Journal of Theoretical Biology: Samuel M. Flaxman, Yuan Lou
Journal of Theoretical Biology: Samuel M. Flaxman, Yuan Lou
Journal of Theoretical Biology: Samuel M. Flaxman, Yuan Lou
a r t i c l e in fo abstract
Article history: We synthesize previous theory on ideal free habitat selection to develop a model of predator movement
Received 19 February 2008 mechanisms, when both predators and prey are mobile. We consider a continuous environment with an
Received in revised form arbitrary distribution of resources, randomly diffusing prey that consume the resources, and predators
8 September 2008
that consume the prey. Our model introduces a very general class of movement rules in which the
Accepted 23 September 2008
overall direction of a predator’s movement is determined by a variable combination of (i) random
Available online 9 October 2008
diffusion, (ii) movement in the direction of higher prey density, and/or (iii) movement in the direction of
Keywords: higher density of the prey’s resource. With this model, we apply an adaptive dynamics approach to two
Predator–prey interactions main questions. First, can it be adaptive for predators to base their movement solely on the density of
Resource tracking
the prey’s resource (which the predators do not consume)? Second, should predator movements be
Reaction–diffusion–advection models
exclusively biased toward higher densities of prey/resources, or is there an optimal balance between
Ideal free distribution
random and biased movements? We find that, for some resource distributions, predators that track the
gradient of the prey’s resource have an advantage compared to predators that track the gradient of prey
directly. Additionally, we show that matching (consumers distributed in proportion to resources),
overmatching (consumers strongly aggregated in areas of high resource density), and undermatching
(consumers distributed more uniformly than resources) distributions can all be explained by the same
general habitat selection mechanism. Our results provide important groundwork for future investiga-
tions of predator–prey dynamics.
& 2008 Elsevier Ltd. All rights reserved.
0022-5193/$ - see front matter & 2008 Elsevier Ltd. All rights reserved.
doi:10.1016/j.jtbi.2008.09.024
ARTICLE IN PRESS
188 S.M. Flaxman, Y. Lou / Journal of Theoretical Biology 256 (2009) 187–200
consumer and its resource. Consumers are assumed to be ‘‘ideal’’ to ignore the prey and pay attention only to the resource, two
in their knowledge of all aspects of the environment (i.e., they trophic levels below, with which they have no direct interaction.
know the qualities of patches and the numbers of competitors and Leap-frogging may seem counterintuitive, so the underlying
predators therein) and ‘‘free’’ from any constraints on or costs of logic of this prediction merits a brief explanation. The following
travel (see Tregenza, 1995 for a review of assumptions and models explanation glosses over the details of individual models, but will
relaxing them). A general prediction of interest from IFD models is hopefully provide the reader with a useful heuristic for under-
the extent to which consumers are predicted to ‘‘match’’ resources standing why so many models predict leap-frogging. First,
(Tregenza, 1995; Flaxman and deRoos, 2007). A consumer consider the perspective of predators. The predator distribution
‘‘matches’’ a resource when the proportion of consumers in each can only be stable if the prey are distributed such that the per
patch is the same as the proportion of resources in each patch. capita predator payoff is equal in the two patches. In many of the
‘‘Undermatching’’ occurs when a species is under-represented in equilibrium models, predators are assumed to (i) compete weakly
the best patches compared to perfect matching, and ‘‘over- (if at all) and (ii) have a linear functional response. With such
matching’’ refers to the reverse of undermatching. (For general assumptions, the predator payoff can only be equalized between
discussions about matching, see Kennedy and Gray, 1993; patches when the prey are (nearly) uniformly distributed between
Milinski, 1994; Tregenza, 1995; Flaxman and Reeve, 2006; Flax- patches. However—now considering the perspective of prey—the
man and deRoos, 2007; Houston, 2008). prey will only remain at such a distribution if the patches are
Most IFD models focus on only one species of consumer equal in their food/risk ratios. The latter condition will only be
(reviewed by Tregenza, 1995; Sih, 2005; see also Lima, 2002; satisfied if predators match the resource. In sum, predators drive
Kacelnik et al., 1992), but a number of models have been prey to undermatch resources, and prey drive predators to match
developed that focus on simultaneous habitat selection by resources. Extremely few empirical studies have both (i) exam-
predators and prey. These models can be broadly categorized as ined simultaneous patch choice by predators and prey while
belonging to one of three types (sensu Kacelnik et al., 1992). In the (ii) measuring patch qualities (Lima, 2002; Sih, 2005). Of those
first category are models that make predictions about the nature that have, there is some support for the leap-frogging prediction
of evolutionarily stable distributions that predators and prey (Bouskila, 2001; Sih, 2005; Hammond et al., 2007). Intriguingly,
could simultaneously adopt (Iwasa, 1982; Hugie and Dill, 1994; counterintuitive distributions similar to leap-frogging have also
Sih, 1998, 2005; Bouskila, 2001; Heithaus, 2001; Alonzo, 2002; been found in the host–parasitoid literature (Fox and Eisenbach,
Rosenheim, 2004). Henceforth, we refer to models in this first 1992; Godfray, 1994; Schreiber et al., 2000).
category as ‘‘equilibrium models’’, since they assume that an The fact that many equilibrium models predict leap-frogging
equilibrium is reached without specifying any behavioral me- (at least as one possible distribution) suggests that a very simple
chanisms that could bring a population to the equilibrium. A behavioral mechanism could be used by predators to reach an
second category assumes fixed populations of predators and prey evolutionarily stable distribution. Instead of tracking the prey
and asks whether certain movement rules will lead the popula- directly, the predators could use a ‘‘resource tracking’’ strategy,
tions to an ESS (Schwinning and Rosenzweig, 1990; Jackson et al., that is, they could make habitat-selection and patch-switching
2004b). In the third category are models that add population decisions based—exclusively or in part—on the distribution of the
dynamics and ask whether certain migration rules lead popula- prey’s resource (hereafter, ‘‘resource’’), similar to how some
tions to spatial distributions that are both ecologically and parasitoids use plant-based cues to find hosts (Godfray, 1994).
evolutionarily stable (van Baalen and Sabelis, 1993, 1999; Křivan, While such a patch-choice mechanism has been mentioned
1997; Cressman et al., 2004; Kimbrell and Holt, 2004, 2005; casually in a number of previous publications (such as the
Abrams, 2007; Abrams et al., 2007). We refer collectively to equilibrium models, above), surprisingly, mechanistic models
models in the second and third categories as ‘‘mechanistic’’ have not formally investigated the effectiveness or consequences
models, since they deal with specific patch-switching or habitat- of resource tracking by predators. Instead, the mechanistic IFD
choice mechanisms. All these models treat the environment as models have all assumed that predators base their patch switch-
being composed of a finite number of patches, and in the large ing decisions directly on their own fitness gradient or on the prey
majority of cases that number is two (but see van Baalen and gradient (i.e., a ‘‘prey tracking’’ mechanism).
