Detecting phase transitions in collective behavior using manifold's curvature

  • If a given behavior of a multi-agent system restricts the phase variable to an invariant manifold, then we define a phase transition as a change of physical characteristics such as speed, coordination, and structure. We define such a phase transition as splitting an underlying manifold into two sub-manifolds with distinct dimensionalities around the singularity where the phase transition physically exists. Here, we propose a method of detecting phase transitions and splitting the manifold into phase transitions free sub-manifolds. Therein, we firstly utilize a relationship between curvature and singular value ratio of points sampled in a curve, and then extend the assertion into higher-dimensions using the shape operator. Secondly, we attest that the same phase transition can also be approximated by singular value ratios computed locally over the data in a neighborhood on the manifold. We validate the Phase Transition Detection (PTD) method using one particle simulation and three real world examples.

    Citation: Kelum Gajamannage, Erik M. Bollt. Detecting phase transitions in collective behavior using manifold's curvature[J]. Mathematical Biosciences and Engineering, 2017, 14(2): 437-453. doi: 10.3934/mbe.2017027

    Related Papers:

    [1] Maoxiang Wang, Fenglan Hu, Meng Xu, Zhipeng Qiu . Keep, break and breakout in food chains with two and three species. Mathematical Biosciences and Engineering, 2021, 18(1): 817-836. doi: 10.3934/mbe.2021043
    [2] Hao Wang, Yang Kuang . Alternative models for cyclic lemming dynamics. Mathematical Biosciences and Engineering, 2007, 4(1): 85-99. doi: 10.3934/mbe.2007.4.85
    [3] Minus van Baalen, Atsushi Yamauchi . Competition for resources may reinforce the evolution of altruism in spatially structured populations. Mathematical Biosciences and Engineering, 2019, 16(5): 3694-3717. doi: 10.3934/mbe.2019183
    [4] Xiyan Yang, Zihao Wang, Yahao Wu, Tianshou Zhou, Jiajun Zhang . Kinetic characteristics of transcriptional bursting in a complex gene model with cyclic promoter structure. Mathematical Biosciences and Engineering, 2022, 19(4): 3313-3336. doi: 10.3934/mbe.2022153
    [5] A. Swierniak, M. Krzeslak, D. Borys, M. Kimmel . The role of interventions in the cancer evolution–an evolutionary games approach. Mathematical Biosciences and Engineering, 2019, 16(1): 265-291. doi: 10.3934/mbe.2019014
    [6] Vaibhava Srivastava, Eric M. Takyi, Rana D. Parshad . The effect of "fear" on two species competition. Mathematical Biosciences and Engineering, 2023, 20(5): 8814-8855. doi: 10.3934/mbe.2023388
    [7] Natalia L. Komarova . Mathematical modeling of cyclic treatments of chronic myeloid leukemia. Mathematical Biosciences and Engineering, 2011, 8(2): 289-306. doi: 10.3934/mbe.2011.8.289
    [8] Fabio Augusto Milner, Ruijun Zhao . A deterministic model of schistosomiasis with spatial structure. Mathematical Biosciences and Engineering, 2008, 5(3): 505-522. doi: 10.3934/mbe.2008.5.505
    [9] Ali Moussaoui, Vitaly Volpert . The influence of immune cells on the existence of virus quasi-species. Mathematical Biosciences and Engineering, 2023, 20(9): 15942-15961. doi: 10.3934/mbe.2023710
    [10] Jim M. Cushing . A Darwinian version of the Leslie logistic model for age-structured populations. Mathematical Biosciences and Engineering, 2025, 22(6): 1263-1279. doi: 10.3934/mbe.2025047
  • If a given behavior of a multi-agent system restricts the phase variable to an invariant manifold, then we define a phase transition as a change of physical characteristics such as speed, coordination, and structure. We define such a phase transition as splitting an underlying manifold into two sub-manifolds with distinct dimensionalities around the singularity where the phase transition physically exists. Here, we propose a method of detecting phase transitions and splitting the manifold into phase transitions free sub-manifolds. Therein, we firstly utilize a relationship between curvature and singular value ratio of points sampled in a curve, and then extend the assertion into higher-dimensions using the shape operator. Secondly, we attest that the same phase transition can also be approximated by singular value ratios computed locally over the data in a neighborhood on the manifold. We validate the Phase Transition Detection (PTD) method using one particle simulation and three real world examples.


    One of interactions between three types of populations is known as non-transitive, cyclic competition or a rock-paper-scissors relationship, that is, type A is stronger than type B, and type B is superior to type C, which in turn is better than A. We can find several examples of this kind of interaction in nature; the most well-known are bacteria (e.g., [1,2,3,4]) and lizards (e.g., [5,6,7]).

    Barreto et al. [8] proposed phenotypic and genotypic cyclic competition models to clarify the fact that three throat morphs in male lizards, known as orange, blue, and yellow, are maintained with a rock-paper-scissors relationship. Both models have the same equilibrium, but the latter gives a wider stability region than the former. Barreto et al. [8] analyzed their model by fixing all payoffs except two elements, which are given by variable parameters. However, such a case is difficult to manage because an internal equilibrium depends on these parameters. Then we consider an alternative setting as an internal equilibrium with equal frequencies for all phenotypes independent of parameters, namely, we use symmetric conditions of parameters to obtain global insights into the dynamics in this simplified system for the convenience of analysis.

