Abstract
We study the evolution of dispersal in advective three-patch models with distinct network topologies. Organisms can move between connected patches freely and they are also subject to the passive, directed drift. The carrying capacity is assumed to be the same in all patches, while the drift rates could vary. We first show that if all drift rates are the same, the faster dispersal rate is selected for all three models. For general drift rates, we show that the infinite diffusion rate is a local Convergence Stable Strategy (CvSS) for all three models. However, there are notable differences for three models: For Model I, the faster dispersal is always favored, irrespective of the drift rates, and thus the infinity dispersal rate is a global CvSS. In contrast, for Models II and III a singular strategy will exist for some ranges of drift rates and bi-stability phenomenon happens, i.e., both infinity and zero diffusion rates are local CvSSs. Furthermore, for both Models II and III, it is possible for two competing populations to coexist by varying drift and diffusion rates. Some predictions on the dynamics of n-patch models in advective environments are given along with some numerical evidence.
Similar content being viewed by others
Avoid common mistakes on your manuscript.
1 Introduction
Since the pioneering work of Speirs and Gurney on the “drift paradox” (Speirs and Gurney 2001), studying population dynamics in advective environments (such as rivers) has become an active research topic, both empirically as well as theoretically (Fagan 2002; Grant et al. 2007; Huang et al. 2016; Jin et al. 2014, 2019; Jin and Lewis 2011, 2012; Lam et al. 2016; Lutscher et al. 2006, 2007, 2005; Pachepsky et al. 2015; Vasilyeva and Lutscher 2011). Most mathematical models in spatial ecology assume that individuals adopt random movement, i.e., the transition probability in all directions are the same. For the organisms in advective environments, they are also subject to the passive, uni-directional drift. Such passive drift may push the organisms to the downstream where the environments could become unfavorable. From the mathematical point of view, the addition of drift makes the differential operators under consideration non-symmetric and thus brings new challenges to the stability analysis, especially for those models for interacting species (Lou et al. 2016, 2019; Vasilyeva 2017; Vasilyeva and Lutscher 2012a, b; Wang and Shi 2019; Wang et al. 2019; Wang and Shi 2020; Zhao and Zhou 2016; Zhou 2016). Almost all of these studies assume that the underlying habitat is an interval in the real line, in order to simplify the mathematical analysis. In contrast, there are rather few studies on the population dynamics in river networks, and they are mostly restricted to the case of a single species (Jin et al. 2019; Ramirez 2012; Samia et al. 2015; Sarhad et al. 2014; Vasilyeva 2019).
One important topic in spatial ecology is the evolution of dispersal. The seminal work of Hastings shows that random dispersal is selected against in spatially heterogeneous but temporally constant environments (Hastings 1983; Dockery et al. 1998), while in spatially and temporally varying environments large dispersal rate can be selected (Hutson et al. 2001; McPeek and Holt 1992). See the review article (Cosner 2014) and the references therein. The evolution of dispersal in advective and continuous habitats has been recently considered: when the carrying capacity is spatially heterogeneous and the drift rates are constants, some intermediate dispersal rate could be selected; see (Golubitsky et al. 2017; Lam et al. 2014). However, for a homogeneous environment where both carrying capacity and drift rates are spatially uniform, it was shown that when the loss at the downstream is not significant, the faster dispersal rate is favored (Lou and Lutscher 2014; Lou and Zhou 2015); see (Hao et al. 2021) for more recent progress. Again, these studies assume that the habitat is a finite interval.
Many of the above work employ the conceptual framework of adaptive dynamics theory (Dieckmann and Law 1996; Geritz et al. 1998). A central idea of adaptive dynamics theory is the evolutionarily stable strategy (ESS) (Maynard-Smith and Price 1973). A strategy is called a global ESS if the resident species adopting such a strategy cannot be invaded by any rare mutant species using different strategy. Another important concept is the convergence stable strategy (CvSS). A strategy is said to be a global CvSS if the mutant species is always able to invade a resident species when the mutant strategy is closer to the CvSS than the resident strategy. Local ESS and CvSS can be similarly defined and interpreted.
Our aim is to study the evolution of dispersal in discrete, advective environments using the conceptual framework of adaptive dynamics theory. Recently, the authors proposed in (Jiang et al. 2020) to study the dynamics of two competing species in three-patch models with different network topology, and to investigate how the topology of directed river network modules may affect the evolution of dispersal. To be specific, we considered the following three types of river network modules in (Jiang et al. 2020):
In Fig. 1a–c we assume that patch 1 is upstream, patch 3 is downstream, and patch 2 is either upstream, or middle stream, or downstream. In Jiang et al. (2020) the carrying capacity of three patches is assumed to be different and the drift rates are assumed to be equal. The main findings in Jiang et al. (2020) are summarized as follows: when the drift rate is small, for all three models the mutant species can invade when rare if and only if it is the slower disperser. However, when the drift rate is large, Models I and II predict that the faster disperser wins, while Model III predicts that fast and slow dispersers may coexist, and that there exists one intermediate strategy which is evolutionarily singular. For the intermediate range of drift, Models I and II predict the existence of one singular strategy, which may or may not be evolutionarily stable, depending upon the topology of modules, while Model III predicts singular strategy may not exist and the faster disperser wins the competition.
The rest of this paper is organized as follows: In Sect. 2 we state the main results for three-patch models. In Sect. 3 we draw the main conclusions and also provide a single framework to unify our main results. In Sect. 4 we present the numerical simulations of some 4-patch models and discuss some predictions on n-patch models. The proofs of the main results for Model I to III are postponed to the Appendices.
2 Main Results for Three-patch Models
In this paper, we assume that the drift rates could be different but the carrying capacity is the same in all three patches. As in Jiang et al. (2020), in all models the two competing species are assumed to be identical except for their dispersal rates.
Our main goal in this paper is to illustrate the effects of varying drift rates and network topology on the evolution of dispersal.
The main findings can briefly be summarized as follows:
-
If all drift rates are identical, then the faster dispersal rate is selected across all three-patch models in which the drift network do not form a closed cycle.
-
For general drift rates, infinite diffusion rate is a local CvSS for all three models.
-
For Models II and III, when a singular strategy (that is neither zero nor infinity) exists, it is not a local CvSS (Numerical simulation suggests that it is not an ESS either).
-
For Models II and III, when bi-stability occurs, it is possible for two competing populations with different dispersal rates to coexist, by varying the drift rates between patches.
2.1 Main Results of Model I
In Model I, the species in patches 1 and 2 are washed down to patch 3 by drift with rates \(q_1, q_2\), respectively. Two competing populations can disperse freely between the upstream patches and the downstream patch, with respective rates d, D. The two upstream patches, however, are not directly connected. The diagram of Model I is shown in Fig. 1a. The dynamics of two competing populations in this river module is described by the following system of ODEs:
Here \(u_i(t),v_i(t)\) \((i=1,2,3)\) denote the numbers of individuals of the respective species at time t in patch i.
The parameter k is the carrying capacity for all patches. For the sake of simplicity, the intrinsic growth rates are assumed to be equal to 1. The initial data of \(u_i\) and \(v_i\) are assumed to be positive for the rest of the paper so that \(u_i(t), v_i(t)\) are positive functions of time \(t>0\).
It can be shown that system (1) has a unique semi-trivial steady state of the form \((U^*,0)=(U^*_1,U^*_2,U^*_3,0,0,0)\), where \(U^*_i>0\) for \(i=1,2,3\).
Theorem 1
For any \(q_1\ge 0\), \(q_2\ge 0\) and \(q_1 +q_2>0\), if \(d>D\), then \((U^*,0)\) is globally asymptotically stable among all solutions of (1) with positive initial data.
This result implies that the faster dispersal is always selected for Model I, provided that the carrying capacity is uniform in the habitat, and the conclusion is independent of the drift rates. The underlying biological intuition is that a single population at equilibrium (i.e., resident) is undermatching the resources in at least one of the two upstream patches and it is always overmatching the resource in the downstream patch; i.e., the downstream patch is always a sink and at least one of the upstream patches is a source. If a mutant with small diffusion rate is introduced, its individuals in the upstream patches will be washed to the downstream patch, where the mutant cannot invade when rare as the downstream patch is a sink. Hence, small diffusion rate is selected against. In contrast, faster diffusion can counterbalance the drift by keeping more mutant individuals in upstream patches, one of which is a source patch, and thus help the mutant populations establish in this upstream source patch.
2.2 Main Results of Model II
Model II assumes that individuals in patch 1 are transported to patch 2 by drift with rate \(q_1\), and individuals in patch 2 are transported to patch 3 by drift with rate \(q_2\). Individuals can also move between patches i and \(i+1\) for \(i=1,2\); see Fig. 1b. The dynamics of two competing species in this network module is governed by the following ODE system:
For Model II, it can also be shown that system (2) has a unique semi-trivial steady state of the form \((U^*,0)=(U_1^*,U_2^*,U_3^*,0,0,0)\), where \(U_i^*>0, i=1,2,3\).
Theorem 2
If \(q_1 \ge 1\) or \(\frac{q_2}{2}<q_1<1\), then \((U^*,0)\) is globally asymptotically stable for \(d>D\); i.e., the faster diffuser wins.
If \(q_1\ge 1\) and the diffusion rate of a species is small, then almost all of its individuals in patch 1 are washed out. Thus, small diffusion is not favored in this scenario. However, large diffusion will be selected as it can counterbalance the uni-dimensional drift by helping more individuals stay in patch 1. Similar intuition applies to the case \(q_2/2<q_1<1\), but the detail is more subtle: when diffusion rate is small, our analysis reveals that there will be more individuals in patch 2 than patch 1 when \(q_2/2<q_1<1\); i.e., the population in patch 1 is undermatching the resource more than in patch 2 (as carrying capacity in patches 1 and 2 is the same), so small diffusion is still not favored in this scenario.
Theorem 3
If \(0<q_1<1\) and \(q_2>2q_1\), there exists some \(d^*=d^*(q_1,q_2)>0\) which is an evolutionarily singular strategy. Moreover, this strategy is not a CvSS, and both zero and infinity dispersal rates are local CvSSs.
It is interesting that zero diffusion rate emerges as a local CvSS under the assumptions of Theorem 3. Suppose the diffusion is zero or very small. On one hand, when \(q_1<1\), the drift out of patch 1 is small enough to allow the population to persist in patch 1, which is a source. On the other hand, \(q_2>2q_1\) drive the population in patch 2 down and that in patch 3 up. Hence, patch 2 becomes a source and patch 3 becomes a sink. Moreover, diffusion takes individuals out of patch 1, but due to the uneven drift rates those individuals are more likely to end up in patch 3 (the sink) than in patch 2 (the source). Hence, increasing diffusion will move individuals from source patch to sink patch. Thus, small diffusion can be favored in this case as the drift forces more individuals to move from patch 1 to 2 to reduce the mismatch in patch 2.
When both zero and infinity dispersal rates are local CvSSs, a natural question is whether two competing populations can coexist. Our next result answers this question partially but positively:
Theorem 4
Fix any \(k, D, q_2 \ge 1\). Then there exists some \(\delta >0\) such that for any \(d\in (0, \delta )\), \(q_1\in (-d, \delta )\), Model II has a globally asymptotically stable positive steady state, denoted by \((U^\delta , V^\delta )\), which satisfies \((U^\delta ,V^\delta )\rightarrow ({\hat{U}}, {\hat{V}})\) as \(d\rightarrow 0\) and \(q_1\rightarrow 0\), where
such that \(({\hat{V}}_2, {\hat{V}}_3)\) is the unique positive solution of the two-patch system
This result suggests that when the drift from patch 1 to patch 2 is very small, slow and fast diffusers can coexist in some interesting way: the slow diffuser will only occupy patch 1 and the fast diffuser is dominant in patches 2 and 3, but not in patch 1. Intuitively, the underlying mechanism for the coexistence is as follows: Consider the case \(d=0\) and \(q_1=0\) for the sake of clarity, so that patch 1 is disconnected from patches 2 and 3 for the species u. It turns out that, due to \(d=0\) and \(q_1=0\), the flux between patches 1 and 2 for the species v is also equal to zero. As a consequence, system (2) for patches 2 and 3 is reduced to a two-patch system for two competing species. It follows from previous work (Hamida 2017; Noble 2015) for two-patch models that the faster diffuser always out-competes the slower diffuser in patches 2 and 3, provided that \(q_2>0\). As patch 2 is at the upstream for the reduced two-patch model, the equilibrium distribution of species v, denoted by \(\hat{V_2}\), satisfies \(\hat{V_2}<k\); i.e., it undermatches the resource in patch 2. As there is no flux for species v between patches 1 and 2 and \(q_1=0\), the equilibrium distribution of species v at patch 1 is also equal to \(\hat{V_2}\), so that the equilibrium distribution of species u at patch 1 is given by \(k-\hat{V_2}>0\). The case of small \(d, q_1\) follows from a perturbation argument.