Sabelis, 1993, 1999; Jackson et al., 2004b; Schreiber and Vejdani, Investigating the effectiveness and the consequences of
2006). predators using a ‘‘resource tracking’’ habitat selection mechan-
With the exception of Iwasa’s (1982) model, the equilibrium ism is important for at least two main reasons. First, there are
models share a common prediction, and one that distinguishes good reasons to expect that resource tracking could be an adaptive
them from most single-species IFD models: for at least some mechanism for predators to employ. When prey are mobile but
assumptions and particular parameter values, the predators’ resources are sessile (e.g., herbivorous prey and the plants they
distribution will match the distribution of the prey’s resource eat), resource tracking may require much less time and energy
(not the distribution of prey), and the prey will be uniformly than prey tracking. Similarly, if the locations of resources are
distributed between patches (for detailed discussions of this obvious and predictable, but the locations of prey are not, resource
prediction, see Rosenheim, 2004; Sih, 2005; Hammond et al., tracking may be the only feasible option. At the very least,
2007). Furthermore, for a broad range of parameter values, a resource tracking could provide predators with a way to narrow
number of equilibrium models predict that (i) prey will under- down search options.
match their resource to some extent (instead they will match the Second, recent theoretical work has revealed that the details of
resource/risk ratio) and (ii) predators will overmatch the prey and patch choice mechanisms can have profound consequences for
match or slightly undermatch the prey’s resource. In other words, both spatial distributions and population dynamics (Kimbrell and
it is frequently predicted that neither predator nor prey will Holt, 2004; Abrams et al., 2007). We might thus expect that
match their own food source; instead, the dynamic interplay of resource tracking by predators could alter the ecological and
behaviors will lead to predators being more common in patches evolutionary dynamics of predator–prey systems (as compared
with high levels of the prey’s resource and prey subsequently with prey tracking). For example, spatial and temporal cycles in
undermatching their own resource to decrease risk. Sih (2005) predator–prey dynamics are a common result in a variety of
termed this prediction ‘‘leap-frogging’’, since, superficially, the predator–prey models (e.g., Schwinning and Rosenzweig, 1990),
predators—rather than matching the prey they consume—appear and much effort has been devoted to understanding factors that
ARTICLE IN PRESS
S.M. Flaxman, Y. Lou / Journal of Theoretical Biology 256 (2009) 187–200 189
could dampen such cycles. An intuitively appealing hypothesis is Our analysis is mainly focused on the success of different
that resource tracking by predators could eliminate such cycles classes of movement mechanisms relative to one another. As such,
altogether, since predators would not chase prey from patch to we do not address the question of how well particular mechan-
patch. As another example, empirical work suggests the hypoth- isms lead populations to an IFD. Most previous mechanistic
esis that—in addition to potentially stabilizing an equilibrium— models—which do look explicitly at achieved distributions
resource tracking predators can speed the approach of a relative to IFDs—have focused on a very narrow definition of the
predator–prey system to an equilibrium (S.M. Flaxman and Y. environment (two discrete patches with a simple ratio of patch
Lou, unpublished). Currently, however, there is no theory that can qualities) and on very narrow classes of movement mechanisms
be used to bolster or refute hypotheses such as these, nor is there (e.g., explicit formulations based on prey numbers or fitness
any theory that has answered a basic question: Under what directly). Both our environment and our movement mechanisms
circumstances is resource tracking a viable habitat selection are considerably more general than these previous approaches.
mechanism for predators? Our models thus provide results which, rather than providing a
To address these issues, we use reaction–diffusion–advection final word, are meant to open up and spur on future work
models to describe a predator–prey system and to develop a very studying both the nature and achievement of spatial distributions,
general class of predator movement mechanisms. In recent years including the IFD. In sum, rather than rehashing the approach of
reaction–diffusion–advection models have been used to model previous equilibrium and mechanistic IFD models, we have taken
biased movement of species along environmental or fitness this rich and successful body of literature and synthesized it in
gradients for both single species (Belgacem and Cosner, 1995; order to begin building a more general theory of animal move-
Cosner and Lou, 2003; Cosner, 2005; Rowell, 2007) and two ment and habitat selection mechanisms and the consequences
competing species (Grindrod, 1988; Cantrell et al., 2006, 2007; thereof.
Chen and Lou, 2008; Chen et al., 2008) (see also Farnsworth and
Beecham, 1997; Cantrell and Cosner, 1999). Kareiva and Odell
(1987) used reaction–diffusion–advection models to study how 2. Models
simple behaviors could lead to ‘‘preytaxis’’, that is, predator
aggregation in areas of high prey density (what we refer to as Our goal is to better understand biased movement of
‘‘prey tracking’’). Our models are most similar to this work. predators—that is, movement in the direction of some feature of
However, whereas Kareiva and Odell were mainly concerned with the environment—and in particular, we wish to know if there are
how prey tracking might arise, our focus is on the ecological and advantages for predators that track the resources of the prey. For
evolutionary consequences of a variety of predator and prey convenience, we refer to the resources of the prey simply as
movement mechanisms (including prey tracking). Furthermore, ‘‘resources’’. Predators that track the gradient of the resource—
Kareiva and Odell did not relate their work to IFD theory. moving toward a location of higher habitat quality (from the
Our investigations unite a mechanistic approach to habitat prey’s perspective)—are termed resource trackers. Similarly,
selection with results from equilibrium IFD models (e.g., the predators that track prey directly are referred to as prey trackers.
‘‘leap-frogging’’ prediction discussed above). We extend the If a species adopts purely random diffusion, we refer to it as a
seminal work of Kareiva and Odell to develop theory on general random tracker or random diffuser. As mentioned above, future
movement mechanisms that can be used to address questions work will consider how effective these various mechanisms are
about if and how organisms can achieve IFDs, or any other spatial for achieving a specific distribution (such as an IFD). For now, we
distributions. begin with the more general question of how adaptive different
We consider a three-trophic-level scenario: mobile predators, classes of mechanisms are relative to each other.