    In this paper, we re-examine stability conditions under the following situations: symmetric parameters for the phenotypic model, cyclic allele dominance rule, and spatial structure. We obtain the following results: (ⅰ) Cyclic allele dominance rule in a genotypic model gives a wider stable region of internal equilibrium than the allele dominance rule observed in lizards. (ⅱ) Spatial structure drastically changes the dynamical behavior, especially when all three phenotypes coexist in almost all the parameter spaces when both competition and dispersal occur locally.

    In the next section, we review the phenotypic model and its corresponding genotypic model proposed by [8] and then restrict these models with two parameters to generate an internal equilibrium independent of the parameters. In section 3, we derive local stability conditions for each model. We consider cyclic allele dominance rule in section 4. In section 5, we investigate the effects of spatial structure on the stability of the internal equilibrium.

    First, we consider a phenotypic model of cyclic competition, or a rock-paper-scissors game, following [8]. There are three phenotypes O,B, and Y. O has a better strategy than B, and B is superior to Y, which is, in turn, surpasses O. When we use a payoff matrix in Table 1, this relation can be expressed by MOY<MOO<MOB,MBO<MBB<MBY, and MYB<MYY<MYO, where MIJ is a payoff of focal individual I{O,B,Y} against an opponent J{O,B,Y}.

    Table 1.  Payoff matrix of rock-paper-scissors game. MIJ is a payoff of focal individual I{O,B,Y} against an opponent J{O,B,Y}. The following magnitude relations hold: MOY<MOO<MOB,MBO<MBB<MBY,MYB<MYY<MYO.
    Individual J
    Individual I O B Y
    O MOO MOB MOY
    B MBO MBB MBY
    Y MYO MYB MYY

     | Show Table
    DownLoad: CSV

    Using these payoffs obtained by the competition between two phenotypes, the dynamics of phenotypic model are defined as

    pO,n+1=WOWpO,n, (2.1)
    pB,n+1=WBWpB,n, (2.2)
    pY,n+1=WYWpY,n, (2.3)

    with pO,n,pB,n, and pY,n as the fractions of phenotypes O,B, and Y, respectively, at generation n. Here WO,WB, and WY are the fitnesses of phenotypes O,B, and Y, respectively, and W is the average fitness of the population:

    WO=MOOpO,n+MOBpB,n+MOYpY,n, (2.4)
    WB=MBOpO,n+MBBpB,n+MBYpY,n, (2.5)
    WY=MYOpO,n+MYBpB,n+MYYpY,n, (2.6)
    W=WOpO,n+WBpB,n+WBpY,n. (2.7)

    This is a standard model of the rock-paper-scissors game (e.g., [9,10]).

    At equilibrium we have

    W=WO=WB=WY (2.8)

    by eqs.(2.1)–(2.3). A unique internal equilibrium solution for the above simultaneous equations (2.8) with pO+pB+pY=1 is

    pO=M1M1+M2+M3, (2.9)
    pB=M2M1+M2+M3, (2.10)
    pY=M3M1+M2+M3, (2.11)

    where

    M1=MOB(MBYMYY)+MBB(MYYMOY)+MYB(MOYMBY), (2.12)
    M2=MOY(MBOMYO)+MBY(MYOMOO)+MYY(MOOMBO), (2.13)
    M3=MOO(MBBMYB)+MBO(MYBMOB)+MYO(MOBMBB). (2.14)

    We have another three equilibria (1,0,0),(0,1,0), and (0,0,1) as the boundary equilibria (ˆpO,ˆpB,ˆpY).

    Next we move to a genotypic model corresponding to the above phenotypic model by the genotypic allele dominance rule in which an allele o dominates over two others, y and b, and an allele y dominates b. In other words, genotypes oo,oy, and ob are phenotype O, genotypes yy and yb are phenotype Y, and genotype bb is phenotype B (Table 2).

    Table 2.  Allele dominance rule (Ⅰ). An allele o is dominant to other alleles y and b, then three genotypes oo, oy, and ob correspond to a phenotype O. An allele y is dominant to an allele b, then two genotypes yy and yb correspond to a phenotype Y. An allele b is recessive to other alleles, then only one genotype bb corresponds to a phenotype B.
    Genotype Phenotype
    oo,oy,ob O
    yy,yb Y
    bb B

     | Show Table
    DownLoad: CSV

    Then phenotypic frequencies can be expressed by the genotypic frequencies gij,n (i,j{o,y,b}) at generation n as follows:

    pO,n=goo,n+goy,n+gob,n, (2.15)
    pY,n=gyy,n+gyb,n, (2.16)
    pB,n=gbb,n. (2.17)

    Therefore the fitness of each genotype becomes

    WO=Woo=Woy=Wob, (2.18)
    WY=Wyy=Wyb, (2.19)
    WB=Wbb. (2.20)

