
  5. If both R1 and R2 are ESSs for A and B, then at most four intersections can exist (see Fig 2I).
  6. If both R1 and R2 are ESSs for A and q* ∈ (0, 1) is an ESS for B, only two intersections exist (see Fig 2J).
  7. If p* is an ESS for A and q* is an ESS for B, the only one intersection exists (see Fig 2K).
  8. If p* is an ESS for A and both R1 and R2 are ESSs for B, then there are at most three intersections (see Fig 2L).
  9. If R2 is the only ESS for A and q* is an ESS for B, then only one intersection exists (see Fig 2M); and, similarly, If R1 is the only ESS for A and q* is an ESS for B, then only one intersection exists (see Fig 2N).
  10. If R2 is the only ESS for A and both R1 and R2 are ESSs for B, then there are at most two intersections (see Fig 2O); and, similarly, if R1 is the only ESS for A and both R1 and R2 are ESSs, then there are at most two intersections (see Fig 2P).

For the stability of the solutions of Eq (7) under dynamics (6), it is easy to see that for given (i.e. corresponds to the stable equilibrium of dynamics (5)), if and is an ESS for the payoff matrix B, then the corresponding equilibrium must be asymptotically stable under dynamics (6). On the hand, let be an interior equilibrium of dynamics (6), and the Jacobian matrix about , denoted by , is given by Clearly, the interior equilibrium is asymptotically stable (i.e. the eigenvalues of have the negative real parts) if

Thus, for given parameter values, stabilities of the interior equilibria of dynamics (6) (i.e., interior solutions of Eq (7)) can be analyzed numerically according to the above conditions (see the figure caption of Fig 2 for detailed parameters, note that the following results may not be true for all parameter values).

  1. If R1 (or R2) is the only ESS for A and B, then the boundary equilibrium with (or with ) is asymptotically stable (see also Fig 2A and 2B).
  2. If both R1 and R2 are ESSs for A but R1 (or R2) is the only ESS for B, then one boundary equilibrium with (or with ) is unstable and the other boundary equilibrium with (or with ) is asymptotically stable. Furthermore, the unique interior equilibrium is also asymptotically stable (see also Fig 2C and 2D).
  3. If p* is an ESS for A and R1 (or R2) is the only ESS for B, then the unique interior equilibrium is asymptotically stable (see also Fig 2E and 2F).
  4. If R2 (or R1) is the only ESS for A but R1 (or R2) is the only ESS for B, then the boundary equilibrium with (or with ) is unstable and the unique interior equilibrium is asymptotically stable (see also Fig 2G and 2H).
  5. If both R1 and R2 are ESSs for A and B, then the boundary equilibrium with , or the boundary equilibrium with , is stable, and for four interior equilibria, two are stable and the other two are unstable (see also Fig 2I).
  6. If both R1 and R2 are ESSs for A and q* is an ESS for B, then the boundary equilibrium with , or the boundary equilibrium with , is unstable, and the two interior equilibria are asymptotically stable (see also Fig 2J).
  7. If p* is an ESS for A and q* is an ESS for B, then the unique interior equilibrium is asymptotically stable (see also Fig 2K).
  8. If p* is an ESS for A and both R1 and R2 are ESSs for B, then there are at most three interior equilibria corresponding to three intersections of h1 and h2, which are denoted by , and with , the two interior equilibria corresponding to and , respectively, are stable and the interior equilibrium corresponding to is unstable (see also Fig 2L).
  9. If R2 (or R1) is the only ESS for A and q* is an ESS for B, then the boundary equilibrium with (or with is unstable and the unique interior equilibrium is asymptotically stable (see also Fig 2M and 2N).
  10. If R2 (or R1) is the only ESS for A and both R1 and R2 are ESSs for B, then there are at most two interior equilibria, the boundary equilibrium for (or for ) is stable, and one interior equilibrium is stable and the other unstable (see also Fig 2O and 2P).

Author Contributions

Conceived and designed the experiments: QL JZ BZ RC YT. Performed the experiments: QL JZ BZ RC YT. Analyzed the data: QL JZ BZ RC YT. Contributed reagents/materials/analysis tools: QL JZ BZ RC YT. Wrote the paper: QL JZ BZ RC YT.