mobile prey, and a static distribution of resources. Unlike most To compare different predator movement mechanisms, our
previous IFD models that consider an environment of only two basic model formulation is a predator–prey system with two
discrete patches, we consider a continuous environment (e.g., species of predators and one species of prey. We assume that the
Kshatriya and Cosner, 2002) with an arbitrary distribution of prey diffuse randomly, but the dispersal of both species of
resources. To make things simple enough to be analytically predators consists of two components: random diffusion and
tractable, we assume that, while prey are mobile, the prey do directed movement along the gradient of some mixture of the
not move in response to the distribution of predators; instead, resources of the prey and the density of prey. In other words, both
prey move by simple diffusion. We will relax this assumption in predator species can sense and evaluate the local environment
future work, but it is beyond the scope of this paper to include and local prey density. More precisely, let Pi ðx; tÞ and Vðx; tÞ denote
what would be such a lengthy analysis here; beginning with the the densities of the i-th predator species (i ¼ 1; 2; i always denotes
simplest possible system is the most appropriate first step. a species of predator) and the prey species, respectively, at
Furthermore, because predators kill prey, the equilibrium prey location x and time t, and let RðxÞ denote the intrinsic growth rate
distribution is not fixed and will depend upon both predator of the prey species which directly reflects the habitat quality (e.g.,
behavior and prey movement. Predator movement is the com- resource density) at location x. We assume that there are no
bined result of (1) random movement, (2) movement in the resource dynamics (i.e., RðxÞ does not change with time). Under
direction of higher prey density (‘‘prey tracking’’), and (3) our assumptions, the dispersal of the predators and prey can be
movement in the direction of higher densities of the prey’s described, respectively, in terms of their fluxes (Okubo, 1980;
resource (‘‘resource tracking’’). We study two main questions. Kareiva and Odell, 1987):
First, considering the special cases of predators that use either
Ji ¼ di rP i þ ai P i rf i ðR; VÞ; J v ¼ dv rV,
resource tracking or prey tracking (but not both simultaneously),
might resource tracking in some cases be more advantageous than where di rPi and dv rV account, respectively, for random diffusion
prey tracking? Second, what are the optimal relative magnitudes of the i-th species of predator and the prey, ai P i rf i ðR; VÞ
of random and biased movement? To answer these questions, we represents biased movement of the i-th predator species along
investigate the adaptive dynamics of predator habitat selection the environmental resource gradient and/or prey density gradient,
mechanisms, where a particular mechanism can be defined by and f i are some functions to be specified later. The positive
parameters that determine the relative contributions of each of constants di , dv are random dispersal rates of the i-th predator
the three types of movement mentioned above. species and prey, respectively; the non-negative constant ai
ARTICLE IN PRESS
190 S.M. Flaxman, Y. Lou / Journal of Theoretical Biology 256 (2009) 187–200
measures the tendency of the i-th predator species to move along The main results of our analysis of this model and particular
the gradient of the resources and/or the prey density gradient. If cases are given below in Section 3. For details of the analysis and
we further assume that there is no flux across the boundary proofs of claims, the reader is referred to Appendix A. Verbally,
(i.e., immigration and emigration balance each other, if they some of the important results can be summarized as follows. First,
occur) and also incorporate the simplest population dynamics for both resource trackers and prey trackers seem to do better than
predator–prey interactions, we are led to the following reaction– random trackers. Second, if the two species of predators are
diffusion–advection system: identical in all ways except in their biased movements, we show
8 that, in some cases, (i) when both predators are rare, the resource
< Pi;t ¼ r ½di rP i ai P i rf i ðR; VÞ þ Pi ðki þ ai VÞ in O ð0; 1Þ;
>
trackers can invade for conditions in which the prey trackers
V t ¼ dv DV þ V½RðxÞ V b1 P 1 b2 P2 in O ð0; 1Þ;
> cannot and (ii) when one predator has invaded and the other is
: ½d rP a P rf ðR; VÞ n ¼ rV n ¼ 0 on qO ð0; 1Þ
i i i i i rare, the resource trackers can invade established prey trackers
(2.1) but not vice versa. The underlying reason for the success of the
resource tracker is the following. Since the prey distribution tracks
for i ¼ 1; 2.
the resources quite well, the dispersal strategy of being a resource
Our model not only can be derived as the above by the
tracker can be fairly efficient since the resource gradient and prey
common approach based upon fluxes (see, e.g., Murray, 2002,
density gradient are in the same direction, but the resource
2003 for many such examples), but also can be derived as limits of
gradient is sharper.
discrete models based on random walks (see, e.g., Okubo and
Levin, 2001). Here ki , ai , bi are all positive constants, where ki is
the mortality rate of the i-th predator, ai is a parameter
3. Results
determining the growth rate of the predator population due to
consumption of prey, and bi is a parameter determining the death
P 2 For each of the two situations outlined above (that is,
rate of prey due to predation by the i-th predator. D:¼ N 2
j¼1 q =qxj
N variable t and variable a), we shall focus on the adaptive
is the Laplace operator in R which describes the random motion
dynamics of rare predator movement mechanisms in two
of prey, the habitat O is assumed to be a bounded domain in RN
scenarios: (i) when both predators are rare, we determine
with smooth boundary, denoted by qO. n denotes the outward
which one can invade in tougher conditions (i.e., which predator
unit normal vector on qO, and the boundary conditions in (2.1)
can invade with a higher predator intrinsic mortality rate, k);
mean that there is no flux across the boundary of the habitat (as
(ii) if one predator has invaded in the absence of the other,
assumed above).
we determine whether the other predator can subsequently
In order to make fair comparisons between predators’ dispersal
invade.
mechanisms, throughout this paper we assume that
When none of the predators are present, the prey can
d1 ¼ d2 ; a1 ¼ a2 ; k1 ¼ k2 ; b1 ¼ b2 . survive under a variety of assumptions about RðxÞ. Consider the
For the sake of simplicity we drop the subscripts and write these following steady state equation for prey (in the absence of
parameters as d, a, k, b, respectively. predators):
To address the questions outlined above, we study cases where dv Dy þ y½RðxÞ y ¼ 0 in O; ry n ¼ 0 on qO, (3.1)
f i ðR; VÞ ¼ ti R þ ð1 ti ÞV (2.2) where dv 40 is the dispersal rate of the prey species and yðxÞ
for some ti 2 ½0; 1 and i ¼ 1; 2. ti ¼ 1 corresponds a situation in is the density of the prey species at location x. We shall always
which the biased movement of the predator is purely affected by assume that the function RðxÞ is Hölder continuous in Ō, and
R
the gradient of habitat quality (i.e., the gradient of resources). O RðxÞ40. Here Ō is the union of O and qO. Under this
Similarly, ti ¼ 0 corresponds to biased movement affected only by assumption, it is well known that (3.1) has a unique positive
the gradient of prey density. classical solution, denoted by y ¼ yðx; dv Þ for every dv 40 (Cantrell
We will focus on two particular situations. and Cosner, 2003).
Clearly, ðP 1 ; P 2 ; VÞ ¼ ð0; 0; yÞ is always a non-negative non-
(i) First, to compare resource tracking with prey tracking, we trivial steady state solution of (2.3). To study its stability, we
study the case in which 0pt2 ot1 p1. In other words, define k :¼k ða; tÞ such that the linear problem
predator 1 is more of a resource tracker than predator 2,
dDc þ ar½tR þ ð1 tÞy rc þ ayc ¼ k c in O; rc njqO ¼ 0
and predator 2 is more of a prey tracker than predator 1. In
(3.2)
order to make fair comparisons, we assume in this case that
the magnitude of biased movement (relative to random has a positive solution. Such k exists, is unique, and is referred to
movement) is the same for both predators, i.e., a1 ¼ a2 . as a principal eigenvalue (Cantrell and Cosner, 2003). See
(ii) Second, we let a1 4a2 X0 (assuming t1 ¼ t2 2 ½0; 1). For this Appendix A for more discussions about k .
case, the two predators give the same relative weight to prey We first consider the case a1 ¼ a2 (for simplicity we write ai as
tracking versus resource tracking (they have the same biased a), and 0pt2 ot1 p1.
movement mechanism), but they differ in their relative
magnitudes of random versus biased movement. Theorem 1. Suppose that O ¼ ð0; 1Þ, a1 ¼ a2 (denoted as a),
0pt2 ot1 p1, Rx ð0Þ40, Rxx X0 and Rxxx p0 in ð0; 1Þ. Then for small
positive a, k ða; t1 Þ4k ða; t2 Þ. Moreover,
Hence from now on, we will study the following system:
8
>
> P ¼ r ½drPi ai P i r½ti R þ ð1 ti ÞV þ P i ðk þ aVÞ
> i;t
>
< in O ð0; 1Þ;
(i) ð0; 0; yÞ is stable for k4k ða; t1 Þ and unstable for kok ða; t1 Þ.