Note that \(q_1=0\) and small negative \(q_1\) are also covered by Theorem 4; the case of negative \(q_1\) applies to Model III.
2.3 Main Results of Model III
Model III assumes patch 1 is upstream, whereas patches 2 and 3 are downstream. Both species in patch 1 are transported to patches 2 and 3 by drift with rates \(q_1\) and \(q_2\), respectively. In this case we have the following ODE system for two competing species:
The dynamics of (5) is more subtle than those for Models I and II. We first consider the global dynamics of Model III.
Theorem 5
If \(q_1=q_2>0\), then the semi-trivial steady state \((U^*,0)\) is globally asymptotically stable for \(d>D\).
Theorem 5 seems to agree with previous results for two-patch models that the faster diffuser always out-competes the slower diffuser (Hamida 2017; Noble 2015). The biological intuition is that both downstream patches are sinks under the assumption of Theorem 5; see also Corollary 6 (in “Appendix C”). Hence, any mutant in the upstream patch with smaller diffusion rate will more likely be pushed to two downstream sinks and thus cannot invade when rare.
Next we consider the local dynamics of Model III.
Theorem 6
If \(q_1,q_2>0\), \(| q_2-q_1 | \le \frac{1}{2}\) and \(\frac{1}{\sqrt{2}}\le \frac{q_2}{q_1}\le \sqrt{2}\), then the semi-trivial steady state \((U^*,0)\) is locally stable for \(d>D\) and unstable for \(d<D\).
Theorem 6 implies that infinite diffusion rate is a global CvSS when two drift rates are comparable. This is in the same spirit as Theorem 5 since both downstream patches are still sinks under the assumptions of Theorem 6; see also Corollary 8 (in “Appendix C”). In contrast, our next result shows that if two drift rates are not comparable, both zero and infinite diffusion rates are local CvSS.
Theorem 7
If \(1<q_1+q_2<(q_1-q_2)^2\), then there exists some \(d^*=d^*(q_1,q_2)>0\) which is an evolutionarily singular strategy. Moreover, \(d^*\) is not a CvSS, and both zero and infinity dispersal rates are local CvSSs.
To see why zero diffusion can be a local CvSS under the assumptions of Theorem 7, first fix \(q_1\) and choose \(q_2\) large. This will drain almost all individuals in patch 1 to drift to patch 3, so that patch 3 becomes a sink patch due to overcrowding. Subsequently, diffusion induces a net flux of individuals from patch 2 to patch 1, so that the population in patch 2 undermatches the resource. Hence, any mutant with smaller diffusion rate can invade when rare by exploiting patch 2, which is a source patch. The same intuitive reasoning applies to the general situation: for the range of \(q_i\) in Theorem 7, our numerical results suggest that one of the two downstream patches is a sink while the other becomes a source patch, and a mutant with smaller diffusion rate can invade when rare in the downstream source patch.
3 Conclusions
In this section, we first summarize the main analytical results, and then we provide a single framework to unify the main results for three models. The main findings are as follows, along with some predictions (see Sect. 4 for further discussions):
-
If all drift rates are identical, then the faster dispersal rate is selected across all three-patch models in which the drift network do not form a closed cycle. A conjecture is that this result holds for n-patch river networks with uniform carrying capacity and identical drift rates, provided that the drift network is not divergence-free (a drift network with identical drift rates is called divergence-free if each individual patch has the same number of upstream and downstream patches);
-
For general drift rates, infinite diffusion rate is a local CvSS for all three models. Biologically this makes good sense as with sufficiently fast dispersal, the spatial distribution of the species approaches the ideal free distribution. However, there are some notable differences for three models: For Model I, the faster dispersal is always favored and thus infinity is a global CvSS. For Models II and III, the answers depend upon the drifts rates: for some ranges of drift rates, infinity is a global CvSS (same as Model I), while for other ranges of drift rates, there exists some intermediate diffusion rate which is a singular strategy so that infinity is a local CvSS but not a global one. A conjecture is that the infinite diffusion rate is a local CvSS for n-patch river networks with uniform carrying capacity and general drift rates;
-
For Models II and III, when a singular strategy (that is neither zero nor infinity) exists, it is not a local CvSS (Numerical simulation suggests that it is not an ESS either). In fact, bi-stability phenomenon happens, i.e., both infinity and zero diffusion rates are local CvSSs;
-
For Models II and III, when bi-stability occurs, it is possible for two competing populations with different dispersal rates to coexist, by varying the drift rates between patches. A conjecture is that any coexistence steady state for Models II–III, if exists, is globally stable.
Next, we provide a single framework to unify the main results for Models I–III. Our idea is to use a single system of ODEs to describe all three models. Without loss of generality, consider system (2) in the \(q_1-q_2\) plane, allowing the drift rates in system (2) to take both positive and negative values. That is, we divide the \(q_1-q_2\) plane into 4 quadrants. Then the first quadrant of Fig. 2 corresponds to Model II with nonnegative drift rates.
-
First quadrant: Theorem 2 implies that the faster diffuser always wins for \(q_1, q_2\) in the red region; for the blue region, Theorem 3 ensures the existence of an evolutionarily singular strategy, where both zero and infinity dispersal rates are local CvSSs.
-
Second quadrant: With \(q_1<0\) and \(q_2>0\) in system (2), the directed flows are from patch 2 to patches 1 and 3. Hence, this corresponds to Model III with patches 1 and 2 switched. Theorem 6 implies that the faster diffuser wins for \(q_1, q_2\) in the red region; for the blue region, Theorem 7 ensures the existence of an evolutionarily singular strategy, in which both zero and infinity dispersal rates are local CvSSs.
-
Third quadrant: With \(q_1<0\) and \(q_2<0\) in system (2), the directed flows are from patch 3 to 2, and from patch 2 to 1. Hence, this corresponds to Model II with patches 1 and 3 switched. Hence the red and blue regions are symmetric to those in the 1st quadrant with respect to the line \(q_1+q_2=0\).
-
Fourth quadrant: With \(q_1>0\) and \(q_2<0\) in system (2), the directed flows are from patches 1 and 3 to patch 2. Hence, this corresponds to Model I with patches 2 and 3 switched. Theorem 1 implies that the faster diffuser always wins for \(q_1, q_2\) in Fourth quadrant.
These discussions suggest that the \(q_1-q_2\) plane can also be divided into three colored regions as in Fig. 2: For the red region, the infinite diffusion rate is a global CvSS; in the blue region, both zero and infinite diffusion rates are local CvSSs; for the white region, numerical simulations suggest that the infinite diffusion rate is a local CvSS but might not be a global one, and the zero diffusion rate might not even be a local CvSS.
4 Discussion and Numerical Results for Four-patch Models
In this section, we will discuss possible extensions to n-patch river network models and raise some conjectures on the evolution of faster dispersal. We will also address the issue of the invasion of slowly diffusing populations and propose to study the coexistence of slow and fast diffusing competing populations.
4.1 Evolution of Fast Dispersal in n-patch Model
Theorems 1, 2 and 5 show that if \(q_1=q_2\), the faster diffusing population always wins the competition for Models I–III. In particular, infinity as a diffusion rate is a global CvSS for Model I and also for wider ranges of parameters in both Models II and III; see Theorems 2 and 6.
Consider the general n-patch river model, i.e.,
Here the connectivity matrix \(M:=(m_{ij})\) is assumed to be symmetric, \(m_{ij}=m_{ji}=1\) when two patches i and j are directly connected, \(m_{ij}=m_{ji}=0\) when they are not directly connected, and \(m_{ii}=-\sum _{j\not =i} m_{ij}\). The drift matrix \(Q:=(q_{ij})\) satisfies \(q_{ii}=-\sum _{j\not =i} q_{ji}\), \(q_{ij}>0\) when patches i and j are connected and the directed flow is from patch i to j, and \(q_{ij}=0\) otherwise. The case \(m_{ij}=1\) but \(q_{ij}=0\) refers to the scenario when patches i, j are directly connected but there is no directed passive flow in between. Note that under our assumptions, the dominant eigenvalue of Q is zero, with left eigenvector being \((1,\ldots ,1)\), i.e.,
The positive constant \(k_i\) is the carrying capacity of patch i.
Definition 1
We say that the drift matrix \(Q=(q_{ij})\) is divergence-free if its right eigenvector, corresponding to the zero eigenvalue, is given by \((1,\ldots ,1)^T\), i.e.,
Conjecture 1
If all positive drift rates are equal, the carrying capacity is the same for all patches, and the drift network \((q_{ij})\) is not divergence-free, then the faster disperser always wins.
For general drift rates, Theorem 1 shows that infinite dispersal rate is a global CvSS for Model I, while for Models II and III, Lemmas 15 (in “Appendix B”) and 36 (in “Appendix C”) find that infinity is always a local CvSS.
Conjecture 2
For n-patch model with general drift rates, when the drift matrix is not divergence-free and that the carrying capacity is the same for all patches, the infinite diffusion rate is always a local CvSS.
From the biological point of view, when \(k_i = k\) for all i, for a single species with sufficiently fast diffusion, its equilibrium will be close to \((k,\ldots , k)\), which is an ideal free distribution. Heuristically, if a strategy can help organisms reach the ideal free distribution at equilibrium, then the strategy is likely to be a local ESS and/or CvSS; see (Cantrell et al. 2007, 2012, 2017; Lou 2019). Again, we may need to exclude the exceptional case \((1,\ldots ,1)^T\) being a right eigenvector of matrix Q.
To support the above predictions for n-patch models, we performed some numerical simulations for the following four-patch models with the network topology as shown in Fig. 3:
For 4-patch model with topology Fig. 3a, our simulation results suggest that for any \({\tilde{q}}>0\), the faster diffusing species always wins the competition, and the conclusion is independent of drift rates. In particular, the faster dispersal rate is selected when \({\tilde{q}}=q\), which is consistent with Conjectures 1 and 2.
Figure 3b can also be viewed as Fig. 3a, c with \({\tilde{q}}=0\). For this special case, \((k,\ldots , k)\) is the unique positive equilibrium for the corresponding single species model. This gives an example of the exceptional case discussed earlier for n-patch models. As predicted earlier, our numerical simulations show that two populations with different dispersal rates coexist, i.e., the faster diffusing species does not win the competition in this exceptional case.
The PIP for 4-patch model with topology Fig. 3c is shown in Fig. 4. We take \(d\in [0,2]\) and \(D\in [0,2]\), and then we discretize the interval [0, 2] with the uniform step \(\varDelta =0.02\). The parameter values \((k_1,k_2,k_3,k_4)\) are set to be (7, 7, 7, 7) and \(q=1\), \({\tilde{q}}\) ranges from 0.01 to 2000. Our simulations (see Fig. 4) suggest more complicated dynamics: when \({\tilde{q}}\in (0, q]\), the faster diffusing population still wins. In particular, the fast dispersal rate is selected when \({\tilde{q}}=q\), which is consistent with Conjectures 1 and 2. However, for \({\tilde{q}}\) larger than q, there are two evolutionarily singular strategies, one is a local ESS and CvSS, and the other is neither an ESS not CvSS. Furthermore, the infinite diffusion rate remains as a local CvSS as predicted, while the zero diffusion rate is not a local CvSS.
4.2 Evolution of Slow Dispersal and Coexistence
Theorems 3 and 7 illustrate the existence of evolutionarily singular strategy for Models II and III, respectively. These singular strategies are not local CvSSs, and numerical simulations suggest they are not local ESSs either. In fact, Lemmas 16 (in “Appendix B”) and 37 (in “Appendix C”) show that zero diffusion rate can be a local CvSS for some parameter ranges in both Models II and III.
A natural question for general n-patch models is when zero diffusion rate can be a local CvSS. The analysis of Model III reveals that if there are more than one downstream patches, then it is possible for one of them to be a source patch, so that a mutant with slow diffusion rate can invade when rare in this source patch. For n-patch models it will be of interest to find sufficient conditions on the existence of some downstream source patch, by taking into account of the river network topology, so that slow diffusing populations can invade such source patch when rare.
For general n-patch models, when both zero and infinite diffusion rate are local CvSSs, it is natural to inquire whether slow and fast diffusers can coexist. It will be of interest to generalize Theorem 4 to n-patch models and to reveal the impact of network topology on the coexistence of competing species.