    Considering the above relationships of fitnesses and frequencies, first, the genotypic frequencies at generation n change:

    ˜gij,n=WijWgij,n(i,j{o,y,b}). (2.21)

    Second, the allele frequencies ˜fi (i{o,y,b}) at generation n are

    ˜fo,n=˜goo,n+12˜goy,n+12˜gob,n, (2.22)
    ˜fy,n=˜gyy,n+12˜gyb,n+12˜goy,n, (2.23)
    ˜fb,n=˜gbb,n+12˜gyb,n+12˜gob,n. (2.24)

    Third, assuming random mating gives the genotypic frequencies at generation n+1, gij,n+1 (i,j{o,y,b}) as

    gij,n+1=2˜fi,n˜fj,nif ij, (2.25)
    gii,n+1=˜f2i,n, (2.26)

    then we can describe the above relations and dynamics (2.15)–(2.26) with only variables representing the genotypic frequencies gij,n(i,j{0,y,b}) as

    goo,n+1=(WOW)2(goo,n+12goy,n+12gob,n)2, (2.27)
    goy,n+1=2WOW(goo,n+12goy,n+12gob,n)×{WYW(gyy,n+12gyb,n)+12WOgoy,nW}, (2.28)
    gob,n+1=2WOW(goo,n+12goy,n+12gob,n)×(WBgbb,nW+12WYgyb,nW+12WOgob,nW), (2.29)
    gyy,n+1={WYW(gyy,n+12gyb,n)+12WOgoy,nW}2, (2.30)
    gyb,n+1=2{WYW(gyy,n+12gyb,n)+12WOgoy,nW}×(WBgbb,nW+12WYgyb,nW+12WOgob,nW), (2.31)
    gbb,n+1=(WBgbb,nW+12WYgyb,nW+12WOgob,nW)2, (2.32)

    where

    WO=MOO(goo,n+goy,n+gob,n)+MOBgbb,n+MOY(gyy,n+gyb,n), (2.33)
    WB=MBO(goo,n+goy,n+gob,n)+MBBgbb,n+MBY(gyy,n+gyb,n), (2.34)
    WY=MYO(goo,n+goy,n+gob,n)+MYBgbb,n+MYY(gyy,n+gyb,n), (2.35)
    W=WO(goo,n+goy,n+gob,n)+WBgbb,n+WY(gyy,n+gyb,n). (2.36)

    Using the above equations, we can obtain the following relations:

    goo,n+1+12goy,n+1+12gob,n+1=WOW(goo,n+12goy,n+12gob,n), (2.37)
    gyy,n+1+12gyb,n+1+12goy,n+1=WYW(gyy,n+12gyb,n)+12WOWgoy,n, (2.38)
    gbb,n+1+12gyb,n+1+12gob,n+1=WBWgbb,n+12WYWgyb,n+12WOWgob,n. (2.39)

    At equilibrium, by eq.(2.37), we have

    W=WO. (2.40)

    By eq.(2.38) and eq.(2.40), we have

    W=WY. (2.41)

    By eqs.(2.39)–(2.41), we have

    W=WB.

    Therefore, because we also have the same relationships (2.8) for genetypic model, we have the same phenotypic equilibrium (2.9)–(2.11) with eqs.(2.12)–(2.14) as the corresponding phenotypic model.

    The payoff matrix includes nine payoffs and is a little complicated. Following [8], we reduce it to two values. We fix the values of the payoffs at 1 for the competitions between the same phenotypes, but we also adopt the symmetric cases --- namely, we give a larger payoff α>1 to a stronger competitor and a smaller one β<1 to a weaker competitor (Table 3), which results in the same equilibrium fractions of 1/3 for all three phenotypes [9].

    Table 3.  Payoff matrix with only two parameters. We obtain the same equilibrium fractions of 1/3 for all three phenotypes. Parameter ranges are given by 0<β<1<α.
    Opponent Individual
    Focal Individual O B Y
    O 1 α β
    B β 1 α
    Y α β 1

     | Show Table
    DownLoad: CSV

    By using this reduced payoff matrix, eqs.(2.1)–(2.7) become

    pO,n+1=(pO,n+αpB,n+βpY,n)pO,n(pO,n+αpB,n+βpY,n)pO,n+(βpO,n+pB,n+αpY,n)pB,n+(αpO,n+βpB,n+pY,n)pY,n, (3.1)
    pB,n+1=(βpO,n+pB,n+αpY,n)pB,n(pO,n+αpB,n+βpY,n)pO,n+(βpO,n+pB,n+αpY,n)pB,n+(αpO,n+βpB,n+pY,n)pY,n, (3.2)
    pY,n+1=(αpO,n+βpB,n+pY,n)pY,n(pO,n+αpB,n+βpY,n)pO,n+(βpO,n+pB,n+αpY,n)pB,n+(αpO,n+βpB,n+pY,n)pY,n. (3.3)

    By eqs.(2.12)–(2.14),

    M1=M2=M3=α2+β2(α+β+αβ)+1,

    so that an internal equilibrium (2.9)–(2.11) becomes

    pO=pY=pB=13. (3.4)

    The linearized system of eqs.(3.1)–(3.3) about an internal equilibrium (3.4) gives the following Jacobian matrix:

    13(1+α+β)(4+α+β2+α2β22α+β22α+β4+α+β2+α2β2+α2β22α+β4+α+β).