>
> V t ¼ dv DV þ V½RðxÞ V bP1 bP2 in O ð0; 1Þ; When both predators are rare and the prey are present, the first
>
>
: ½drP a P r½t R þ ð1 t ÞV n ¼ rV n ¼ 0 on qO ð0; 1Þ predator can invade if and only if kok ða; t1 Þ, and the second
i i i i i
predator can invade if and only if kok ða; t2 Þ.
(2.3) (ii) There exists some small d40 independent of a such that for
for i ¼ 1; 2. every small positive a, a branch of steady state solutions of (2.3)
ARTICLE IN PRESS
S.M. Flaxman, Y. Lou / Journal of Theoretical Biology 256 (2009) 187–200 191
with the form ðP 1 ; 0; V 1 Þ bifurcates from ð0; 0; yÞ at k ¼ k ða; t1 Þ,
and it can be parameterized by k for the range k 2 ðk ða; t1 Þ
d; k ða; t1 ÞÞ; another branch of steady state solutions with
the form ð0; P2 ; V 2 Þ bifurcates from ð0; 0; yÞ at k ¼ k ða; t2 Þ,
and it can be parameterized by k for the range
k 2 ðk ða; t2 Þ d; k ða; t2 ÞÞ. Both Pi and V i , i ¼ 1; 2, are positive
functions in O. P2
(iii) Furthermore, for every small positive a, ðP1 ; 0; V 1 Þ is locally
stable for k 2 ðk ða; t1 Þ d; k ða; t1 ÞÞ, i.e., the second predator
cannot invade when rare; however, ð0; P 2 ; V 2 Þ is unstable for
k 2 ðk ða; t2 Þ d; k ða; t2 ÞÞ, i.e., the first predator can invade
when rare.
0
The proofs of all parts of Theorem 1 are given in Appendix A. We
note that the conditions placed on the resource function RðxÞ in 0 P1
Theorem 1 are technical necessities that make the analysis and
proof of the theorem possible. How to relax these assumptions is
an interesting open problem, and one which we will explore in
future work.
192 S.M. Flaxman, Y. Lou / Journal of Theoretical Biology 256 (2009) 187–200
Theorem 2. Suppose that O ¼ ð0; 1Þ, a1 4a2 , t1 ¼ t2 (denoted as t), to achieve, or at least approximate, an evolutionarily stable
and Rx 40 in ½0; 1. Then there exists some small positive constant d distribution. Second, a great deal of theoretical work on habitat
such that if either a2 oa1 pd or a2 od and a1 41=d then selection has focused on trying to elucidate general mechanisms
k ða2 ; tÞok ða1 ; tÞ. Moreover, whenever k ða2 ; tÞok ða1 ; tÞ, that would lead populations to (or prevent them from achieving)
equilibrium (e.g., Abrahams, 1986; Bernstein et al., 1988, 1991;
(i) ð0; 0; yÞ is stable for k4k ða1 ; tÞ, i.e., none of the two predators Schwinning and Rosenzweig, 1990; Hugie and Grand, 1998, 2003;
can invade for k4k ða1 ; tÞ, and unstable for kok ða1 ; tÞ; Ruxton and Humphries, 1999; Cressman et al., 2004; Jackson et al.,
(ii) the first predator can invade if kok ða1 ; tÞ; the second 2004a; Kimbrell and Holt, 2004; Safran, 2004; Cressman and
predator can not invade for k 2 ðk ða2 ; tÞ; k ða1 ; tÞÞ and can Křivan, 2006). Predator habitat selection mechanisms based on
invade only when kok ða2 ; tÞ. knowledge of prey or knowledge of fitness directly are more
complicated than resource tracking and frequently do not lead to
The proof of Theorem 2 is given in Appendix A. stable dynamics and/or the IFD. Resource tracking is simple
and—since it does not involve predators chasing prey directly—
We were able to prove Theorem 2 for some ranges of a, but we do is potentially more likely to produce stable habitat use dynamics
not know whether a1 4a2 always implies that k ða1 ; tÞ4k ða2 ; tÞ. In (although we need to investigate this hypothesis with future
other words, we have been able to prove that, for some ranges of a, work). Third, the details of behavioral mechanisms matter, not
higher values of a are more adaptive than lower values. That is, just for individual fitness, but also for population level patterns
faster movement up the gradient of prey/resources is more adaptive. and the stability of predator–prey systems (Kimbrell and Holt,
However, we have not shown this for all values of a. If larger values 2004; Schreiber and Vejdani, 2006; Abrams et al., 2007; Safran
of a were always more adaptive, from an evolutionary point of view et al., 2007). The details of habitat selection behaviors determine
this would mean that natural selection would favor ever-more-rapid whether habitat selectors can converge on a predicted equilibrium
movements in directions of higher densities of prey/resources. That distribution and whether predator–prey abundances can stabilize.
is, selection would favor rapid biased movement and would disfavor As such, understanding habitat selection mechanisms can inform
any movement mechanisms that caused movements down the questions at a variety of spatial and temporal scales.
resource/prey gradients. One could imagine the latter being adaptive We have developed a very general model of predator–prey
if predators operated in complete isolation of one another. However, spatial distributions in which predators can move (i) randomly,
we suspect that there is some intermediate optimal a for the (ii) by tracking the gradient of the prey’s resource (‘‘resource
following reasons. If a were very large, all the predators in the local tracking’’), and/or (iii) by tracking the gradient of the prey (‘‘prey
population—whether resource- or prey-tracking—would aggregate tracking’’). Our model is not restricted to a finite number of
tightly in a single location (i.e., extreme overmatching). None of the patches or a particular resource distribution (as many previous
equilibrium predator–prey IFD models predict such a distribution to IFD models are). By studying the adaptive dynamics of different
be evolutionarily stable. Furthermore, prey trackers with extremely strategies, we have shown that, in at least some cases, a resource
large a would respond dramatically to even small changes in prey tracking predator may perform better than prey tracking predator.
density. This would likely lead to cyclic, non-equilibrium distribu- This result is not necessarily expected, since, apart from their
tions of predators and prey, as predators would continually chase movement mechanisms, prey trackers and resource trackers were
prey from location to location. assumed to be identical (i.e., they had the same intrinsic mortality
rates, birth rates, etc.). For example, if prey are mobile but
resources are sessile, we might expect that a prey tracker would
4. Discussion often pay higher search costs than a resource tracker. We did not
include such costs in our basic model, but doing so would enhance
Predator–prey habitat selection models fall into two broad any relative advantage the resource tracker has.
categories (Kacelnik et al., 1992): (i) equilibrium models which In addition to demonstrating the effectiveness of resource
make predictions about the nature of evolutionarily stable tracking, our model provides a very general treatment of animal
locations (i.e., simultaneous IFDs) of predators and prey and movement mechanisms. In our model, predator movement is the
(ii) mechanistic models which examine the distributions resulting combined result of random movements and biased movements
from particular movement rules. Since the inception of IFD theory, toward particular environmental features. It could be further
many models of both types have been developed. As discussed generalized to incorporate any number of sources of biased
above, a common prediction from the equilibrium models is that movement (e.g., movement in the direction of potential mates,
the distribution of top predators will match the distribution of movement due to physical features of the environment such as
their prey’s resource (i) more closely than the predators’ wind or water currents). While both predators and prey are
distribution will match that of their own prey and (ii) more mobile in its current formulation, an important extension of the
closely than the prey will match the resources (Rosenheim, 2004; model will be to allow the prey biased movement as well so that
Sih, 2005). This prediction has received some empirical support they can respond to predators in an adaptive fashion. We were
(Bouskila, 2001; Hammond et al., 2007), and it suggests that many unable to produce analytic results for the latter case, but will
predators in nature could achieve optimal distributions simply by explore it in future work. Furthermore, we also need to explore
using a habitat selection mechanism involving tracking the the equilibria (if any) that result from the kinds of movement
locations of the prey’s resource, a mechanism we have called mechanisms we have proposed and whether these are IFDs.