References
Cantrell RS, Cosner C, DeAngelis DL, Padrón V (2007) The ideal free distribution as an evolutionarily stable strategy. J Biol Dyn 1:249–271
Cantrell RS, Cosner C, Lam K-Y (2017) On resident-invader dynamics in infinite dimensional dynamical systems. J Differ Equ 263:4565–4616
Cantrell RS, Cosner C, Lou Y (2012) Evolutionary stability of ideal free dispersal strategies in patchy environments. J Math Biol 65:943–965
Cantrell RS, Cosner C, Lou Y, Schreiber S (2017) Evolution of natal dispersal in spatially heterogeneous environments. Math Biosci 283:136–144
Cosner C (2014) Reaction-diffusion-advection models for the effects and evolution of dispersal. Discrete Contin Dyn Syst Ser A 34:1701–1745
Dieckmann U, Law R (1996) The dynamical theory of coevolution: a derivation from stochastic ecological processes. J Math Biol 34:579–612
Dockery J, Hutson V, Mischaikow K, Pernarowski M (1998) The evolution of slow dispersal rates: a reaction-diffusion model. J Math Biol 37:61–83
Fagan WF (2002) Connectivity, fragmentation, and extinction risk in dendritic metapopulations. Ecology 83:3243–3249
Geritz SAH, Kisdi E, Meszena G, Metz JAJ (1998) Evolutionarily singular strategies and the adaptive growth and branching of the evolutionary tree. Evol Ecol 12:35–57
Golubitsky M, Hao W, Lam K-Y, Lou Y (2017) Dimorphism by singularity theory in a model for river ecology. Bull Math Biol 79:1051–1069
Grant EHC, Lowe WH, Fagan WF (2007) Living in the branches: population dynamics and ecological processes in dendritic networks. Ecol Lett 10:165–175
Hamida Y (2017) The evolution of dispersal for the case of two patches and two-species with travel loss, M.S. Thesis, Ohio State University
Hao WR, Lam K-Y, Lou Y (2021) Ecological and evolutionary dynamics in advective environments: critical domain size and boundary conditions. Discrete Contin Dyn Syst Ser B 26:367–400
Hastings A (1983) Can spatial variation alone lead to selection for dispersal? Theor Popul Biol 24:244–251
Hess P (1991) Periodic-parabolic boundary value problems and positivity, pitman research notes in mathematics series, vol 247. Longman/Wiley, Harlow
Hess P, Lazer AC (1991) On an abstract competition model and applications. Nonlinear Anal TMA 16:917–940
Hutson V, Mischaikow K, Polacik P (2001) The evolution of dispersal rates in a heterogeneous time-periodic environment. J Math Biol 43:501–533
Huang Q, Jin Y, Lewis MA (2016) R0 analysis of a benthic-drift model for a stream population. SIAM J Appl Dyn Syst 15:287–321
Hsu SB, Smith HL, Waltman P (1996) Competitive exclusion and coexistence for competitive systems on ordered Banach spaces. Trans Am Math Soc 348:4083–4094
Jiang HY, Lam K-Y, Lou Y (2020) Are two-patch models sufficient? The evolution of dispersal and topology of river network modules. Bull Math Biol 82:131
Jin Y, Hilker F, Steffler P, Lewis MA (2014) Seasonal invasion dynamics in a spatially heterogeneous river with fluctuating flows. Bull Math Biol 76:1522–1565
Jin Y, Huang Q, Steffler PM, Lewis MA (2019) Population persistence of a benthic-drift model in a depth-averaged river. AIMS Math 4:1768–1795
Jin Y, Lewis MA (2011) Seasonal influences on population spread and persistence in streams II: critical domain size. SIAM J Appl Math 71:1241–1262
Jin Y, Lewis MA (2012) Seasonal influences on population spread and persistence in streams: spreading speeds. J Math Biol 65:403–439
Jin Y, Peng R, Shi JP (2019) Population dynamics in river networks. J Nonlinear Sci 29:2501–2545
Lam K-Y, Lou Y, Lutscher F (2014) Evolution of dispersal in closed advective environments. J Biol Dyn 9(Suppl. 1):188–212
Lam K-Y, Lou Y, Lutscher F (2016) The emergence of range limits in advective environments. SIAM J Appl Math 76:641–662
Lam K-Y, Munther D (2016) A remark on the global dynamics of competitive systems on ordered Banach spaces. Proc Am Math Soc 144:1153–1159
Levin SA, Cohen D, Hastings A (1984) Dispersal strategies in patchy environments. Theor Popul Biol 26:165–191
Lou Y, Lutscher F (2014) Evolution of dispersal in open advective environments. J Math Biol 69:1319–1342
Lou Y (2019) Ideal free distribution in two patches. J Nonlinear Model Anal 2:151–167
Lou Y, Xiao D, Zhou P (2016) Qualitative analysis for a Lotka–Volterra competition system in advective homogeneous environment. Discrete Contin Dyn Syst Ser A 36:953–969
Lou Y, Zhao X-Q, Zhou P (2019) Global dynamics of a Lotka–Volterre competition-diffusion-advection system in heterogeneous environments. J Math Pure Appl 121:47–82
Lou Y, Zhou P (2015) Evolution of dispersal in advective homogeneous environment: the effect of boundary conditions. J Differ Equ 259:141–171
Lutscher F, Lewis MA, McCauley E (2006) Effects of heterogeneity on spread and persistence in rivers. Bull Math Biol 68:2129–2160
Lutscher F, McCauley E, Lewis MA (2007) Spatial patterns and coexistence mechanisms in systems with unidirectional flow. Theor Popul Biol 71:267–277
Lutscher F, Pachepsky E, Lewis MA (2005) The effect of dispersal patterns on stream populations. SIAM Rev 47:749–772
Maynard-Smith J, Price GR (1973) The logic of animal conflict. Nature 246:15–18
McPeek MA, Holt RD (1992) The evolution of dispersal in spatially and temporally varying environments. Am Nat 140:1010–1027
Meyer C (2000) Matrix analysis and applied linear algebra, Society for Industrial and Applied Mathematics (SIAM), vol xii+718. Philadelphia, PA
Noble L (2015) Evolution of dispersal in patchy habitats. Ohio State University Ph.D Thesis
Pachepsky E, Lutscher F, Nisbet RM, Lewis MA (2015) Persistence, spread and the drift paradox. Theor Popul Biol 67:61–73
Ramirez JM (2012) Population persistence under advection-diffusion in river networks. J Math Biol 65:919–942
Samia Y, Lutscher F, Hastings A (2015) Connectivity, passability and heterogeneity interact to determine fish population persistence in river networks. J R Soc Interface 12:20150435
Sarhad J, Carlson R, Anderson KE (2014) Population persistence in river networks. J Math Biol 69:401–448
Smith H (1995) Monotone dynamical system. Dynamical system. An introduction to the theory of competitive and cooperative systems, Math. Surveys Monogr., vol. 41, American Mathematical Society, Providence, RI
Speirs DC, Gurney WSC (2001) Population persistence in rivers and estuaries. Ecology 82:1219–1237
Vasilyeva O (2017) Competition of multiple species in advective environments. Bull Math Biol 79:1274–1294
Vasilyeva O (2019) Population dynamics in river networks: analysis of steady states. J Math Biol 79:63–100
Vasilyeva O, Lutscher F (2011) Population dynamics in rivers: analysis of steady states. Can Appl Math Q 18:439–469
Vasilyeva O, Lutscher F (2012) Competition of three species in an advective environment. Nonlinear Anal RWA 13:1730–1748
Vasilyeva O, Lutscher F (2012) Competition in advective environments. Bull Math Biol 74:2935–2958
Wang Y, Shi JP (2019) Persistence and extinction of population in reaction-diffusion-advection model with weak Allee effect growth. SIAM J Appl Math 79:1293–1313
Wang Y, Shi JP, Wang JF (2019) Persistence and extinction of population in reaction-diffusion-advection model with strong Allee effect growth. J Math Biol 78:2093–2140
Wang Y, Shi JP (2020) Dynamics of a reaction-diffusion benthic-drift model with strong Allee effect growth. J Differ Equ 269:7605–7642
Zhang B, Lam K-Y, Ni W-M, Collins KM, Fu Z, Zhai L, Lou Y, DeAngelis DL, Hastings A (2020) Directed movement changes coexistence outcomes in heterogeneous environments, submitted 28pp
Zhao X-Q, Zhou P (2016) On a Lotka–Volterra competition model: the effects of advection and spatial variation. Calc Var Partial Differ 55:73
Zhou P (2016) On a Lotka–Volterra competition system: diffusion vs advection. Calc Var Partial Differ 55:137
Acknowledgements
The authors sincerely thank Prof. Mark Lewis and the referees for their suggestions which help improve the presentation of the manuscript. HYJ is partially supported by the NSFC Grant No. 11571364. KYL and YL are partially supported by the NSF grant DMS-1853561. YL is also supported by Institute of Modern Analysis-A Frontier Research Center of Shanghai and the NICE Visiting Program.
Author information
Authors and Affiliations
Corresponding author
Additional information
Publisher's Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Appendices
Appendix A: The Global Dynamics of Model I
In this section, we mainly study the global dynamics of Model I. By the monotone dynamical system theory (Hess and Lazer 1991, Theorem 1.5) (see also Hsu et al. 1996; Lam and Munther 2016; Smith 1995), in order to show the global stability of the semi-trivial steady state \((U^*,0)\), we need to show the linear instability of the other semi-trivial steady state \((0,V^*)\) and the non-existence of positive steady state of (1).
By replacing \(u_i\) and \(v_i\) by \(ku_i\) and \(kv_i\), for all i, we may assume without loss of generality that \(k=1\). Henceforth, we will prove our theorems concerning Models I, II and III only for the case \(k=1\).
1.1 A.1. Preliminary Estimates on Nonnegative, Non-trivial Steady States
In this subsection, we consider nonnegative and non-trivial solutions of Model I. After setting \(k=1\), they satisfy the following system:
When \(d,D>0\), it follows from irreducibility of system (6) that there are at most three types of nonnegative and non-trivial solutions, namely: semi-trivial equilibria \((U^*,0)\), \((0,V^*)\) and positive solutions for which \(U_i>0\), \(V_i>0\), \(i=1,2,3\). Hence, for the simplicity of notation, in this subsection we may denote all of these different types of solutions as (U, V), with the understanding that there are only three possibilities for all \(i=1,2,3\): \(U_i>0\) and \(V_i=0\), or \(U_i=0\) and \(V_i>0\), or \(U_i>0\) and \(V_i>0\). We shall establish some a prior estimates of (U, V).
Lemma 1
Assume \(q_1,q_2\ge 0 \), \((q_1,q_2)\not = (0,0)\) and \(d,D>0\).
-
(i)
If \(q_1\ge q_2\), then \(U_1\le U_2\) and \(V_1\le V_2\).
-
(ii)
If \(q_1\le q_2\), then \(U_1\ge U_2\) and \(V_1\ge V_2\).
In particular, if \(q_1=q_2\), then \(U_1=U_2\) and \(V_1=V_2\).
Proof
For part (i) we shall prove \(U_1\le U_2\) only, as \(V_1\le V_2\) follows from a similar argument. We argue by contradiction: if not, assume that there exist some \(q_1 \ge q_2\) and a nonnegative, non-trivial solution of (6) such that \(U_1> U_2\). By the first and second equations of (6), we get
so that \(-d-q_i+1-U_i-V_i<0\), \(i=1,2\). Due to \(U_1> U_2\), (7) implies
and thus
which together with \(q_1\ge q_2\) implies that \(U_1+V_1< U_2+V_2\). This implies \((V_1,V_2)\ne (0,0)\) and \(V_2>V_1>0\).
Therefore, similar to (8), the equations of \(V_1\) and \(V_2\) from (6) imply
which implies \( q_2-q_1<(U_1+V_1)-(U_2+V_2)\). This, however, contradicts (9). Therefore, \(U_1\le U_2\) holds. This proves (i). The conclusion in (ii) follows by exchanging patches 1 and 2. \(\square \)
Lemma 2
Assume \(q_1,q_2\ge 0 \) and \((q_1,q_2)\not = (0,0)\). For any \(d,D>0\),
-
(i)
if \(q_1\ge q_2\), then \(U_1+V_1<1<U_3+V_3\);
-
(ii)
if \(q_1\le q_2\), then \(U_2+V_2<1<U_3+V_3\).
In particular, when \(q_1=q_2\), \(U_1+V_1=U_2+V_2<1<U_3+V_3\).