References

  1. 1. Maynard Smith J, Price GR. The logical animal conflict. Nature. 1973; 246: 15–18.
  2. 2. Taylor PD, Jonker LB. Evolutionarily stable strategies and game dynamics. Math. Biosci. 1978; 40: 145–156.
  3. 3. Maynard Smith J. Evolution and the Theory of Games. Cambridge University Press; 1982.
  4. 4. Cressman R. The stability concept of evolutionary game theory: a dynamic approach. Lect Notes Biomath. vol. 94, Berlin, Heildelberg New York: Springer; 1992.
  5. 5. Hofbauer J, Sigmund K. Evolutionary Games and Population Dynamics. Cambridege University Press; 1998.
  6. 6. Prior TG, Hines WGS, Cressman R. Evolutionary games for spatially dispersed populations. J. Math. Biol. 1993; 32: 55–65.
  7. 7. Ludwig D, Levin SA. Evolutionary stability of plant communities and the maintenance of multiple dispersal types. Theor. Popul. Biol. 1991; 40: 285–307.
  8. 8. Comins HN, Hamilton WD, May RM. Evolutionary stable dispersal strategies. J. Theor. Biol. 1980; 82: 205–230.
  9. 9. Levin SA, Cohen D, Hastings A. Dispersal strategies in patchy environments. Theor. Popul. Biol. 1984; 26: 165–191.
  10. 10. Hutson VCL, Vickers GT. Traveling waves and dominance of ESS’s. J. Math. Biol, 1993; 30: 457–471.
  11. 11. Hofbauer J, Hutson VCL, Vickers GT. Travelling waves for games in economics and biology. Nonlinear. Anal. Theor. 1997; 30:1235–1244.
  12. 12. Dieckmann U, O’Hara B, Weisser W. The evolutionary ecology of dispersal. Trends Ecol. Evol. 1999; 14: 88–90.
  13. 13. Travis JMJ, Murrell DJ, Dytham C. The evolution of density dependent dispersal. Proc. R. Soc. Lond. B. 1999; 266: 1837–1842.
  14. 14. Hines WGS, Maynard Smith J. Games between relatives. J. Theor. Biol. 1979; 79: 19–30. pmid:513800
  15. 15. Fretwell SD, Lucas HL. On Terriotorial behavior and other factors influcing habitat distribution in birds. Acta Bio-theoretica. 1970; 19: 16–32.
  16. 16. Cressman R, Krivan V. Migration dynamics for the ideal free distribution. Am. Nat. 2006; 168: 384–397. pmid:16947113
  17. 17. Sun GQ, Jin Z, Li L, Haque M, Li BL. Spatial patterns of a predator-prey model with cross diffusion. Nonlinear. Dynam. 2012; 69: 1631–1638.
  18. 18. Sun GQ, Zhang J, Song LP, Jin Z, Li BL. Pattern formation of a spatial predator-prey system. Appl. Math. Comput. 2012; 218: 11151–11162.
  19. 19. Sun GQ, Chakraborty A, Liu QX, Jin Z, Anderson KE, Li BL. Influence of time delay and nonlinear diffusion on herbivore outbreak. Commun. Nonlinear. Sci. 2014; 19: 1507–1518.
  20. 20. Sun GQ, Wang SL, Ren Q, Jin Z, Wu YP. Effects of time delay and space on herbivore dynamics: linking inducible defenses of plants to herbivore outbreak. Sci. Rep. 2015; 5: 11246. pmid:26084812
  21. 21. Hofbauer J. The spatially dominant equilibrium of a game. Ann. Oper. Res. 1999; 89: 233–251.
  22. 22. Hugie DM, Grand TC. Movement between patches, unequal competitors and the ideal free distribution. Evol. Ecol. 1998; 12: 1–19.
  23. 23. Cressman R, Krivan V, Garay J. Ideal free distributions, evolutionary games, and population dynamics in multiple-species environments. Am. Nat. 2004; 164: 473–489. pmid:15459879