‘‘resource tracking’’. While previous work has suggested that As an example of the generality of our model, consider the
predators may indeed be resource trackers, mechanistic models following. Several previous IFD models (Hugie and Grand, 1998,
have not formally considered the effectiveness or consequences of 2003; Ruxton and Humphries, 1999) have considered random
resource tracking. movements as ‘‘non-ideal’’ explanations that can account for
Investigating resource tracking is important for at least three undermatching (i.e., too few consumers found in good patches
reasons. First, tracking a sessile resource may be much less costly relative to the predicted IFD). Our model provides a different
than trying to track mobile prey, especially if prey are fast moving perspective on matching and random movements. Suppose
and/or unpredictable. In such cases, resource tracking might be a that—as most researchers believe to be true in nature—predators
highly adaptive mechanism providing predators with an easy way have only limited, mostly local information about prey and
ARTICLE IN PRESS
S.M. Flaxman, Y. Lou / Journal of Theoretical Biology 256 (2009) 187–200 193
competitors. Such predators must use simple rules of thumb if appendix works for general domains and might be useful
they are to achieve (nearly) optimal distributions. These rules of for later studies, we shall use a general domain unless otherwise
thumb are most likely to be based on good predictors of fitness specified.
(i.e., the prey gradient or the gradient of the prey’s resource). As
our model shows, if predators always responded rapidly and A.1. Properties of yðxÞ
strongly to local gradients (i.e., large a), predators would likely
aggregate more strongly than optimal in ‘‘good’’ areas (i.e., In this subsection of the appendix we discuss several proper-
overmatching). Conversely, if they responded only weakly to local ties of yðxÞ which are used in our analysis.
gradients, they would be more uniformly distributed than is
optimal (i.e., undermatching). It is thus a balance between Lemma A.1. Suppose that R is a non-constant function. Then
random and biased movements that can lead a population to an min RoyðxÞo max R
evolutionarily stable distribution. Thus, while previous models Ō Ō
have seen random movements as causing ‘‘mistakes’’ which for every x 2 Ō.
lead to undermatching deviations from the IFD, our model
highlights how a single parameter ðaÞ can tune the relative Proof. If minŌ Rp0, the first inequality automatically holds.
influences of random and biased movement, and thus how a Hence we may assume that R40 in Ō. Choose some x0 2 Ō
single movement rule could evolve to produce matching, under- such that yðx0 Þ ¼ minŌ y. By Eq. (3.1) and Proposition 2.2
matching, or overmatching, depending on what is favored by of Lou and Ni (1996), we have yðx0 ÞXRðx0 ÞXminŌ R, i.e.,
selection. minŌ yXminŌ R. Set w ¼ y minŌ R, which is non-negative in O.
Finally, an important conclusion is that very simple rules can It remains to show that w40 in Ō. We argue by contradiction. If
produce adaptive behavior. Without knowing the locations of prey not, suppose that wðx0 Þ ¼ 0 for some x0 2 Ō. Direct calculation
or conspecifics, but simply by responding to local environmental shows that
gradients, predators might be able (at least approximately) to
achieve distributions that game theory models predict to be dv Dw þ w R 2 min R w ¼ min R R min R p0
Ō Ō Ō
evolutionarily stable. Simply by tuning the strength of biased
movements, selection can fashion behavioral mechanisms to suit in O. Since wX0, by choosing some positive constant C such that
a wide variety of habitat selection challenges. h:¼R 2minŌ R w Cp0 in Ō we have
dv Dw þ hðxÞwp Cwp0
in O. If x0 2 qO, by the Hopf boundary lemma (Protter and
5. Conclusions
Weinberger, 1984) we have ðrw nÞðx0 Þ ¼ 0, which contradicts the
zero Neumann boundary condition of w; if x0 2 O, by the strong
Our model is a step toward building a more general and
maximum principle (Protter and Weinberger, 1984) we see that
predictive theory of predator movement mechanisms and predator–
w 0 in O, i.e., y is a positive constant. Then, by (3.1) we see that
prey interactions than currently exists. We have synthesized
R is also a positive constant, which contradicts the assumption
previous equilibrium and mechanistic IFD models to investigate
that R is a non-constant function. Hence, w40 in Ō. The second
the effectiveness of simple movement mechanisms that could—with
inequality can be proved similarly, so we omit it. &
appropriate tuning of parameters—allow predators to achieve or
approximate evolutionarily stable distributions. These results lay Lemma A.2. Suppose that O ¼ ð0; 1Þ, Rx X0 and Rx c0. Then yx 40
essential groundwork for future investigations of both ecological and in ð0; 1Þ.
evolutionary consequences of optimal predator movement. Several
intriguing questions await answers. For example, does resource Proof. Differentiate Eq. (3.1) with respect to x and set w ¼ yx .
tracking by predators impact the dynamics and equilibrium of prey Then
habitat selection? Does resource tracking eliminate cycles in dv wxx þ ðR 2yÞw ¼ yRx in ð0; 1Þ; wð0Þ ¼ wð1Þ ¼ 0. (A.1)
behavioral and/or population dynamics? Does resource tracking by
Set u ¼ w=y. Then u satisfies
predators create the opportunity for prey to use similarly simple
rules to optimally balance resource gathering with predation risk? dv uxx þ 2dv ðln yÞx ux yu ¼ Rx p0 in ð0; 1Þ; uð0Þ ¼ uð1Þ ¼ 0.