Proof
As the proofs of (i) and (ii) are similar, we only prove (i). Firstly, we show
We argue by contradiction: If not, there exist some \(q_1\ge q_2\) and a nonnegative non-trivial solution such that \(U_1+V_1 \ge 1\); i.e., \(1-U_1-V_1 \le 0\). We claim that
Without loss of generality, we may assume \((U_i)\) is non-trivial, so that the first equation of (6) implies that \(U_3 \ge \frac{d+q_1}{d}U_1>U_1\). If \((V_i)\) is trivial then (11) holds. If not, then applying similar reasoning to the fourth equation of (6), we also get \(V_3 \!\ge \! \frac{D+q_1}{D}V_1\!>\!V_1\). This proves (11) for any nonnegative solutions. Due to \(q_1\ge q_2\), we get \(U_2+V_2\ge U_1+V_1 \ge 1\) by Lemma 1, thus
However, adding the equations of \(U_1, U_2, U_3\) in (6), we get
This is a contradiction. This proves (10).
Next, we claim
We again argue by contradiction and assume that there exist some \(q_1 \ge q_2\) and a nonnegative non-trivial solution such that \((U_i)\) is non-trivial and \(U_3+V_3 \le 1\). From the third equation of (6), we obtain
\(d(U_1+U_2-2U_3)+q_1U_1+q_2U_2 \le 0\),
which together with \(U_1\le U_2\) (Lemma 1) implies
Hence \(U_1 < U_3\). If \((V_i)\) is non-trivial, then we can get \(V_1<V_3\) by the same method. Therefore, \(U_1+V_1<U_3+V_3 \le 1\). In view of (12), we have also
Using the second equation of (6), we get \(U_3>U_2\), which implies \(V_3<V_2\). Hence, by the equation of \(V_2\) in (6), we get \(1-U_2-V_2>0\), i.e., \(U_2+V_2<1\), which is again impossible. Hence, we proved (13). The proof of (i) is completed. \(\square \)
Lemma 3
Assume \(q_1,q_2\ge 0 \), \((q_1,q_2)\not = (0,0)\), and \(d,D>0\).
-
(i)
If \(q_1\ge q_2\), and \((U_1,U_2,U_3)\ne (0,0,0)\), then \(U_1\le U_2<U_3\).
-
(ii)
If \(q_1\ge q_2\), and \((V_1,V_2,V_3)\ne (0,0,0)\), then \(V_1\le V_2<V_3\).
-
(iii)
If \(q_1\le q_2\), and \((U_1,U_2,U_3)\ne (0,0,0)\), then \(U_2\le U_1<U_3\).
-
(iv)
If \(q_1\le q_2\), and \((V_1,V_2,V_3)\ne (0,0,0)\), then \(V_2\le V_1 < V_3\).
In particular, if \(q_1=q_2\) then every positive equilibrium satisfies \(U_1=U_2<U_3\), \(V_1=V_2<V_3\).
Proof
We only prove (i) as (ii)-(iv) follow from a similar argument. To this end, we assume \((U_1,U_2,U_3)\ne (0,0,0)\) and prove \(U_1\le U_2<U_3\). From Lemma 1, it suffices to prove \(U_3>U_2\). Suppose to the contrary that \(q_1\ge q_2\) and there is a nonnegative solution such that \((U_1,U_2,U_3)\ne (0,0,0)\) and \(U_3 \le U_2\). We claim that \(U_3+V_3 \le U_2 + V_2\). This is immediate if \((V_i)\) is trivial. If \((V_i)\) is non-trivial, then the second equation of (6) implies
By way of the fifth equation of (6), we obtain \(V_3\le V_2\), which again implies \(U_3+V_3 \le U_2 + V_2\).
By Lemma 2, we have \(1-U_2-V_2\le 1-U_3-V_3<0\). Again using the second equation of (6), we get \(U_3>U_2\), a contradiction. The assertions (ii)-(iv) are analogous, by exchanging the role of U and V or the patches one and two. \(\square \)
Lemma 4
Assume \(q_1,q_2\ge 0\), \((q_1,q_2)\not = (0,0)\), and \(d,D>0\). Then we have
Proof
By exchanging the two species if necessary, we may assume without loss of generality that \((U_i)\) is non-trivial. Adding the equations of \(U_i\), \(i=1,2,3\), in (6), we get
If \(q_1\ge q_2\), applying (15), \(U_1+V_1<1<U_3+V_3\) (Lemma 2) and \(U_1\le U_2<U_3\) (Lemma 3). Hence
that is, (14) holds.
The proof of the case \(q_1<q_2\) is similar and thus omitted. \(\square \)
Lemma 5
Assume \(q_1,q_2\ge 0 \), \((q_1,q_2)\not = (0,0)\) and \(d,D>0\).
-
(i)
If \((U_i)\) is non-trivial, then \(-q_1+1-U_1-V_1<0\).
-
(ii)
If \((V_i)\) is non-trivial, then \(-q_2+1-U_2-V_2<0\).
Proof
By Lemma 3 and the first and second equation of (6), we get
This proves (i). The proof of (ii) is omitted. \(\square \)
From the above results, we can establish the non-existence of positive solution of (6).
Lemma 6
Assume \(q_1,q_2\ge 0 \), \((q_1,q_2)\not = (0,0)\), and \(d,D>0\) satisfy \(d\not = D\). Then (6) has no positive solution.
Proof
If there exists a positive solution (U, V) with \(U_i>0\) and \(V_i>0\), then we can rewrite (6) as \(E_0(U_1,U_2,U_3)^T=(0,0,0)^T\) and \(F_0(V_1,V_2,V_3)^T=(0,0,0)^T\), where the matrices \(E_0\) and \(F_0\) are defined as
Direct calculation gives
and
for some constant P depending only on \(U_i, V_i\) (\(i=1,2,3\)) and \(q_j\) (\(j=1,2\)). Multiplying the above two equations by D, d, respectively, and subtracting the resulting equations, in view of \(D \ne d\), we obtain
From Lemmas 2 and 5, it follows that the right-hand side of (17) is negative. However, the left-hand side of (17) is positive, as implied by Lemma 4. This contradiction finishes the proof. \(\square \)
1.2 A.2. The Global Stability of Semi-trivial Steady State
In this subsection, we mainly establish Theorem 1. We first study the linear instability of the semi-trivial steady state \((0,V^*):=(0,0,0,V^*_1,V^*_2,V^*_3)\) for Model I, where \(V^*\) satisfies
with matrix \(F_1\) given by
The linear instability of \((0,V^*)\) is determined by the sign of the principal eigenvalue, denoted as \(\varLambda _1\), of the eigenvalue problem
where matrix \(E_1\) is given by
Note that \(\varLambda _1=\varLambda _1(d,D)\) depends on D by way of \(V^*_i\). We first study the sign of \(\varLambda _1\) for the case \(q_1=q_2=q\). From Jiang et al. (2020), we recall the following two results concerning \(\varLambda _1\):
Proposition 1
(Jiang et al. 2020, Proposition 1) Suppose \(q_1=q_2=q>0\). Then the derivative of \(\varLambda _1\) with respect to d, at d=D, is given by
Proposition 2
(Jiang et al. 2020, Proposition 2) Assume \(q_1=q_2=q>0\) and \(V_1^*+V_2^*+V_3^*\ne 3\). Then \(\mathrm{det}(E_1)=0\) if and only if either \(d=D\), or \(d=F(D)\), where function F is given by
Corollary 1
Assume \(q_1=q_2=q>0\). For any \(d, D>0\), we have \(\frac{\partial \varLambda _1}{\partial d}{\big |_{d=D}} < 0\).
Proof
By Lemma 3, we have \(V_3^*-V_1^*>0\) and \(V_3^*-V_2^*>0\). Using (i) of Lemma 1 and Lemma 2, we get \(V_2^*=V_1^*<1\), which together with the equations of \(V_1^*\) and \(V_2^*\) in (18) yields
Therefore, the right-hand side of (20) is negative. \(\square \)
Corollary 2
Assume \(q_1=q_2=q>0\). Then for any \(d, D>0\), the right-hand side of (21) is strictly negative.
Proof
This lemma is a direct consequence of Lemmas 2, 4 and 5. \(\square \)
Theorem 8
Assume \(q_1=q_2=q>0\). Then for any \(d, D>0\), we have
Proof
Since the right-hand side of (21) is strictly negative (by Corollary 2), Proposition 2 says that \(\varLambda _1(d,D)=0\) if and only if \(d=D\). Therefore, by Corollary 1 and the continuity of \(\varLambda _1\), \(\varLambda _1(d,D)>0\) holds for \(D>d>0\) and \(\varLambda _1(d,D)<0\) holds for \(0<D<d\). \(\square \)
Next, we consider the sign of \(\varLambda _1\) for any \(q_1,q_2\ge 0 \) and \((q_1,q_2)\not = (0,0)\).
Lemma 7
For any \(q_1,q_2\ge 0 \) and \((q_1,q_2)\not = (0,0)\), we have \(\varLambda _1(d,D)<0\) for \(d>D>0\).
Proof
Fix \(d> D >0\). By Theorem 8, if \(q_1=q_2=q\), then \(\varLambda _1(d,D)<0\). Then by the continuity of \(\varLambda _1\) in \(q_1,q_2\), it is sufficient to show \(\varLambda _1 \ne 0\) for any \(q_1 \ne q_2\). If not, we assume there exist some \(q_1\ne q_2\) such that \(\varLambda _1 = 0\). By direct calculation, we get
By (18), we also get
Here M depends on \(V_i^* (i=1,2,3), q_1\) and \(q_2\). Multiplying the above two equations by d, D, respectively, and subtracting them, we obtain
Due to \(d>D\), we have
Lemmas 2 and 5 imply that the right-hand side of (22) is negative. However, the left-hand side of (22) is positive, as implied by Lemma 4. This contradiction finishes the proof. \(\square \)
Proof of Theorem 1
Fix \(d>D\). By Lemmas 6 and 7, \((0,V^*)\) is linearly unstable, and Model I has no positive equilibria. By the theory of monotone dynamical systems (Hess and Lazer 1991, Theorem 1.5), \((U^*,0)\) is globally asymptotically stable among all nonnegative, non-trivial solutions of (1).\(\square \)
Appendix B: The Dynamics of Model II
In this section, we mainly study the dynamics of Model II and establish Theorems 2 to 4.
1.1 B.1. Preliminary Estimates on Nonnegative, Non-trivial Steady States
In this subsection, we study the nonnegative and non-trivial solutions of the system
Once again, we set \(k=1\) and observe that system (23) has at most three types of nonnegative and non-trivial solutions, that is, semi-trivial solution \((U^*,0),(0,V^*)\) and positive solutions for which \(U_i>0, V_i>0\) (\(i=1,2,3\)). In the following, we denote by (U, V) a nonnegative non-trivial solution of (23), in which \(U_i>0\) and \(V_i=0\), or \(U_i=0\) and \(V_i>0\), or \(U_i>0\) and \(V_i>0\) for all \(i=1,2,3\). We shall establish a priori estimates of the nonnegative and non-trivial solutions (U, V).
Lemma 8
For any \(d,D>0\) and \(q_1,q_2>0\), we have \(U_1+V_1<1<U_3+V_3\).
Proof
We will prove this conclusion for the case \(U_i>0\). The case \(U_i \equiv 0\) and \(V_i>0\) can be proved by a similar argument.
Step 1: We prove \(U_1+V_1<1\). We argue by contradiction and assume that there exist some \(q_1,q_2\) such that \(U_1+V_1\ge 1\). Then by the first equation of (23), we have \(d(U_2-U_1)-q_1U_1 \ge 0\), i.e., \(U_2 \ge \frac{d+q_1}{d}U_1 >U_1\). Following from the similar argument, we also get \(V_i=0\) for all i, or \(V_2>V_1\). Hence \(U_2+V_2>U_1+V_1 \ge 1\).
Clearly, we have \(U_3+V_3 \ge 1\). If not, assume that \(U_3+V_3 <1\) for some \(q_1,q_2\). By the equations of \(U_3\) and \(V_3\), we have \(U_3>\frac{d+q_2}{d}U_2>U_2\) and \(V_3 \ge \frac{D+q_2}{D}V_2 \ge V_2\) (where equality holds in case \(V_i=0\) for all i), which imply \(U_3+V_3>U_2+V_2 >1\), a contradiction.
Therefore, we get
However, adding the equations of \(U_i\), \(i=1,2,3\), in (23), we find that the left-hand side of the above inequality is equal to zero. This is a contradiction. Hence, \(U_1+V_1<1\) holds.