    Then the characteristic equation of the linearized system becomes

    λ(λ2+a1λ+a2)=0,

    where

    a1=4+α+β1+α+β, (3.5)
    a2=4+2(α+β)+α2+β2(1+α+β)2. (3.6)

    Jury conditions or Schur-Cohn criteria (e.g., [11]) for second-degree characteristic equations λ2+a1λ+a2=0 are known as

    |a1|<1+a2<2. (3.7)

    Because a1<0 from eq.(3.5), the left inequality of eq.(3.7) becomes a1<1+a2, that is, a1+a2+1>0, which holds by the calculation using eqs.(3.5)–(3.6):

    a1+a2+1=(1α)2+(1β)2+(αβ)22(1+α+β)2>0.

    The right inequality of eq.(3.7) 1+a2<2, that is, a21<0, gives

    αβ>1. (3.8)

    Notice that it is symmetric with respect to α and β; namely, eq.(3.8) does not change when α and β are exchanged.

    Indeed, it is known that eq.(3.8) becomes a globally asymptotic stable condition, which can be checked using the Lyapunov function [9]. If eq.(3.8) does not hold, then we can observe the trajectory of the heteroclinic cycle between three vertices in the triangular space for three phenotypic frequencies [9] similar to the three-species Lotka-Volterra cyclic competition model [12] (See Figure 1(a) and (b)).

    Figure 1.  Trajectories of phenotypic frequencies. (a)-(c) for eqs.(3.1)-(3.3), (d)-(f) for eqs.(2.27)-(2.36) with Table 3. β is fixed to 0.3. α=2 for (a) and (d), α=3 for (b) and (e), α=4 for (c) and (f). Figure 2(a) tells us that internal equilibrium is unstable for (a), (b) and (d), but stable for (c), (e) and (f). Colorbars indicate the time steps of the dynamics.
    Figure 2.  Stable and unstable region on parameter spaces of the equilibrium point (3.4) for (a) allele dominance rule (Ⅰ) and (b) allele dominance rule (Ⅱ). α and β indicate payoffs for strong and weak competitors, respectively. Green, red and yellow regions indicate stable for both phenotypic and genotypic models, stable for genotypic model but unstable for phenotypic one and unstable for both phenotypic and genotypic models, respectively. Allele dominance rule (Ⅱ) has larger stability region than (Ⅰ).

    On the other hand, for a genotypic model, we get the following unique internal equilibrium by eqs.(2.27)–(2.36):

    goo=(32)22, (3.9)
    goy=(32)(21), (3.10)
    gob=23(32), (3.11)
    gyy=13(21)2, (3.12)
    gyb=23(21), (3.13)
    gbb=13. (3.14)

    Of course,

    pO=goo+goy+gob=13,pY=gyy+gyb=13,PB=gbb=13.

    We can say that an internal equilibrium does not depend on the parameters α nor β on either the phenotyic or genotypic models.

    The characteristic equation of the linearized system of eqs.(2.27)–(2.32) with eqs.(2.33)–(2.36) about an internal equilibrium (3.9)–(3.14) becomes

    λ4(λ2+a1λ+a2)=0,

    where

    a1=233(1+α+β){(22+23)+(2321)(α+β)}, (3.15)
    a2=1(1+α+β)2[8211+43(22)+{22(4+3)(63+5)}(α2+β2)+2{4+332(4+3)}(α+β)2{22(2+3)(7+23)}αβ]. (3.16)

    Noticing that a1<0 from eq.(3.15), the left inequality of eq.(3.7) holds because, using eqs.(3.15)–(3.16),

    a1+a2+1=43(32)(21)3(1+α+β)2{(1α)2+(1β)2+(αβ)2}>0.

    The right inequality of eq.(3.7) a21<0 becomes

    2(3+1)(3+2)(αβ1)+α(α1)+β(β1)>0, (3.17)

    which is a quadratic inequation on α or β and is also symmetric with respect to α and β.

    Figure 2(a) shows the local stability region obtained by eq.(3.8) and eq.(3.17). We can observe their dynamics by numerical simulations (Figure 1).

    Common side-blotched lizards have an allele dominance rule (Ⅰ), that is, an allele o for an orange throat is the most dominant over others, an allele y for a yellow one is intermediate, and an allele b for a blue one is the most recessive (Table 2). In other words, this allele dominance rule has three kinds of alleles: most dominant, intermediate, and most recessive. We refer to these as o, y, and b, respectively.

    Although perhaps we have not yet discovered it in realistic genetic systems, we can theoretically consider another allele dominance rule (Ⅱ), that is, all alleles having intermediate dominance: o dominant over y, y over b, and b over o (Table 4). This rule could be called as a "cyclic allele dominance rule." Similarly, in this rule, we name these alleles o, y, and b, and any allele can be replaced by another one.