These questions will be foci of future studies. By the maximum principle, u must attain its non-positive
minimum at the boundary. Since uð0Þ ¼ uð1Þ ¼ 0, we have uX0
in O. To show that u40 in ð0; 1Þ, we argue by contradiction. If
Acknowledgments
uðx0 Þ ¼ 0 for some x0 2 ð0; 1Þ, by the strong maximum principle
we have that u 0 in ð0; 1Þ, i.e., w 0 in O. Then, by (A.1) we see
S.M.F. thanks Simon Levin, Princeton University’s Council on that Rx 0, which is a contradiction since R is a non-constant
Science and Technology, Princeton’s Department of Ecology and function. Hence u40 in ð0; 1Þ, i.e., yx 40 in ð0; 1Þ. &
Evolutionary Biology, and the Santa Fe Institute for support. Y.L.
thanks Simon Levin and Princeton’s Department of Ecology and Lemma A.3. Suppose that O ¼ ð0; 1Þ, Rx ð0Þ40, Rxx X0 and Rxxx p0
Evolutionary Biology for the hospitality during his visit. Y.L. is in ð0; 1Þ. Then Rx Xyx and Rx cyx in ½0; 1.
partially supported by NSF Grant DMS-0615845. The authors are
Proof. We argue by contradiction. If not, suppose that
grateful to two anonymous referees for their careful readings and
min½0;1 ðRx yx Þo0. By assumption, Rx 40 in ½0; 1. Since
constructive comments and suggestions that significantly im-
yx ð0Þ ¼ yx ð1Þ ¼ 0, Rx ð0Þ40 and Rx ð1Þ40, we see that there exists
proved the manuscript.
some x0 2 ð0; 1Þ such that Rx ðx0 Þ yx ðx0 Þ ¼ min½0;1 ðRx yx Þo0. In
particular,
Appendix A. Proofs of theorems and other details of analysis Rxx ðx0 Þ ¼ yxx ðx0 Þ; Rxxx ðx0 ÞXyxxx ðx0 Þ.
By Eq. (3.1), we have
While our main results (Theorems 1 and 2) are restricted to a
one-dimensional domain, since some of the analysis in this yðx0 Þ½yðx0 Þ Rðx0 Þ ¼ dv yxx ðx0 Þ ¼ dv Rxx ðx0 ÞX0,
ARTICLE IN PRESS
194 S.M. Flaxman, Y. Lou / Journal of Theoretical Biology 256 (2009) 187–200
not, suppose that j1 ¼ j2 0 and jv c0. Then l and jv satisfies It is rather difficult to determine the sign of k1 ðtÞ and its
dv Djv þ ðR 2yÞjv ¼ ljv in O; rjv n ¼ 0 on qO. derivative in general. From now on we restrict to the case when O
is an interval and R is monotone.
By (3.1) and the positivity of y, the smallest eigenvalue of the
operator dv D ðR yÞ with the zero Neumann boundary Lemma A.6. Suppose that O ¼ ð0; 1Þ, Rx X0 and Rx c0. Then c0;x 40
condition is equal to zero. By the comparison principle for in ð0; 1Þ.
eigenvalues and the positivity of y, the smallest eigenvalue of Proof. Note that c0 satisfies
the operator dv D ðR 2yÞ with the zero Neumann boundary
condition is positive. By the assumption ReðlÞo0, we must have dc0;xx ¼ ðk0 ayÞc0 in ð0; 1Þ; c0;x ð0Þ ¼ c0;x ð1Þ ¼ 0.
jv 0, i.e., ðj1 ; j2 ; jv Þ ¼ ð0; 0; 0Þ. This contradiction shows that By integrating the above equation, we see that
either j1 c0 or j2 c0. If j1 c0, the principle eigenvalue l1 Z 1
satisfies l1 pReðlÞo0, which implies that minðl1 ; l2 Þo0; the case
ðk0 ayÞc0 ¼ 0.
j2 c0 is similar. 0
Suppose that ð0; 0; yÞ is stable. We argue by contradiction, i.e., c0 is positive, so k0 ay must change sign. As k0 ay changes
suppose that minfl1 ; l2 gp0. Without loss of generality we assume sign in ð0; 1Þ and y is strictly increasing (Lemma A.2), k0 ay is
that l1 p0 and let j1 40 be an eigenfunction of l1 . Consider the strictly decreasing and changes sign exactly once, say, at some
ARTICLE IN PRESS
S.M. Flaxman, Y. Lou / Journal of Theoretical Biology 256 (2009) 187–200 195
x 2 ð0; 1Þ. Therefore, c0;xx 40 in ð0; x Þ and c0;xx o0 in ðx ; 1Þ. Since Since k1 ðtÞ is an affine function of t, Lemmas A.8 and A.9 imply
c0;x ð0Þ ¼ c0;x ð1Þ ¼ 0, we see that c0;x 40 in ð0; 1Þ. & that the following holds for k1 ðtÞ:
By Lemma A.3, Rx yx X0 and is not identically equal to zero. Lemma A.10. Suppose that O ¼ ð0; 1Þ, Rx X0 and Rx c0. Then
Since c0;x 40 by Lemma A.6 and c0 40 in ð0; 1Þ, we see from k1 ðtÞ40 for every t 2 ½0; 1.
Eq. (A.9) that the following holds:
Corollary A.7. Suppose that O ¼ ð0; 1Þ, Rx ð0Þ40, Rxx X0 and Rxxx p0 A.4. Asymptotic behavior of k ða; tÞ for large a
in ð0; 1Þ. Then dk1 =dt40.
Recall that lðaÞ denotes the principal eigenvalue of the
Proof of part (i) of Theorem 1. By Corollary A.7, dk1 =dt40. This problem (A.3). We will need the following asymptotic result for
implies that k ða; t1 Þ4k ða; t2 Þ for small positive a. The rest of part lðaÞ as established by Chen and Lou (2008):
(i) follows from Lemma A.5 and its proof. &
Lemma A.11. Assume that m 2 C 2 ðŌÞ and c 2 CðŌÞ and all critical
The following result concerns k1 ð1Þ, and is an immediate points of m are non-degenerate. Let M be the set of points of local
consequence of Lemma A.6: maximum of m. Then
Lemma A.8. Suppose that O ¼ ð0; 1Þ, Rx X0 and Rx c0. Then lim lðaÞ ¼ min cðxÞ.
a!1 x2M
k1 ð1Þ40.
For precise definitions of critical points (interior or boundary)
For k1 ð0Þ, we have a more general result which holds for all and their non-degeneracy, see Chen and Lou (2008, Sub-
convex domains. section 2.1).
Lemma A.9. Suppose that O is convex. Then k1 ð0Þ40.
Lemma A.12. Suppose that O ¼ ð0; 1Þ, Rx 40 in ½0; 1. Then for every
Proof. Apply the r operator onto (A.8) and then take the dot t 2 ½0; 1, all critical points of tR þ ð1 tÞy are non-degenerate and
product with rc0 we have its set of local maxima consists of only x ¼ 1.
drðDc0 Þ rc0 ¼ ðk0 ayÞjrc0 j2 aðry rc0 Þc0 .
Proof. By the assumption we have Rx 40 in ½0; 1 and Lemma A.2
Let Hess c0 denote the Hessian of c0 , i.e., the matrix gives yx 40 in ð0; 1Þ. Hence the only possible critical points of
2 P 2
ðq c0 =qxi qxj Þ1pi;jpN and define jHess c0 j2 ¼ N 2
i;j¼1 ðq c0 =qxi qxj Þ . tR þ ð1 tÞy are x ¼ 0; 1, and 1 is the only local maximum. For the
Substituting the identity non-degeneracy of the critical points x ¼ 0; 1, since d=dx½tR þ ð1
tÞy ¼ 0 at x ¼ 0; 1 if and only if t ¼ 0, it suffices to further show
rðDc0 Þ rc0 ¼ jHess c0 j2 þ 12Dðjrc0 j2 Þ
that yxx ð1Þo0 and yxx ð0Þ40. Since R is not a constant function,
into the previous equation and integrating in O, we have Lemma A.1 implies that maxŌ yomaxŌ R, which together with the
Z Z monotonicity of R and y imply that yð1ÞoRð1Þ. This along with
a ðry rc0 Þc0 ¼ ½djHess c0 j2 þ ðk0 ayÞjrc0 j2 (3.1) implies that yxx ð1Þo0. Similarly, yxx ð0Þ40. &
O O
Z
d By Lemmas A.11 and A.12, we have the following asymptotic
Dðjrc0 j2 Þ.