Step 2: We show \(U_3+V_3>1\). If not, we assume \(U_3+V_3 \le 1\) for some \(q_1, q_2\). By the equations of \(U_3\) and \(V_3\) in (23), we deduce that \(U_2<U_3\) and \(V_2\le V_3\) (with equality holds if \(V_i =0\)). Thus \(U_2+V_2<U_3+V_3 \le 1\), which together with \(U_1+V_1<1\) implies
Similarly, adding the equations of \(U_i\), \(i=1,2,3\), in (23), we find that the left-hand side of the above inequality is equal to zero. This is a contradiction. \(\square \)
Lemma 9
For any \(d,D>0\), \(q_1, q_2>0\), it holds that
Proof
We consider two cases:
Case I. Either \(U_3> U_2\) or \(V_3> V_2\). Without loss of generality, assume \(U_3> U_2\), so that \(U_i >0\) for all i. Adding equations of \(U_1\) and \(U_2\) in (23), we have
It is easy to see that (24) follows from (25), \(U_3\ge U_2\) and \(U_1+V_1<1\) (Lemma 8).
Case II. \(U_3\le U_2\) and \(V_3\le V_2\). For this case,
The last inequality follows from Lemma 8. This completes the proof. \(\square \)
Lemma 10
Let \(d,D>0\), and either \(q_1\ge 1\) or \(\frac{q_2}{2}<q_1<1\).
-
(i)
Either \(U_i \equiv 0\) or \(U_1 < U_2\).
-
(ii)
Either \(V_i \equiv 0\) or \(V_1 < V_2\).
Proof
We only prove (i), as (ii) follows in a completely analogous manner. Assume \(U_i >0\) for all i, we need to show \(U_1 < U_2\). Obviously, for \(q_1 \ge 1\), this conclusion is true from the first equation of (23).
Next, assume to the contrary that there exist some \(\frac{q_2}{2}<q_1<1\) such that \(U_1 \ge U_2\). By the first equation of (23), \(-q_1+1-U_1-V_1\ge 0\), i.e., \(U_1+V_1 \le 1-q_1\). Hence \(U_1+V_1<1-\frac{q_2}{2}\). Using the 4th equation of (23), we get \(V_2 \le V_1\). So we have
From the second equation of (23), we get
which together with \(U_1 \ge U_2\) indicates \(d(U_3-U_2)+(q_1- \frac{q_2}{2})U_2<0\). Since \(q_1>\frac{q_2}{2}\), we have \(U_3<U_2\).
We claim that \(U_2 + V_2 > U_3 + V_3\). If \(V_i \equiv 0\), then it follows from \(U_3<U_2\) and we are done. If \(V_i >0\) for all i, then we can repeat the above argument to show that \(V_3 < V_2\). Using Lemma 8, we have \(U_2 + V_2> U_3 + V_3 >1\). This is in contradiction with (26). \(\square \)
Lemma 11
For any \(d,D>0\), if \(q_1 \ge 1\) or \(\frac{q_2}{2}<q_1<1\), then
Proof
Adding all six equations of (23), we have
We consider two cases:
Case I. \(U_2+V_2\le U_3+V_3\). For this case, by \(U_3+V_3>1\) we have
By Lemma 10, we have \(U_1+V_1<U_2+V_2\). This together with \(U_1+V_1<1\) implies
It is easy to see that (27) follows directly from (28), (29) and (30).
Case II. \(U_2+V_2\ge U_3+V_3\). For this case, by \(U_3+V_3>1>U_1+V_1\) (by Lemma 8) we have
Since \(U_2 + V_2 \ge U_3 + V_3>1 > U_1 + V_1\), we can similarly derive
It is easy to see that (27) follows directly from (28), (31) and (32). Note that the above reasoning is valid also when \(U_1=U_2=U_3=0\) or \(V_1=V_2=V_3=0\). \(\square \)
Note that the above results are valid when \(q_1=q_2>0\). The following result implies that (23) has no positive solution when \(q_1 \ge 1\) or \(\frac{q_2}{2}<q_1<1\).
Corollary 3
If \(q_1 \ge 1\) or \(\frac{q_2}{2}<q_1<1\), then system (23) has no positive solutions for \(d\not =D\).
Proof
We argue by contradiction. If there exists some positive solution, denoted by (U, V), for (23). Direct calculation, as in the proof of Lemma 6, gives
Due to \(U_1<U_2\), the equation of \(U_1\) implies \(-q_1+1-(U_1+V_1)<0\). By Lemma 8, \(1-(U_3+V_3)<0.\) By Lemma 9, we see \(-q_2+1-(U_2+V_2)<0.\) Hence, the right-hand side of (33) is negative. However, the left-hand side of (33) is positive from Lemma 11, which is a contradiction. \(\square \)
1.2 B.2. The Global Dynamics of Model II When \(q_1 \ge 1\) or \(\frac{q_2}{2}<q_1<1\)
In this subsection, we shall show that the faster diffuser always wins when \(q_1 \ge 1\) or \(\frac{q_2}{2}<q_1<1\). We first study the local instability of \((0,V^*):=(0,0,0,V_1^*,V_2^*,V_3^*)\), as determined by the sign of the principal eigenvalue \(\varLambda _2\) of the eigenvalue problem
where matrix \(E_2\) is given by
Proposition 3
When \(d=D\), the derivative of \(\varLambda _2\) with respect to d satisfies
Proof
Differentiate (34) with respect to d, we get
where \(\varphi _i'=\frac{\partial \varphi _i}{\partial d}\), \(i=1,2,3\). Note that when \(d=D\),
and when \(d=D\), we may choose
Set \(d=D\) in (36) and multiplying it by \(\begin{pmatrix}(D+q_1)V_1^*, DV_2^*, \frac{D^2}{D+q_2}V_3^*\end{pmatrix}\), using (37), (38) and \(\varLambda _2(D,D)= 0\), we obtain (35). This completes the proof. \(\square \)
1.2.1 B.2.1. The Sign of \(\varLambda _2\) When \(q_1=q_2\)
Our goal in this subsection is to determine the sign of \(\varLambda _2\) when \(q_1=q_2\). We first recall the following result:
Proposition 4
(Jiang et al. 2020, Proposition 4) Assume \(q_1=q_2=q\) and \({V_1^*+V_2^*+V_3^*}\ne 3\). Then \(\mathrm{det}(E_2)=0\) if and only if either \(d=D\), or \(d=F(D)\), where F(D) is given by (21).
Lemma 12
Suppose \(d,D>0\) and \(q_1=q_2=q>0\), then \(V_2^*<V_3^*\).
Proof
We argue by contradiction. If not, we assume there exist some \(d,D,q>0\) such that \(V_i >0\) for all i and \(V_2\ge V_3\). By the sixth equation of (23), we get
which implies that \(V_3^* \ge 1+q >1\).
Therefore, \(V_2^*\ge V^*_3 >1\). Hence,
where we also used the assumption \(V_2^*\ge V_3^*\) and \(V_1^*<V_2^*\) (Lemma 10). This is in contradiction with the fifth equation of (23). \(\square \)
Corollary 4
Suppose \(q_1=q_2=q>0\), then for any \(d,D>0\), the quantity F(D) given in (21) is strictly negative.
Proof
By Lemma 10 and the first equation of (23), we get \(-q+1-V_1^*<0.\) Using also Lemma 11, the quantity F(D), given in (21), is strictly negative. \(\square \)
Lemma 13
Suppose \(q_1=q_2=q>0\), then for any \(d,D>0\), we have \(\frac{D+q}{D}V_1^*>V_2^*>\frac{D}{D+q}V_3^*\).
Proof
By the fourth equation of (23) and Lemma 8, we get \(d(V_2^*-V_1^*)-qV_1^*<0\), which implies \(V_2^*<\frac{D+q}{D}V_1^*\). Similarly, by the sixth equation of (23) and Lemma 8, we have \(d(V_2^*-V_3^*)+qV_2^*>0\), i.e., \(V_2^*>\frac{D}{D+q}V_3^*\). \(\square \)
Corollary 5
Suppose \(q_1=q_2=q>0\), then for any \(d,D>0\), we have \(\frac{\partial \varLambda _2}{\partial d}\Big |_{d=D}<0\).
Proof
If \(q_1=q_2=q\), (35) can be rewritten as
Note that
which together with Lemmas 10, 12 and 13 yields the conclusion. \(\square \)
Theorem 9
Assume \(q_1=q_2=q>0\). Then for any \(d,D>0\), we have
Proof
Since the quantity F(D), which is given in (21), is strictly negative (by Corollary 4), Proposition 4 says that \(\varLambda _2(d,D)=0\) if and only if \(d=D\). Therefore, by Corollary 5 and the continuity of \(\varLambda _2\), \(\varLambda _2(d,D)>0\) holds for \(D>d>0\) and \(\varLambda _2(d,D)<0\) holds for \(0<D<d\). \(\square \)
1.2.2 B.2.2. The Proof of Theorem 2
In this subsection, we will study the global dynamics of Model II for \(q_1 \ge 1\) or \(\frac{q_2}{2}<q_1<1\).
Lemma 14
If \(q_1 \ge 1\) or \(\frac{q_2}{2}<q_1<1\), \(\varLambda _2(d,D)<0\) for \(d>D\).
Proof
The proof is similar as that of Lemma 7. It follows from Theorem 9 that, if \(q_1=q_2\), then \(\varLambda _2(d,D)<0\) for \(d>D\). Since \(\varLambda _2\) is continuous with respect to parameters \(q_1,q_2\), it suffices to show that \(\varLambda _2 \ne 0\) for any \(q_1\not =q_2\) and \(q_1 \ge 1\) or \(\frac{q_2}{2}<q_1<1\). We argue by contradiction and assume that there exists some q satisfying the assumptions such that \(\varLambda _2 = 0\). By proceeding similarly as in Lemma 7, we derive (22) again. Note that \(3-V_1^*-V_2^*-V_3^*>0\) holds, which implies the left-hand side of (22) is positive. Using \(V_1^*<V_2^*\) and the first equation of (23), we deduce \(-q_1+1-V_1^*<0\). By Lemma 9, \(-q_2+1-V_2^*<0\). These together with \(V_3^*>1\) imply the right-hand side of (22) is negative, which is a contradiction. \(\square \)
Proof of Theorem 2
By Corollary 3 and Lemma 14, the equilibrium \((0,V^*)\) is linearly unstable and Model II has no positive equilibria. By the theory of monotone dynamical systems (Hess and Lazer 1991, Theorem 1.5), the equilibrium \((U^*,0)\) is globally asymptotically stable. \(\square \)
1.3 B.3. Existence of Evolutionarily Singular Strategy
In this subsection, we consider the existence of evolutionarily singular strategy and establish Theorem 3. The linear stability of the semi-trivial steady state, \((U^*,0):=(U_1^*,U_2^*,U_3^*,0,0,0)\), is determined by the sign of the principal eigenvalue \({\tilde{\varLambda }}_2\) of the eigenvalue problem
where matrix \(F_2\) is given by
Note that \(U_i^*\), \(i=1,2,3,\) satisfy
By exchanging the role of the two species, we can rewrite Proposition 3 as follows:
Proposition 5
When \(D=d\), the derivative of \({\tilde{\varLambda }}_2\) with respect to D satisfies
Lemma 15
For any \(q_1,q_2>0\), we have \( \frac{\partial {\tilde{\varLambda }}_2}{\partial D}(d,d)<0\) for sufficiently large d.
Proof
Set
By (23), we can rewrite (42) as
Note that \((U_1^*,U_2^*,U_3^*) \rightarrow (1,1,1)\) as \(d \rightarrow \infty \). As \((U_1^*, U_2^*, U_3^*)\) is the unique stable positive solution of (40), it can be shown that it is smooth at \(d=\infty \) so that we can expand \(U_i\) as
To determine \({\tilde{U}}_i\), we substitute the above expansion of \(U^*_i\) into the first and third equation in (40) to get
By adding the first three equations of (40), we obtain \(\sum _{i=1}^3 U^*_i(1-U^*_i)=0\), from which we deduce \(\sum _{i=1}^3 {\tilde{U}}_i = 0\). Combining this with (44), we obtain
Having determined \({\tilde{U}}_i\), we may substitute \(U^*_i = 1 + {\tilde{U}}_i/d + O({1}/{d^2})\) into (43) to get
Therefore, by Proposition 5, we have \( \frac{\partial {\tilde{\varLambda }}_2}{\partial D}(d,d)<0\) for \(d \gg 1\). \(\square \)
Lemma 16
If \(0<q_1<1\) and \(q_2>2q_1\), we have \( \frac{\partial {\tilde{\varLambda }}_2}{\partial D}(d,d)>0\) for sufficiently small d.