    Table 4.  Allele dominance rule (Ⅱ). An allele o is dominant to an allele y, then two genotypes oo and oy correspond to a phenotype O. An allele y is dominant to an allele b, then two genotypes yy and yb correspond to a phenotype Y. An allele b is dominant to an allele o, then two genotypes bb and ob correspond to a phenotype B.
    Genotype Phenotype
    oo,oy O
    yy,yb Y
    bb,ob B

     | Show Table
    DownLoad: CSV

    A calculation for allele dominance rule (Ⅱ) similar to that for allele dominance rule (Ⅰ) reveals that the boundary of local stability obtained from the Jury condition (3.7) is determined by

    8(αβ1)+α(α1)+β(β1)=0, (4.1)

    which only differs from the first calculation in the smaller coefficient of αβ1 compared to eq.(3.17). Figure 2(b) shows the local stability region by allele dominance rule (Ⅱ) using eq.(4.1), which clearly has a larger stability region than rule (Ⅰ).

    From a theoretical point of view, we should also consider the cases with the same allele dominance rules but with different cyclic competitive strengths. Then another possible combination exists, that is, Table 2 with α<1 and β>1. However, this combination gives the same stability regions as Figure 2(a) by the symmetry of the boundary equation (3.17) on the parameters α and β; eq.(3.17) does not change when α and β are exchanged. On the other hand, it is trivial that no qualitative difference exists between the two cases: Table 4 with α>1 and β<1 and Table 4 with α<1 and β>1.

    Barreto et al. [8] showed the same stabilizing effect in a genetic system of lizards as the previous section of this paper. Here we introduce a spatial structure into a phenotypic model and consider its effects on population dynamics.

    The phenotypic model includes two processes: competition and reproduction. When we add a spatial structure into this model, we restrict the spatial range, both for the opponents against a focal individual and for the dispersion of offspring by reproduction. When we set such a spatial restriction as a whole or as a neighborhood, we consider four distinct cases: (a) global competition and global dispersion, (b) global competition and local dispersion, (c) local competition and global dispersion, and (d) local competition and local dispersion.

    The Monte Carlo simulation procedures for the above four cases use the following algorithm (See the C program code in the Electronic Supplementary Materials):

    (ⅰ) We prepare a two-dimensional square lattice space, each-side with a size of 100, so that the total number of sites on a whole lattice becomes N=100×100=10000. We use a periodic boundary condition. The states O, Y, and B are randomly distributed according to the initial fractions: (pO,pY,pB)=(0.3,0.3,0.4), respectively.

    (ⅱ) The first process is competition, which determines the payoffs after games against randomly chosen opponents. All individuals experience this process. The payoffs for all the sites i (=1,,N) in a whole population with size N are determined in order; the site (,m) with x-coordinate  (=1,,100) and y-coordinate m (m=1,,100) on the two-dimensional square lattice is numbered (1)×100+m. An individual I{O,B,Y} is included on the site i. Another site j is randomly chosen in a whole population in the cases of (a) and (b) or, in the cases of (c) and (d), in the nearest neighboring four individuals. An individual J{O,B,Y} is included on the site j. Two individuals I and J compete and an individual I gets a payoff by a payoff matrix (Table 3). All the payoffs are transferred to their offspring, and they are gathered in one offspring pool in the cases of (a) and (c) or, in the cases of (b) and (d), in local offspring pools. Here, we use lower cases for the sites and corresponding upper cases for the states of the individuals on those sites.

    (ⅲ) The second process is dispersion, which gives positions of offspring selected from the offspring pool. The model is discrete-generation: all individuals die, and the states at the next time steps for all the sites i (=1,,N) in a whole population with size N are replaced in order. For each site, an individual is randomly chosen from a whole pool for (a) and (c) or from each local pool for (b) and (d); this choice is made in proportion to the relative payoff against total payoffs in a whole population in the cases of (a) and (c) or in the nearest neighboring four individuals in the cases of (b) and (d). In other words, this random selection depends on WOW,WYW, and WBW. After the previous process (ⅱ), all the individuals have their own payoffs. In the cases of (a) and (c), we can then calculate W,WO,WY, and WB as the total payoffs for, respectively, a whole population, phenotype O, phenotype Y, and phenotype B. In the cases of (b) and (d), we can obtain W,WO,WY, and WB as the total payoffs for, respectively, a local population, phenotype O in the local population, phenotype Y in the local population, and phenotype B in the local population. Here a local population is restricted to individuals on five sites: the focal site and its nearest neighboring four sites.

    (ⅳ) The above procedures (ⅱ) and (ⅲ) are repeated for 10000 (=100×100) time steps for each parameter combination of α=1,1.05,1.10,,5 and β=0,0.05,0.10,,1. The repetition of the simulations is 100 for each parameter combination, and we record the fraction of the coexistence of the three phenotypes.

    We show the results of the Monte Carlo simulation by the above algorithm in Figure 3. Case (a) completely coincides with Figure 2(a). Case (b) gives a wider coexistence region due to local dispersion. Case (c) has the same result as case (a) because the global dispersal produces the same effect with random choices of opponents from the population. Case (d) shows coexistence in almost all the simulations for the entire parameter space. Therefore, we can conclude that locally limited interactions between individuals strongly promote the coexistence of all phenotypes.