2 O behavior of k ða; tÞ:
By the definition of c0 , since c0 40 in O, we see that the smallest
Lemma A.13. Suppose that O ¼ ð0; 1Þ, Rx 40 in ½0; 1. Then
eigenvalue of the operator dD þ ðk0 ayÞ with zero Neumann
boundary condition is zero (with c0 as its eigenfunction), which lim k ða; tÞ ¼ ayð1Þ.
a!1
implies that
Z Z Proof of Theorem 2. By Lemma A.10 we see that qk =qað0; tÞ40
d jrcj2 X ðay k0 Þc
2 for every t 2 ½0; 1. Hence there exists d1 40 such that if
O O 0pa2 oa1 od1 and 0ptp1 we have k ða1 ; tÞ4k ða2 ; tÞ. For large
for any c 2 H1 ðOÞ, with equality holds if and only if c is a scalar a, by Lemma A.13 we see that k ða; tÞ ! ayð1Þ as a ! 1. Since
multiplier of c0 . Setting c ¼ c0;xi for every 1pipN and summing ay k0 must change sign, we see that k ð0; tÞ k0 oa maxŌ y ¼
up in i, we have ayð1Þ. Hence there exists some small d2 40 such that if a2 2 ½0; d2 Þ
Z Z and a1 41=d2 , k ða1 ; tÞ4k ða2 ; tÞ. Set d ¼ minfd1 ; d2 g, this finishes
d jHess c0 j2 4 ðay k0 Þjrc0 j2 , the proof of the first part of Theorem 2. The rest follows from
O O
Lemma A.5 and its proof. &
where the strict inequality follows from the fact that c0;xi is
not a scalar multiplier of c0 for at least some i (otherwise if c0;xi ¼ A.5. Local bifurcation of steady states
gi c0 for
P
some constants gi and every 1pipn, then c0 ¼
C expð N gi xi Þ for some positive constant C. Hence Dc0 ¼
P i¼1 We are interested in a complete classification of all non-
ð N 2
i¼1 gi Þc0 . Substituting this into (A.8) we would obtain that y negative steady state solutions of (2.3) for k close to k ða; ti Þ
is constant, a contradiction).
(i ¼ 1; 2) and ðP1 ; P2 ; VÞ close to ð0; 0; yÞ. As we will show in this
Since qc0 =qn ¼ 0 on qO and O is convex, by a result of Casten subsection, for small positive a, there are two types of such
and Holland (1978) and Matano (1979), we have qðjrc0 j2 Þ=qnp0 solutions, one with the form ðP 1 ; 0; V 1 Þ and the other of the form
on qO. In particular, this implies that ð0; P2 ; V 2 Þ, which are close to ð0; 0; yÞ near k ¼ k ða; t1 Þ and
Z Z
k ¼ k ða; t2 Þ, respectively. We first show that under suitable
Dðjrc0 j2 Þ ¼ qðjrc0 j2 Þ=qnp0. conditions there is no interior equilibria close to ð0; 0; yÞ when k
O qO
is close to k ða; ti Þ (i ¼ 1; 2).
Hence, we have
Z Lemma A.14. Suppose that dk1 =dta0. There exist some small
ðry rc0 Þc0 40, positive constant d3 such that (2.3) has no interior equilibria
O ðP 1 ; P 2 ; VÞ satisfying kP1 kC 2 ðŌÞ þ kP2 kC 2 ðŌÞ þ kV ykC 2 ðŌÞ pd3 for a 2
i.e., k1 ð0Þ40. & ð0; d3 Þ and k 2 ðk ða; t1 Þ d3 ; k ða; t1 ÞÞ [ ðk ða; t2 Þ d3 ; k ða; t2 ÞÞ.
ARTICLE IN PRESS
196 S.M. Flaxman, Y. Lou / Journal of Theoretical Biology 256 (2009) 187–200
dDjl þ ðay k0 Þjl ¼ 0 in O; rjl n ¼ 0 on qO Lemma A.15. There exists some small d4 40, some function k̃ðsÞ 2
C 2 ðd4 ; d4 Þ with k̃ð0Þ ¼ k such that all non-negative steady state
which contradicts our assumption dk1 =dta0. & Note that we choose s40 such the corresponding P; V are both
positive. Also note that we can choose d4 independent of both
By Lemma A.14 and Corollary A.7 we see that under the t 2 ½0; 1 and all small positive a. This fact will be useful later.
assumptions of Theorem 1 there is no interior equilibria for Finally, the functions ci ðsÞ are continuous differentiable for small s
ðk; P 1 ; P 2 ; VÞ close to ðk ða; ti Þ; 0; 0; yÞ (i ¼ 1; 2) and small positive a. (see Crandall and Rabinowitz, 1971, p. 325).
This reduces the local bifurcation analysis of (2.3) to that of some It is important to determine the bifurcation direction of the
one predator–one prey systems. Hence, it suffices to classify all solution branch. In this connection, we have
steady state solutions of the following reduced system (i.e., one 0
dimension less than original system (2.3)): Lemma A.16. For small positive a, k̃ ð0Þo0.
8
Proof. Substituting the expansion (A.16) into the first equation of
< r ½drP aP r½tR þ ð1 tÞV þ Pðk þ aVÞ ¼ 0 in O;
>
dv DV þ V½RðxÞ V bP ¼ 0 in O; (A.12), applying (A.15), and dividing the result by s we have
(A.12)
>
: ½drP aP r½tR þ ð1 tÞV n ¼ rV n ¼ 0 on qO
k k̃
ðj þ sc1 Þ þ aj jv að1 tÞr ðj rjv Þ
s
for ðk; P; VÞ close to ðk ða; tÞ; 0; yÞ. To find such classifications, we
þ fr ½drc1 ac1 r½tR þ ð1 tÞy þ ðay k Þc1 g ¼ OðsÞ.
will apply the local bifurcation theorem (Crandall and Rabinowitz,
(A.17)
1971, Theorem 1.7). In order to use such abstract bifurcation
result, we need to introduce some functional spaces and Multiplying (A.17) by j eða=dÞ½tRþð1tÞy, integrating in O, using
operators. For g 2 ð0; 1Þ, set the equation of j and no-flux boundary conditions for j and c1 ,
applying k̃ð0Þ ¼ k , and finally passing to the limit we have
X ¼ fðP; VÞ 2 C 2;g ðŌÞ C 2;g ðŌÞ: Z
0
ðdrP atPrRÞ n ¼ rV n ¼ 0 on qOg k̃ ð0Þ eða=dÞ½tRþð1tÞy ðj Þ2
O
Z
and Y ¼ C g ðŌÞ C g ðŌÞ. Define the operator Fðk; P; VÞ :
ð0; 1Þ X ! Y by ¼a eða=dÞ½tRþð1tÞy jv ðj Þ2
O
! Z
r ½drP aPr½tR þ ð1 tÞV þ Pðk þ aVÞ þ að1 tÞ j rðeða=dÞ½tRþð1tÞy j Þ rjv . (A.18)
Fðk; P; VÞ ¼ . O
dv DV þ V½RðxÞ V bP
By elliptic regularity we have j ! c0 and jv ! cv in C 1 ðŌÞ as
(A.13) a ! 0, where cv is the unique solution of
Note that Fðk; 0; yÞ 0 for all k. Clearly, the derivatives Dk F,
dv Dcv þ cv ðR 2yÞ ¼ byc0 in O; rcv njqO ¼ 0. (A.19)
DðP;VÞ F, and Dk DðP;VÞ F exist and are continuous near ðk; 0; yÞ.