Proof
When \(d \rightarrow 0\), we have \(U_1^* \rightarrow {\overline{U}}_1:= 1-q_1\) and, passing to a subsequence if necessary, \(U_2^*\rightarrow {\overline{U}}_2\) for some nonnegative \({\overline{U}}_2\). We claim that if \(2q_1<q_2\), then \({\overline{U}}_2<{\overline{U}}_1\). If not, we assume for some \(2q_1<q_2\), \({\overline{U}}_2\ge {\overline{U}}_1\). By the equation of \(U_2^*\) and let \(d \rightarrow 0\),
Then we have \( q_1{\overline{U}}_2-q_2{\overline{U}}_2+{\overline{U}}_2(1-{{\overline{U}}_2})\ge 0\), which implies that \({\overline{U}}_2 \le 1+q_1-q_2\). Therefore, \(1+q_1-q_2 \ge 1-q_1\), i.e., \(2q_1 \ge q_2\). This contradiction shows that \({\overline{U}}_2<{\overline{U}}_1\).
Note that \(M \rightarrow q_1{\overline{U}}_1({\overline{U}}_2-{\overline{U}}_1)<0\) as \(d\rightarrow 0\), where M is given by (42). Note that \(0<q_1<1\). Hence, for sufficiently small d, we have \( \frac{\partial {\tilde{\varLambda }}_2}{\partial D}(d,d)>0\). \(\square \)
Proof of Theorem 3
Since \(d\mapsto \frac{\partial {\tilde{\varLambda }}_2}{\partial D}(d, d)\) is analytic, all the roots are discrete. By Lemmas 15 and 16, \(\frac{\partial {\tilde{\varLambda }}_2}{\partial D}(d, d)<0\) for \(d\gg 1\) and \(\frac{\partial {\tilde{\varLambda }}_2}{\partial D}(d, d)>0\) for \(0<d\ll 1\). This says that the infinity and zero diffusion rates are local CvSSs. Furthermore, there exists at least one \(d^*=d^*(q_1,q_2)\) such that \( \frac{\partial {\tilde{\varLambda }}_2}{\partial D}(d^*, d^*)=0\), and \( \frac{\partial {\tilde{\varLambda }}_2}{\partial D}(d, d)\) changes sign from positive to negative in a neighborhood of \(d^*\); i.e., \(d^*\) is an evolutionary singular strategy which is not a CvSS. \(\square \)
1.4 B.4. The Proof of Theorem 4
The proof of Theorem 4 is divided into a series of lemmas. First, we recall that \(({\hat{V}}_2,{\hat{V}}_3)\) is the unique positive solution of (4) with \(k=1\).
Lemma 17
Let \(d=q_1=0\) and \(D,d_2>0\) and let \(({\hat{V}}_2,{\hat{V}}_3)\) be the unique positive solution of (4). Then \({\hat{V}}_2<1 < {\hat{V}}_3\) and \((1-{\hat{V}}_2,0,0,{\hat{V}}_2,{\hat{V}}_2,{\hat{V}}_3)\) is a nonnegative solution of system (23).
Proof
It is clear that \((1-{\hat{V}}_2,0,0,{\hat{V}}_2,{\hat{V}}_2,{\hat{V}}_3)\) is a nonnegative solution of system (23) when \(d=q_1=0\).
Adding the equations of (4), we have
In view of (46), it is enough to show \({\hat{V}}_3>1\). Suppose not, then \({\hat{V}}_3\le 1\) and the 2nd equation of (4) implies \({\hat{V}}_2 < {\hat{V}}_3 \le 1\), which contradicts (46). \(\square \)
Lemma 18
The matrix
is invertible.
Proof
Observe that zero is an eigenvalue of the cooperative matrix
with a strictly positive eigenvector \(({\hat{V}}_2,{\hat{V}}_3)\). Hence, zero is the principal eigenvalue of \({\hat{E}}_2\). Since the principal eigenvalue is strictly monotone with respect to the diagonal entries, we deduce that zero is not an eigenvalue of \({\hat{E}}_1\). \(\square \)
Set \(U=(U_1, U_2, U_3)\) and \(V=(V_1, V_2, V_3)\). Define map \(F(d,q_1,U, V): {\mathbb {R}}^8\rightarrow {\mathbb {R}}^6\) by
It is clear that \((U,V)=(U_1,U_2,U_3,V_1,V_2,V_3)\) is a steady state of Model II if and only if \(F(d,q_1, U, V)=0\). Now, observe that \(F(0,0,{{\hat{U}}}, {{\hat{V}}})=0\). One can further compute
Lemma 19
Let \(D>0\) and \({q_2\ge 1}\) and consider the eigenvalue problem
Then every eigenvalue of (48) lies in \(\{z \in {\mathbb {C}}:\, Re \, z <0\}\). In particular, \(D_{(U, V)}F(0, 0, {\hat{U}}, {\hat{V}})\) is invertible.
Proof
First, note that system (4) implies
It suffices to show that the principal eigenvalue of (48), denoted as \(\lambda _1^*\), is strictly negative. Suppose to the contrary that (48) holds for some \(\varphi ,\psi \in {\mathbb {R}}^3\) and \(\lambda ^*_1 \in {\mathbb {R}}\) such that
We will show that \(D=0\), which gives a contradiction.
Now, multiply both sides of equation (48) with \(\lambda =\lambda _1^*\) on the left by the row vector
with \(M >0\) to be chosen later. We obtain
where the strict inequality follows from (50), and \(\mathbf {R}\) can be computed (using (49) for \(R_5\)) as
Next choose \(M \gg 1\) so that \(R_2,R_3<0\) and \(R_5,R_6<0\). By inspecting (51) in conjunction with (50), we deduce that
But if we substitute this into the 5th component of (48), we have \(D/2=0\). This is a contradiction. \(\square \)
Lemma 20
Fix any \(D>0\) and \({q_2\ge 1}\). Then there exists some \(\delta >0\) such that for any \(d\in (0, \delta )\), \(q_1\in (-\delta , \delta )\) and \(d+q_1>0\), Model II has a positive steady state, denoted by \((U^\delta , V^\delta )\), which satisfies \((U^\delta ,V^\delta )\rightarrow ({\hat{U}}, {\hat{V}})\) as \((d,q_1)\rightarrow (0,0)\), where \(({\hat{U}}, {\hat{V}})\) is given in (3) and (4) with \(k=1\).
Proof
It is easy to check that \(F(0,0, {\hat{U}}, {\hat{V}})=0\). Moreover, we have shown in Lemma 19 that \(D_{(U, V)}F (0, 0, {\hat{U}}, {\hat{V}})\) is invertible.
By the implicit function theorem, there exists some \(\delta >0\) such that for \(|d|, |q_1|\le \delta \), there exists \((U^\delta , V^\delta )\in {\mathbb {R}}^6\) such that
and \((U^\delta , V^\delta )\rightarrow ({\hat{U}}, {\hat{V}})\) as \(d, q_1\rightarrow 0.\) Finally we show that for all \(d,q_1\) small such that \(d>0\) and \(d+q_1>0\), each component of \((U^\delta , V^\delta )\) is also positive. Since \({\hat{U}}_1, {\hat{V}}_2,{\hat{V}}_3>0\), it suffices to show that \(U_2^\delta >0\) and \(U_3^\delta >0\). Recall from Lemma 17 that \({\hat{V}}_2< 1 < {\hat{V}}_3\). By setting the second component of (47) to zero we have
Using \(q_2 \ge 1\), we deduce that \(U^\delta _2 >0\). Next, we set the third component of (47) to get
Since \({\hat{V}}_3 >1\), we deduce that \(U^\delta _3 >0\). In summary, we have proved that \(U_2^\delta >0\) and \(U_3^\delta >0\) for \(d\in (0, \delta )\), \(q_1\in (-\delta , \delta )\) and \(d+q_1>0\). \(\square \)
Lemma 21
Suppose that \(q_2\ge 1\). Let (U, V) denote any positive solution of Model II. Then as \(d\rightarrow 0\) and \(q_1\rightarrow 0\), \((U, V)\rightarrow ({\hat{U}}, {\hat{V}})\).
Proof
First it is easy to see that \(U_i, V_i\), \(i=1,2,3,\) are uniformly bounded with respect to small \(d, q_1\). Hence, passing to a sub-sequence if necessary we may assume \(U_i\rightarrow {\bar{U}}_i\) and \(V_i\rightarrow {\bar{V}}_i\) as \(d, q_1\rightarrow 0\), where \({\bar{U}}_i, {\bar{V}}_i\ge 0\) satisfy \(F(0, 0, {\bar{U}}, {\bar{V}})=0\), with F defined in (47).
Step 1: \({\bar{U}}_2=0.\)
This is a consequence of assumption \(q_2\ge 1\) and
Step 2: If \({\bar{U}}_3>0\), then \({\bar{U}}_1>0\).
Suppose to the contrary that \({\bar{U}}_3>0\) and \({\bar{U}}_1=0\). The first component of \(F(0,0,{{\bar{U}}},{{\bar{V}}})=0\) yields \({{\bar{U}}_3+{\bar{V}}_3}=1.\) Therefore, the 4th to 6th component of \(F(0,0,{{\bar{U}}},{{\bar{V}}})=0\) can be rewritten as
By \({\bar{U}}_3+{\bar{V}}_3=1\) and \({\bar{U}}_3>0\), we have \({\bar{V}}_3<1.\) By the third equation above, \({\bar{V}}_2=D/(D+q_2) {\bar{V}}_3<1\). Adding three equations we find
which together with \({\bar{V}}_2<1\) implies \({\bar{V}}_1>1\). By the first equation we then obtain \({\bar{V}}_2>{\bar{V}}_1\), which is a contradiction. This completes Step 2.
Step 3: If \({\bar{U}}_3 =0\), then \({\bar{U}}_1>0\).
Suppose to the contrary that \({\bar{U}}_1={\bar{U}}_3=0.\) Then \({\bar{U}}_3\) and \({\bar{V}}_i\) satisfy
If \({\bar{U}}_3=0\), then either \({\bar{V}}_i=0\) for \(i=1,2,3\), or \({\bar{V}}_i={V}^*_i>0\) for \(i=1,2,3\), where \((0, V^*)\) is one of the semi-trivial steady states of Model II. We rule out both cases as follows:
-
1.
For the case \(({\bar{U}}, {\bar{V}})=(0,0,0,0,0,0)\), we have \((U,V)\rightarrow (0,0,0,0,0,0)\) as \(d, q_1\rightarrow 0\), which implies that \(1-(U_i+V_i)>0\) for small \(d, q_1\). Adding the equations of \(U_i\) for \(i=1,2,3\), we have
$$\begin{aligned} U_1(1-U_1-V_1)+U_2(1-U_2-V_2)+U_3(1-U_3-V_3)=0, \end{aligned}$$which is a contradiction as each term in the left-hand side is positive.
-
2.
For the case \(({\bar{U}}, {\bar{V}})=(0, V^*)\), we normalize \(U_i\) by setting \({\tilde{U}}_i\!=\!U_i/(U_1\!+\!U_2\!+\!U_3)\). Then by similar argument we have, by passing to a subsequence if necessary, \({\tilde{U}}_i\rightarrow {\check{U}}_i\ge 0\) as \(d, q_1\rightarrow 0\), and \({\check{U}}_i\) satisfy \({\check{U}}_1+{\check{U}}_2+{\check{U}}_3=1\) and
$$\begin{aligned} {\check{U}}_1(1-V_1^*) =-q_2 {\check{U}}_2+{\check{U}}_2(1-{V_2^*})=q_2{\check{U}}_2+{\check{U}}_3(1-{V_3^*})=0 \end{aligned}$$Since \(V_1^*<1\), so \({\check{U}}_1=0\). It follows from \(q_2\ge 1\) that \({\check{U}}_2=0\). This together with the last equation and \(V_3^*>1\) implies that \({\check{U}}_3=0.\) This contradicts \({\check{U}}_1+{\check{U}}_2+{\check{U}}_3=1\).
Having ruled out both cases above, we proved Step 3.
Step 4: \({\bar{U}}_1>0\).
This is a consequence of Steps 2 and 3.
Step 5: \({\bar{U}}_3=0\).