    Figure 3.  Phase diagram for coexistence and non-coexistence. Each parameter α or β is set for every 0.05 increment. Initial values are given as (pO,pB,pY)=(0.3,0.3,0.4). Simulation runs are repeated for 100 times for each parameter set and their fractions are shown by continuous gradation between green and yellow. Green indicates coexistence until the end of simulations. Yellow corresponds to non-coexistence in which one phenotype goes extinct, and then an inferior phenotype disappears and ultimately only one phenotype survives, due to finite size of the system. (Upper left panel) global competition and global dispersion; (Upper right panel) global competition and local dispersion; (Lower left panel) local competition and global dispersion; (Lower right panel) local competition and local dispersion.

    In case (d), individuals with the same phenotypes tend to cluster and adopt collective behavior, so changes of state occur only on the boundaries between these clusters, causing a slower change of total population and stabilization (See Figure 4). Indeed [13] already reached such a conclusion from a continuous-time cyclic competition model that also calculated the average domain size or boundary length at a steady state. This spatial structure in the population is gradually produced by both local competition and local dispersion, and, in turn, also gives different results in competition and dispersion, either globally or locally.

    Figure 4.  Snapshot of spatial pattern by Monte Carlo simulation of phenotypic model on two-dimensional square lattice space after some time. Both competition and dispersion occur locally. We can observe that several dozens of same phenotypic individuals are clustering. (Right panel) The whole spatial pattern with 1000×1000 lattice sites. (Left panel) The upper left part of spatial pattern with 100×100 lattice sites. Colors indicate O,B,Y as orange, blue, yellow, respectively. α=2.0 and β=0.3.

    Similar to [8], we show the difference between the phenotypic model and its corresponding genotypic model, but we can also give an explicit condition using a simplified system. In addition, we investigate other allele dominance rules to clarify their effects on the stability of internal equilibrium compared to the allele dominance rule observed in lizards. Unfortunately, however, we cannot specify the mechanism by which the genotypic model stabilizes the system more than the phenotypic model. To discover it, we need to find models other than cyclic competition in which the genotypic models give different dynamic behaviors than the corresponding phenotypic models.

    A recent proposal by [14] may have important implications for future work in this area. As they point out, the present three-strategy payoff matrix can be built up as the sum of nine independent Fourier components, g(1),,g(9), with orthogonality and normalization:

    (1αββ1ααβ1)=(1+α+β)13(111111111)+22(2αβ)118(211121112)+62(βα)16(011101110)=(1+α+β)g(1)+22(2αβ)g(8)+62(βα)g(9),

    where we have g(1),g(8), and g(9) as irrelevant constant terms, a coordination (Potts) component that equally favors the formation of one of the homogeneous states, and a cyclic (rock-paper-scissors) component, respectively. This formulation may offer more detailed insights into dynamical behaviors if we examine cases with various combinations of coefficients corresponding to the strengths of these components.

    The intuitive reason local competition and local dispersion both stabilize an internal equilibrium is that several dozens of identical phenotypic individuals gather and adopt collective behavior, restricting phenotypic changes on the boundary. To investigate their stabilization mechanisms in detail, we should further rely on other kinds of analysis, especially spatial pattern formation, which may play an important role in stabilizing the dynamics. So far, the concept of "vortex" has been proposed to characterize the spatial pattern [15,16,17], and it has also been shown that clockwise or anti-clockwise rotating spiral patterns with characteristic sizes emerge that yield finite size effects by cyclic competition models or rock-paper-scissors games in physics [18,19,20,21] and evolutionary games [22,23]. More precisely, the coexistence of three strategies transforms into one of the homogeneous states after a suitable relaxation time if the characteristic length of patterns exceeds a threshold value comparable to the system size. Here we use a small lattice space size (100×100) for Figure 3, but we should further investigate these finite size effects using a larger lattice space, such as 1000×1000, as in Figure 4.

    In this paper, we only show whether three phenotypes can coexist or not in the long run using a Monte Carlo simulation. However, we can expect an internal equilibrium to be asymptotically stable for the case of coexistence in larger Monte Carlo systems than in smaller ones (Figure 5).

    Figure 5.  Time series of frequencies by Monte Carlo simulation of phenotypic model on two-dimensional square lattice space. Both competition and dispersion occur locally. We can confirm that fluctuations are due to the effect of finite size of lattice sites. The numbers of lattice sites are (a) 1000×1000, (b) 100×100, (c) 10×10. All phenotypes survived after 2000 Monte Carlo steps in (a) and (b), but in (c) B went extinct around 70 generations by chance, and then O died out almost inevitably (but not definitely because of finite system size) because Y is stronger than O. Colors indicate O,B,Y as orange, blue, yellow, respectively. Initial frequencies are (pO,pB,pY)=(0.3,0.3,0.4). α=2.0 and β=0.3.

    Here we can only show the condition of the locally asymptotic stability of a genotypic model. We expect that condition to be replaced by global stability, but we leave it as a future problem.

    In addition, we would like to examine a genotypic model in a lattice space and clarify whether it becomes more stable than a phenotypic model and to what degree its stability can be evaluated by, for example, the return time of perturbation to an internal equilibrium.

    Throughout this paper, we use only symmetric conditions of parameters to obtain global insights into the dynamics, then we should study more complicated models with symmetry-breaking interactions, including the parameter setting of [8], in future research.

    I sincerely appreciate Prof. Akira Sasaki for giving me useful comments and encouraging me. I would also like to thank three anonymous reviewers whose comments were helpful in revising the manuscript.

    The author declares no conflict of interest.

    [1] [ Birds flying away, shutterstock. Available from: https://www.shutterstock.com/video/clip-3003274-stock-footage-birds-flying-away.html?src=search/Yg-XYej1Po2F0VO3yykclw:1:19/gg.
    [2] [ Data set of detection of unusual crowd activity available at robotics and vision laboratory, Department of Computer Science and Engineering, University of Minnesota. Available from: http://mha.cs.umn.edu/proj_events.shtml.
    [3] [ Data set of pet2009 at Computational Vision Group, University of Reading, 2009. Available from: http://ftp.pets.reading.ac.uk/pub/.
    [4] [ N. Abaid,E. Bollt,M. Porfiri, Topological analysis of complexity in multiagent systems, Physical Review E, 85 (2012): 041907.
    [5] [ G. Alfred, Modern Differential Geometry of Curves and Surfaces with Mathematica, CRC press, 1998.
    [6] [ I. R. de Almeida and C. R. Jung, Change detection in human crowds, in Graphics, Patterns and Images (SIBGRAPI), 2013 26th SIBGRAPI-Conference on, IEEE, (2013), 63-69.
    [7] [ E. L. Andrade, S. Blunsden and R. B. Fisher, Hidden markov models for optical flow analysis in crowds, in Pattern Recognition, 2006. ICPR 2006. 18th International Conference on, IEEE, 1(2006), 460-463.
    [8] [ M. Ballerini,N. Cabibbo,R. Candelier,A. Cavagna,E. Cisbani,I. Giardina,A. Orlandi,G. Parisi,A. Procaccini,M. Viale, Empirical investigation of starling flocks: A benchmark study in collective animal behavior, Animal Behaviour, 76 (2008): 201-215.
    [9] [ C. Becco,N. Vandewalle,J. Delcourt,P. Poncin, Experimental evidences of a structural and dynamical transition in fish school, Physica A: Statistical Mechanics and its Applications, 367 (2006): 487-493.
    [10] [ M. Beekman,D. J. T. Sumpter,F. L. W. Ratnieks, Phase transition between disordered and ordered foraging in pharaoh's ants, Proceedings of the National Academy of Sciences, 98 (2001): 9703-9706.
    [11] [ A. C. Bovik, Handbook of Image and Video Processing, Academic press, 2010.
    [12] [ R. Bracewell, Fourier Analysis and Imaging, Springer Science & Business Media, 2010.
    [13] [ I. D. Couzin, Collective cognition in animal groups, Trends in cognitive sciences, 13 (2009): 36-43.
    [14] [ I. D. Couzin,J. Krause,N. R. Franks,S. A. Levin, Effective leadership and decision-making in animal groups on the move, Nature, 433 (2005): 513-516.
    [15] [ A. Deutsch, Principles of biological pattern formation: Swarming and aggregation viewed as self organization phenomena, Journal of Biosciences, 24 (1999): 115-120.
    [16] [ J. H. Friedman,J. L. Bentley,R. A. Finkel, An algorithm for finding best matches in logarithmic expected time, ACM Transactions on Mathematical Software, 3 (1977): 209-226.
    [17] [ K. Gajamannage,S. Butailb,M. Porfirib,E. M. Bollt, Model reduction of collective motion by principal manifolds, Physica D: Nonlinear Phenomena, 291 (2015): 62-73.
    [18] [ K. Gajamannage,S. Butailb,M. Porfirib,E. M. Bollt, Identifying manifolds underlying group motion in Vicsek agents, The European Physical Journal Special Topics, 224 (2015): 3245-3256.
    [19] [ J. J. Gerbrands, On the relationships between SVD, KLT and PCA, Pattern recognition, 14 (1981): 375-381.
    [20] [ R. Gerlai, High-throughput behavioral screens: The first step towards finding genes involved in vertebrate brain function using zebra fish, Molecules, 15 (2010): 2609-2622.
    [21] [ G. H. Golub,C. Reinsch, Singular value decomposition and least squares solutions, Numerische Mathematik, 14 (1970): 403-420.