If not, then \({\bar{U}}_3>0\), which leads to \({\bar{U}}_3+{\bar{V}}_3=1\). Therefore,
Adding the above two equations we find \({\bar{V}}_2(1-{{\bar{V}}_2})=0\). Since \({\bar{V}}_2<1\), the only possibility is \({\bar{V}}_2=0\), from which we have \({\bar{V}}_i=0\) for \(i=1,2,3\) and \({\bar{U}}_1={\bar{U}}_3=1.\) That is, \((U, V)\rightarrow (1, 0, 1, 0, 0, 0)\) as \(d, q_1\rightarrow 0\). We normalize \(V_i\) by setting \({\tilde{V}}_i=V_i/(V_1+V_2+V_3)\). Then by passing to a subsequence if necessary, \({\tilde{V}}_i\rightarrow {\check{V}}_i\ge 0\) as \(d, q_1\rightarrow 0\), and \({\check{V}}_i\) satisfy \({\check{V}}_1+{\check{V}}_2+{\check{V}}_3=1\), and
from which we conclude that \({\check{V}}_i=0\) for all i, which is a contradiction. This completes Step 5.
Step 6: \({\bar{V}}_1={\hat{V}}_2\) and \({\bar{U}}_1=1-{\hat{V}}_2.\)
As \({\bar{U}}_1>0\) and \({\bar{U}}_2={\bar{U}}_3=0\), we have \({\bar{U}}_1+{\bar{V}}_1=1\), \({\bar{V}}_1={\bar{V}}_2\), and
By similar normalization argument we can show that \({\bar{V}}_i>0\) for all i. Hence, \({\bar{V}}_i={\hat{V}}_i\) for \(i=2,3.\) Thus \({\bar{V}}_1={\hat{V}}_2\) and \({\bar{U}}_1=1-{\hat{V}}_2.\) This completes the proof. \(\square \)
Lemma 22
Fix any \(D, q_2>0\). Then there exists some \(\delta >0\) such that for any \(d\in (0, \delta )\) and \(q_1\in (-d, \delta )\), the positive steady state \((U^\delta , V^\delta )\), which is given by Lemma 20, is locally stable.
Proof
By previous result, there exists some \(\delta >0\) such that for \(|d|, |q_1|\le \delta \), there exist \((U^\delta , V^\delta )\in {\mathbb {R}}_+^6\) such that \( F(d, q_1, U^\delta , V^\delta )=(0,0,0,0,0,0)^T. \) Since the two-species competition models II and III are strongly monotone, the linearized system at \((U^\delta , V^\delta )\) has a principal eigenvalue (it is real, simple and has the largest real part among all eigenvalues), which we denote as \(\lambda ^\delta _1\); i.e.,
where \(\varphi ^\delta :=(\varphi _1^\delta , \varphi _2^\delta , \varphi _3^\delta )^T\) and \(\phi ^\delta :=(\phi _1^\delta , \phi _2^\delta , \phi _3^\delta )^T\). Furthermore, we may choose \(\varphi _i^\delta <0\) and \(\phi _i^\delta >0\) for \(i=1,2,3\), and normalize by
We proceed to show that \((U^\delta , V^\delta )\) is stable, that is, \(\lambda _1^\delta <0\). To this end, we argue by contradiction and assume \(\lambda _1^\delta \ge 0\). Let \(\delta \rightarrow 0\) (so that \(d\rightarrow 0, q_1\rightarrow 0)\), by passing to a subsequence, we may assume that \(\lambda _1^\delta \rightarrow \lambda _1^* \ge 0\), so that \(D_{(U,V)}F(0,0,{\hat{U}},{\hat{V}})\) has at least one nonnegative eigenvalue, i.e., that (48) holds for some non-trivial eigenvector \((\varphi ,\psi )\) and nonnegative eigenvalue \(\lambda ^*_1\). However, Lemma 19 asserts that \(\lambda ^*_1 <0\). This is a contradiction.
Therefore, \(\lambda _1^\delta <0\) for sufficiently small \(\delta >0\), i.e., \((U^\delta , V^\delta )\) is stable. \(\square \)
In the following two results, we consider the linear instability of the semi-trivial steady states of (2), denoted by \((U^*,0):=(U_1^*,U_2^*,U_3^*,0,0,0)\) and \((0,V^*):=(0,0,0,V_1^*,V_2^*,V_3^*)\).
Lemma 23
If \(q_2\ge 1\), then there exists \(\delta >0\) such that \((U^*,0)\) is linearly unstable for every \(0 \le d , q_1 \le \delta \).
Proof
Setting \(d= 0, q_1= 0\), \((U_1^*,U_2^*,U_3^*)\) is the unique solution of
By direct calculations and using \(q_2\ge 1\), we get
For \(q_2\ge 1\), we have \((U_1^*,U_2^*,U_3^*)=(1,0,1)\), and its linear stability of \((U^*,0)\) is determined by eigenvalue problem
where \(\varphi =(\varphi _1,\varphi _2,\varphi _3)^{T}\) and
We will test \({\tilde{E}}_2\) by multiplying on the right with the vector \(({\hat{V}}_2-\epsilon , {\hat{V}}_2, {\hat{V}}_3)^T\), where \({\hat{V}}_2, {\hat{V}}_3\) is given in Theorem 4 and \(\epsilon \) is a small positive constant.
Since all of the entries of the right-hand side are positive, we can apply the Collatz–Wielandt Formula (Meyer 2000, P. 667) to get
that is, \((U^*,0)\) is linearly unstable when \(d=q_1=0\). By continuity, it remains linearly unstable for all small d and \(q_1\). \(\square \)
Lemma 24
For each \(D, q_2 >0\), there exists \(\delta >0\) such that \((0,V^*)\) is linearly unstable for every \(0 \le d,q_1 \le \delta \).
Proof
Setting \(d=0\) and \(q_1=0\), the linear instability of \((0,V^*)\) is determined by the principal eigenvalue of the following problem
Clearly, \(\varLambda =V^*_3-1\) is an eigenvalue with eigenfunction \((0,0,1)^T\). Recalling that \(V_1^*<1\) (Lemma 8), we deduce that there is at least one negative eigenvalue. Thus \((0,V^*)\) is linearly unstable. \(\square \)
Proof of Theorem 4
By Lemma 20, there exists some \(\delta >0\) such that for any \(d\in (0, \delta )\), \(q_1\in (-d, \delta )\), Model II has a unique positive steady state \((U^\delta , V^\delta )\) in a small neighborhood of \(({\hat{U}}, {\hat{V}})\). Lemma 21 further ensures that this is the only positive steady state for small positive d and \(q_1\). By Lemma 22, \((U^\delta , V^\delta )\) is locally stable. We can then conclude by the theory of monotone dynamical systems (Hess 1991; Hess and Lazer 1991; Smith 1995) and Lemmas 23 and 24 that \((U^\delta , V^\delta )\) is globally stable. \(\square \)
Appendix C: The Dynamics of Model III
In this section, we mainly consider the dynamics of Model III, i.e., system (5), in homogeneous environments. We consider the nonnegative and non-trivial solutions of
There are three types of nonnegative and non-trivial solutions of this system. We denote three different types of solutions as (U, V), in which \(U_i>0\) and \(V_i=0\), or \(U_i=0\) and \(V_i>0\), or \(U_i>0\) and \(V_i>0\) for all \(i=1,2,3\). The semi-trivial steady state \((U^*,0):=(U_1^*,U_2^*,U_3^*,0,0,0)\) satisfies
The linear stability of \((U_1^*,U_2^*,U_3^*,0,0,0)\) is determined by the sign of the principal eigenvalue \({\tilde{\varLambda }}_3\) of the eigenvalue problem
where matrix \(F_3\) is given by
Proposition 6
When \(D=d\), the derivative of \({\tilde{\varLambda }}_3\) with respect to D satisfies
Proof
Differentiate (54) with respect to D, we get
where \(\varphi _i'=\frac{\partial \varphi _i}{\partial D}\), \(i=1,2,3\). Note that when \(D=d\),
and when \(D=d\), we may choose
Set \(D=d\) in (56) and multiplying it by \(\begin{pmatrix}U_1^*, \frac{d}{d+q_1}U_2^*, \frac{d}{d+q_2}U_3^*\end{pmatrix}\), using (57), (58) and \({\tilde{\varLambda }}_3(d,d)= 0\), we obtain (55). This completes the proof. \(\square \)
Next, we establish some a prior estimates of \(U_i^*\), \(i=1,2,3\).
Lemma 25
For any \(d,D>0\) and \(q_1,q_2>0\), the following results hold:
-
(i)
If \(q_1 \ge q_2\), then \(U_2^* \ge U_3^*\);
-
(ii)
If \(q_1 \le q_2\), then \(U_2^* \le U_3^*\).
In particular, if \(q_1=q_2\), \(U_2^*=U_3^*\) holds.
Proof
We prove (i) only and (ii) can be shown similarly. We will assume \(U^*_2 < U^*_3\) and deduce \(q_1 < q_2\). By the second and third equation of (53),
This implies
Combining the above, we have
This implies \(q_2 > q_1\). This proves (i). \(\square \)
Lemma 26
For any \(d,D>0\) and \(q_1,q_2>0\), \(U_1^*<1\) always holds.
Proof
By exchanging patches 2 and 3 if necessary (the river network is symmetric), we may assume without loss of generality that \(q_1\ge q_2\).
We argue by contradiction and assume that \(U_1^* \ge 1\) for some \(q_1 \ge q_2>0\). By the first equation of (53), we get
Using \(U_2^* \ge U_3^*\) (from Lemma 25), we have
so \(U_2^*>U_1^*\ge 1\).
In view of \(\sum _{i=1}^3 U_i^*(1-U_i^*)=0\) (upon summing (53)), we must have \(U_3^*<1\). By the third equation of (53), we get \(U_3^*>\frac{d+q_2}{d}U_1^*>U_1^*\ge 1\). This is a contradiction. This finishes the proof for the case \(q_1\ge q_2\), and the case \(q_1\le q_2\) can be treated similarly. \(\square \)
1.1 C.1. The Global Dynamics of Model III When \(q_1=q_2\)
In this part, we shall show Theorem 5. We first establish some a priori estimates of nonnegative and non-trivial steady state of (52) as \(q_1=q_2\).
1.1.1 C.1.1. Preliminary Results on Nonnegative, Non-trivial Steady States
Lemma 27
Suppose \(q_1=q_2>0\), then for any \(d, D>0\), we have \(U_2=U_3\) and \(V_2=V_3\).
Proof
By the similar argument in the proof of Lemma 1, we can obtain this lemma. \(\square \)
Lemma 28
Suppose \(q_1=q_2:=q>0\), then for any \(d, D >0\), we have \(U_3+V_3>1\).
Proof
Assume that \((U_i)\) is non-trivial and is thus positive for all i. The same argument applies to the case \(U_1=U_2=U_3=0\). Assume to the contrary that \(U_3+V_3 \le 1\) for some \(d,D,q>0\), then by the third equation of (52), we get \(d(U_1-U_3)+qU_1 \le 0\). Thus \(U_3 \ge \frac{d+q}{d}U_1>U_1\).
Similarly, we can show \(V_3>V_1\), if \(V_i>0\) for all i. Hence \(1\ge U_3+V_3>U_1+V_1\) holds. By the first equation of (52) and Lemma 27, we obtain
i.e., \(d(U_3-U_1)-qU_1<0\). This together with the third equation of (52) implies that \(U_3+V_3>1\), which is a contradiction with our assumption. \(\square \)
Lemma 29
Suppose \(q_1=q_2=q>0\), then for any \(d, D >0\), we have \(U_1+V_1<1\).
Proof
We argue by contradiction. If \(U_1+V_1 \ge 1\) for some \(d,D,q>0\), then by the first equation of (52) and Lemma 27, we have
so that \(d(U_2-U_1)-qU_1 \ge 0\), which together with the second equation of (52) implies \(U_2+V_2 \le 1\). Using Lemma 27, we have \(U_3+V_3 = U_2+V_2 \le 1\). But this contradicts Lemma 28. \(\square \)
Lemma 30
Suppose \(q_1=q_2:=q>0\) and \(d, D >0\).
-
(i)
If \(U_i>0\) for all i, then \(U_1 < U_3\).
-
(i)
If \(V_i>0\) for all i, then \(V_1 < V_3\).
Proof
In case of \((U^*,0)\) and \((0,V^*)\), the lemma follows from Lemmas 28 and 29. It therefore suffices to consider positive equilibria (U, V). We will prove (i), as (ii) follows from a similar argument.
If \(U_1 \ge U_3\) for some \(d,D,q>0\), by Lemmas 28 and 29, we have \(V_3>V_1\), which together with the 6th equation of (52) implies
i.e., \(q+1-U_3-V_3>0\). However, by the third equation of (52) and \(U_1 \ge U_3\), we get
i.e., \(q+1-U_3-V_3\le 0\). This is a contradiction. \(\square \)
The following result is a direct consequence of Lemmas 27 and 30, and it provides some insight for the biological interpretation of Theorem 5.
Corollary 6
Assume \(q_1=q_2>0\) and \(d, D >0\). Then we have
Lemma 31
Suppose \(q_1=q_2>0\), then for any \(d,D>0\), we have
Proof
By possibly exchanging the role of U and V, we may assume \(U_i>0\) for all i. Adding the equations of \(U_i \ (i=1,2,3)\) in (52) and using \(U_2+V_2=U_3+V_3>1\) (Lemmas 27 and 28), \(U_1+V_1<1\) (Lemma 29) and \(U_1<U_3=U_2\) (Corollary 6), we obtain
which establishes (59). \(\square \)
Lemma 32
Suppose \(q_1=q_2:=q>0\) and \(d,D>0\). Then we have \(-2q+1-U_1-V_1<0.\)
Proof
Corollary 6 and the first equation of (52) indicate that \(-2q+1-U_1-V_1<0\). \(\square \)
Theorem 10
If \(q_1=q_2>0\) and \(d, D > 0\), then system (52) has no positive solution.
Proof
We argue by contradiction. If there exists a positive solution (U, V) for (52), by direct calculation, we obtain
Lemma 31 shows that the left-hand side of (60) is positive, but the right-hand side of (60) is negative, due to Lemmas 27, 28 and 32. This completes the proof. \(\square \)
1.1.2 C.1.2. Global Stability of \((U^*,0)\) When \(q_1=q_2\)
We assume \(q_1=q_2:=q\) throughout this subsection. The local instability of \((0,V^*)\) is determined by the sign of the principal eigenvalue, denoted as \(\varLambda _3\), of the system
where \(E_3\) is rewritten as
Setting \(q_1=q_2:=q\) and exchanging the role of the two species, (55) can be rewritten as
We can obtain the following result by direct calculations:
Proposition 7
(Jiang et al. 2020, Proposition 6) Assume \(q_1=q_2:=q>0\) and \({V_1^*+V_2^*+V_3^*}\ne 3\). Then \(\mathrm{det}(E_3)=0\) if and only if either \(d=D\), or
Corollary 7
Suppose \(q_1=q_2:=q>0\), then for any \(d,D>0\), \(\frac{\partial \varLambda _3}{\partial d}\big |_{d=D}<0.\)
Proof
By Corollary 6, we have \(V_1^*-V_2^*<0, V_1^*-V_3^*<0\). Using \(V_2^*>1\) and the fifth equation of (52), we get
Similarly, by \(V_3^*>1\) and the sixth equation of (52), we get
Therefore, the right-hand side of (61) is strictly negative. \(\square \)
Lemma 33
Suppose \(q_1=q_2>0\), then for any \(d,D>0\), the right-hand side of (62) is negative.
Proof
Using Lemmas 27 and 28, we get \(V_2^*=V_3^*>1\), hence \((1-V_2^*)(1-V_3^*)>0\), which together with Lemmas 31 and 32 shows that the right-hand side of (62) is strictly negative. \(\square \)
Theorem 11
Suppose \(q_1=q_2>0\), then for any \(d,D>0\), we have
Proof
Equation (62) cannot hold since the right-hand side is strictly negative, by Lemma 33. Hence, Proposition 7 says that \(\varLambda _3(d,D)=0\) if and only if \(d=D\). Therefore, by Corollary 7 and the continuity of \(\varLambda _3\), \(\varLambda _3(d,D)>0\) holds for \(D>d>0\) and \(\varLambda _3(d,D)<0\) holds for \(0<D<d\). The result for \({\tilde{\varLambda }}_3\) follows from the identity \({\tilde{\varLambda }}_3(d,D)=\varLambda _3(D,d)\) for all d, D. \(\square \)
Proof of Theorem 5
For \(d>D\), Theorems 11 and 10 say that \((0,V^*)\) is linearly unstable, and that Model III has no positive equilibria. It follows from the theory of monotone dynamical systems (Hess and Lazer 1991, Theorem 1.5) that the equilibrium \((U^*,0)\) is globally asymptotically stable.\(\square \)
1.2 C.2. The Local Stability of \((U^*,0)\)
In this subsection, we determine the local stability of the semi-trivial steady sate \((U^*,0)\) for more general \(q_1, q_2\).
Lemma 34
Suppose \(0<q_2\le q_1+\frac{1}{2}\), \(0<\frac{q_2}{q_1}\le \sqrt{2}\). Then \(U_2^*>1\) holds for all \(d>0\).
Proof
Since we have shown \(U_2^*>1\) for \(q_1=q_2\), it is sufficient to show \(U_2^*\not =1\) for any \(q_1, q_2\) satisfying the assumptions. We argue by contradiction: Suppose that \(U_2^*=1\) for some \(q_1, q_2\). By the second equation of (53), we have
Adding the equations of \(U_1^*, U_2^*, U_3^*\) in (53) and using \(U_2^*=1\), we get
Substituting (63) and (64) into the third equation of (53), we get
By \(U_1^*<1\) (from (63)) and (64), we see that \(U_3^*>1\). This, together with (65), implies that
Hence, \(q_2-q_1>0\). Therefore, we can rewrite (66) to get
which the last inequality follows from \(0<q_2-q_1\le \frac{1}{2}\). By (63), (64) and (65), after simplifications, we have
It follows from (68) that \(dq_1<(q_2-q_1)(d+q_2)\), which can be rewritten as
By (67) and (69), note that \(2q_1-q_2>0 \) by assumption, we have
By assumption \(0<q_2-q_1 \le \frac{1}{2}\), we have \(\frac{q_1}{q_2-q_1} \ge 2q_1\). Hence, by (70)
which is equivalent to \(q_2>\sqrt{2}q_1\), a contradiction to assumption \(q_2 \le \sqrt{2}q_1\). \(\square \)
Lemma 35
Suppose that \(0<q_1\le q_2+ \frac{1}{2}\) and \(\frac{q_2}{q_1}\ge \frac{1}{\sqrt{2}}\), then \(U_3^*>1\) holds for all \(d>0\).
Proof
This proof is similar to Lemma 34, by exchanging the role of patches 2 and 3. We omit the proof. \(\square \)
Directly by Lemmas 34 and 35, we obtain the following result, which also provides some insight for the biological interpretation of Theorem 6.
Corollary 8
Let \(q_1,q_2,d,D\) be positive. If \(| q_2-q_1 | \le \frac{1}{2}\) and \(\frac{1}{\sqrt{2}}\le \frac{q_2}{q_1}\le \sqrt{2}\), then \(U_2^*>1\) and \(U_3^*>1\) hold.
Proof of Theorem 6
Fix \(d > D\). We have shown that if \(q_1=q_2\), \({\tilde{\varLambda }}_3(d,D)>0\) in Theorem 11. By the continuity of \({\tilde{\varLambda }}_3\) in \(q_1,q_2\), we just need to prove \({\tilde{\varLambda }}_3 \not =0\). By contradiction, we assume that there exist some \(q_1,q_2\) such that \({\tilde{\varLambda }}_3 =0\). Then by direct calculation, we get
Adding the equations of (53), we get
Due to \(U_2^*>1\) and \( U_3^*>1\), we have \(U_1^*<1\), so
Substituting this into (72), we obtain \(3-U_1^*-U_2^*-U_3^*>0\). Again using \(U_1^*<U_2^*, U_1^*<U_3^*\) and the first equation of (53), \(-(q_1+q_2)+(1-{U_1^*})<0\). This together with \(U_2^*>1\) and \(U_3^*>1\) (Corollary 8) yields the right-hand side of (71) is negative. This contradiction finishes the proof. \(\square \)
1.3 C.3. Existence of Evolutionarily Singular Strategy
The goal of this subsection is to establish Theorem 7.
Lemma 36
For any \(q_1,q_2>0\), we have \( \frac{\partial {\tilde{\varLambda }}_3}{\partial D}(d,d)<0\) for sufficiently large d.
Proof
By Proposition 6, the sign of \(\frac{\partial {\tilde{\varLambda }}_3}{\partial D}(d,d)\) is the opposite of that of N, where
By (53), we can rewrite (73) as
Note that \((U_1^*,U_2^*,U_3^*) \rightarrow (1,1,1)\) as \(d \rightarrow \infty \). As \((U_1^*, U_2^*, U_3^*)\) is the unique stable positive solution of (53), it can be shown that it is smooth at \(d=\infty \) so that we can expand \(U_i^*\) as \(U_i^*=1+{\tilde{U}}_i/d+O(1/d^2)\), \(i=1,2,3\), for sufficiently large d. From the second and third equation of (53) we have
Recall (72) (resulting from adding three equations in (53)), it follows that
By solving (75) and (76), \({\tilde{U}}_1=-(q_1+q_2)/3\), \({\tilde{U}}_2=(2q_1-q_2)/3\), \({\tilde{U}}_3=(2q_2-q_1)/3\). Hence, for large d it holds that
Substituting into (74), we obtain
provided that \((q_1, q_2)\not =(0,0)\). Therefore, we conclude by Proposition 6 that \(\frac{\partial {\tilde{\varLambda }}_3}{\partial D}(d,d)<0\) for sufficiently large d. \(\square \)
Lemma 37
Let \(q_1,q_2>0\). For sufficiently small d, we have
Proof
We consider three cases:
Case I. \(q_1+q_2< 1\). By the first equation of (53), we see that \(U_1^* \rightarrow {\overline{U}}_1:= 1-(q_1+q_2)>0\) as \(d \rightarrow 0\). By the second and third equation of (53), \(U_2^*\rightarrow {\overline{U}}_2\ge 1, U_3^*\rightarrow {\overline{U}}_3 \ge 1\). Thus \(N \rightarrow {\overline{U}}_1({\overline{U}}_2+{\overline{U}}_3-2{\overline{U}}_1)>0\) as \(d\rightarrow 0\), where N is given by (73). Here we used \(q_1>0\) and \(q_2>0\). By Proposition 6, we have \( \frac{\partial {\tilde{\varLambda }}_3}{\partial D}(d,d)<0\) for sufficiently small d when \(q_1+q_2<1\).
Case II. \(q_1+q_2=1\). For this case, we have \(U_1^* \rightarrow 0\) and \(U_i^* \rightarrow 1\) \((i=2,3)\) as \(d\rightarrow 0\). By the first equation of (53), we get
Thus \(N=2\sqrt{2} d^{1/2}+ o(1)\) is positive for sufficiently small d. Therefore, \( \frac{\partial {\tilde{\varLambda }}_3}{\partial D}(d,d)<0\) when \(q_1+q_2=1\).
Case III. \(q_1+q_2>1\). For this case, we have \(U_1^* \rightarrow 0\) and \(U_i^* \rightarrow 1\) \((i=2,3)\) as \(d\rightarrow 0\). By the first equation of (53), we get
Substituting into (73), we get
Hence,
Having determined the sign of N for d sufficiently small, (78) follows from Proposition 6. \(\square \)
Proof of Theorem 7
Since \(d\mapsto \frac{\partial {\tilde{\varLambda }}_3}{\partial D}(d, d)\) is analytic, all the roots are discrete. By Lemmas 36 and 37, \(\frac{\partial {\tilde{\varLambda }}_3}{\partial D}(d, d)>0\) for d small and \(\frac{\partial {\tilde{\varLambda }}_3}{\partial D}(d, d)<0\) for \(d\gg 1\). This says that the infinity and zero diffusion rates are local CvSSs. Furthermore, there exists at least one \(d^*=d^*(q_1,q_2)\) such that \( \frac{\partial {\tilde{\varLambda }}_3}{\partial D}(d^*, d^*)=0\), and \( \frac{\partial {\tilde{\varLambda }}_3}{\partial D}(d, d)\) changes sign from positive to negative in a neighborhood of \(d^*\); i.e., \(d^*\) is an evolutionary singular strategy which is not a CvSS. \(\square \)
Rights and permissions
About this article
Cite this article
Jiang, H., Lam, KY. & Lou, Y. Three-patch Models for the Evolution of Dispersal in Advective Environments: Varying Drift and Network Topology. Bull Math Biol 83, 109 (2021). https://doi.org/10.1007/s11538-021-00939-8
Received:
Accepted:
Published:
DOI: https://doi.org/10.1007/s11538-021-00939-8