    [22] [ D. Helbing,J. Keltsch,P. Molnar, Modelling the evolution of human trail systems, Nature, 388 (1997): 47-50.
    [23] [ J. M. Lee, Riemannian Manifolds: An Introduction to Curvature, volume 176, Springer, 1997.
    [24] [ J. M. Lee, Introduction to Smooth Manifolds, Graduate Texts in Mathematics, 218. Springer-Verlag, New York, 2003.
    [25] [ R. Mehran, A. Oyama and M. Shah, Abnormal crowd behavior detection using social force model, in Computer Vision and Pattern Recognition, 2009. CVPR 2009. IEEE Conference on, IEEE, (2009), 935-942.
    [26] [ M. M. Millonas, Swarms, Phase Transitions, and Collective Intelligence, Technical report, Los Alamos National Lab., New Mexico, USA, 1992.
    [27] [ S. R. Musse and D. Thalmann, A model of human crowd behavior: Group inter-relationship and collision detection analysis, in Computer Animation and Simulation, Springer, (1997), 39-51.
    [28] [ M. Nagy,Z. Ákos,D. Biro,T. Vicsek, Hierarchical group dynamics in pigeon flocks, Nature, 464 (2010): 890-893.
    [29] [ B. O'neill, null, Elementary Differential Geometry, Academic press, New York, 1966.
    [30] [ T. Papenbrock and T. H. Seligman, Invariant manifolds and collective motion in many-body systems, AIP Conf. Proc. , 597(2001), p301, arXiv: nlin/0206035.
    [31] [ B. L. Partridge, The structure and function of fish schools, Scientific American, 246 (1982): 114-123.
    [32] [ W. Rappel,A. Nicol,A. Sarkissian,H. Levine,W. F. Loomis, Self-organized vortex state in two-dimensional dictyostelium dynamics, Physical Review Letters, 83 (1999): p1247.
    [33] [ E. M. Rauch,M. M. Millonas,D. R. Chialvo, Pattern formation and functionality in swarm models, Physics Letters A, 207 (1995): 185-193.
    [34] [ V. Y. Rovenskii, Topics in Extrinsic Geometry of Codimension-one Foliations, Springer, 2011.
    [35] [ S. T. Roweis,L. K. Saul, Nonlinear dimensionality reduction by locally linear embedding, Science, 290 (2000): 2323-2326.
    [36] [ R. V. Solé,S. C. Manrubia,B. Luque,J. Delgado,J. Bascompte, Phase transitions and complex systems: Simple, nonlinear models capture complex systems at the edge of chaos, Complexity, 1 (1996): 13-26.
    [37] [ D. Somasundaram, Differential Geometry: A First Course, Alpha Science Int'l Ltd., 2005.
    [38] [ D. Sumpter,J. Buhl,D. Biro,I. Couzin, Information transfer in moving animal groups, Theory in Biosciences, 127 (2008): 177-186.
    [39] [ J. B. Tenenbaum,V. De Silva,J. C. Langford, A global geometric framework for nonlinear dimensionality reduction, Science, 290 (2000): 2319-2323.
    [40] [ E. Toffin,D. D. Paolo,A. Campo,C. Detrain,J. Deneubourg, Shape transition during nest digging in ants, Proceedings of the National Academy of Sciences, 106 (2009): 18616-18620.
    [41] [ C. M. Topaz,A. L. Bertozzi, Swarming patterns in a two-dimensional kinematic model for biological groups, SIAM Journal on Applied Mathematics, 65 (2004): 152-174.
    [42] [ T. Vicsek,A. Cziró,E. Ben-Jacob,I. Cohen,O. Shochet, Novel type of phase transition in a system of self-driven particles, Physical Review Letters, 75 (1995): 1226-1229.
    [43] [ E. Witten, Phase transitions in m-theory and f-theory, Nuclear Physics B, 471 (1996): 195-216.
    [44] [ P. N. Yianilos, Data structures and algorithms for nearest neighbor search in general metric spaces, in Proceedings of the Fourth Annual ACM-SIAM Symposium on Discrete Algorithms, Society for Industrial and Applied Mathematics, (1993), 311-321.
    [45] [ T. Zhao and R. Nevatia, Tracking multiple humans in crowded environment, in Computer Vision and Pattern Recognition, 2004. CVPR 2004. Proceedings of the 2004 IEEE Computer Society Conference on, IEEE, (2004), Ⅱ-406.
  • This article has been cited by:

    1. A. Szolnoki, B. F. de Oliveira, D. Bazeia, Pattern formations driven by cyclic interactions: A brief review of recent developments, 2020, 131, 1286-4854, 68001, 10.1209/0295-5075/131/68001
    2. Sourav Kumar Sasmal, Yasuhiro Takeuchi, Editorial: Mathematical Modeling to Solve the Problems in Life Sciences, 2020, 17, 1551-0018, 2967, 10.3934/mbe.2020167
  • Reader Comments
  • © 2017 the Author(s), licensee AIMS Press. This is an open access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/4.0)
通讯作者: 陈斌, bchen63@163.com
  • 1. 

    沈阳化工大学材料科学与工程学院 沈阳 110142

  1. 本站搜索
  2. 百度学术搜索
  3. 万方数据库搜索
  4. CNKI搜索

Metrics

Article views(3440) PDF downloads(511) Cited by(2)

Figures and Tables

Figures(9)

Other Articles By Authors

/

DownLoad:  Full-Size Img  PowerPoint
Return
Return

Catalog