For the simplicity of notation, we abbreviate k ða; tÞ as Hence,
k . We first check that the operator DðP;VÞ Fjðk ;0;yÞ has one- R
0 a c20 cv
dimensional kernel and Y=RangeðDðP;VÞ Fjðk ;0;yÞ Þ is also one lim k̃ ð0Þ ¼ RO 2
. (A.20)
a!0þ
dimensional, where RangeðDðP;VÞ Fjðk ;0;yÞ Þ denotes the range O c0
ARTICLE IN PRESS
S.M. Flaxman, Y. Lou / Journal of Theoretical Biology 256 (2009) 187–200 197
By (3.1) and the positivity of y and c0 , similar to the proof of and V 1 yo0 in O. This establishes our assertion that P1 0,
Lemma 7.2 we can show that cv o0 in O. This implies that the from which it follows that V 1 y. Hence, P 1 ! 0 and V 1 ! y in
right-hand side of (A.20) is negative. Hence, for positive small a, C 1 ðŌÞ.
0
we have k̃ ð0Þo0. & Set P̃1 ¼ P1 =kP1 k1 . By elliptic regularity we may assume that
P̃1 ! P^ 1 , where P^ 1 X0 satisfies kP^ 1 k1 ¼ 1 and
The following result is the main result of this subsection:
Lemma A.17. Suppose that dk1 =dta0. Given any 0pt2 ot1 p1, dDP^ 1 þ P^ 1 ðay k0 Þ ¼ 0 in O; rP^ 1 njqO ¼ 0
there exists some small d5 40 such that 0oaod5 , all solutions of as V 1 ! y. Hence, P^ 1 c0 , i.e., P 1 =kP 1 k1 ! c0 . As P1 ! 0 and
(2.3) close to ð0; 0; yÞ for the parameter range k 2 ðk ða; t1 Þ
V 1 ! y, we can similarly show that lða; kÞ ! 0 and c ! c0 in
d5 ; k ða; t1 ÞÞ [ ðk ða; t2 Þ d5 ; k ða; t2 ÞÞ consist exactly of three
C 1 ðŌÞ as ða; kÞ ! ð0; k0 Þ. &
branches in the space C 2 ðŌÞ C 2 ðŌÞ C 2 ðŌÞ: ð0; 0; yÞ, ðP 1 ; 0; V 1 Þ
with k 2 ðk ða; t1 Þ d5 ; k ða; t1 ÞÞ and ð0; P2 ; V 2 Þ with
Lemma A.19. The following holds:
k 2 ðk ða; t2 Þ d5 ; k ða; t2 ÞÞ, where P i ; V i ; i ¼ 1; 2; are positive
Z
functions which are infinitely differentiable in k. lða; kÞ t1 t2
lim ¼ R c0 rðR yÞ rc0 . (A.23)
ða;kÞ!ð0;k0 Þ 2 a c0 O
Proof. By Lemma A.14 we see that there is no interior equilibria O
for ðk; P1 ; P 2 ; VÞ close to ðk ða; ti Þ; 0; 0; yÞ (i ¼ 1; 2) and small Proof. Rewriting the equation of c as
positive a. We then apply Lemmas 7.15 and 7.16 on (2.3)
with either P1 0 or P2 0, respectively. Finally, the smooth- dr ½eða=dÞf 2 ðR;V 1 Þ rðeða=dÞf 2 ðR;V 1 Þ cÞ þ cðk þ aV 1 Þ ¼ lða; kÞc,
ness of the solutions with respect to k follows from Lemma A.16
multiplying it by eða=dÞf 1 ðR;V 1 Þ P 1 and integrating in O, we have
and the implicit function theorem (see Crandall and Rabinowitz,
Z
1971, Theorem 1.18 for such an argument). &
d eða=dÞf 2 ðR;V 1 Þ r½eða=dÞf 1 ðR;V 1 Þ P 1 r½eða=dÞf 2 ðR;V 1 Þ c
Proof of part (ii) of Theorem 1. By Corollary A.7, dk1 =dt40. ZO
Then, part (ii) follows from Lemma A.17. & þ eða=dÞf 1 ðR;V 1 Þ P 1 cðk þ aV 1 Þ
O
Z
A.6. Stability of steady states ðP 1 ; 0; V 1 Þ and ð0; P 2 ; V 2 Þ ¼ lða; kÞ eða=dÞf 1 ðR;V 1 Þ P1 c. (A.24)
O
We first study the stability of ðP 1 ; 0; V 1 Þ for small positive a and Similarly, rewrite the equation of P1 as
k close to k0 . To this end, consider the corresponding linear
dr ½eða=dÞf 1 ðR;V 1 Þ rðeða=dÞf 1 ðR;V 1 Þ P1 Þ þ P1 ðk þ aV 1 Þ ¼ 0,
eigenvalue problem
8 multiplying it by eða=dÞf 2 ðR;V 1 Þ c and integrating in O, we have
>
> r ½drj1 aj1 rf 1 ðR; V 1 Þ að1 t1 ÞP1 rjv þ j1 ðk þ aV 1 Þ
>
> Z
< þaP 1 jv ¼ lj1 ;
d eða=dÞf 1 ðR;V 1 Þ r½eða=dÞf 1 ðR;V 1 Þ P 1 r½eða=dÞf 2 ðR;V 1 Þ c
>
> r ½drj2 aj2 rf 2 ðR; V 1 Þ þ j2 ðk þ aV 1 Þ ¼ lj2 ; ZO
>
>
: dv Dj þ j ½RðxÞ 2V bP bV j bV j ¼ lj
v v 1 1 1 1 1 2 v þ eða=dÞf 2 ðR;V 1 Þ P 1 cðk þ aV 1 Þ ¼ 0. (A.25)
O
(A.21)
We subtract Eq. (A.24) from Eq. (A.25) and divide it by kP 1 k1 to
in O and corresponding boundary conditions
find
½drjl ajl rf l ðR; V 1 Þ n ¼ rjv n ¼ 0 on qO, (A.22) Z
P
where l ¼ 1; 2. Note that for the boundary condition of j1 , the
lða; kÞ eða=dÞf 1 ðR;V 1 Þ 1 c
O kP 1 k1
term að1 t1 ÞP 1 ðrjv nÞ vanishes as rjv n ¼ 0 on qO. Z
ða=dÞf 2 ðR;V 1 Þ P
We first consider the principal eigenvalue, denoted by lða; kÞ, of ¼ d ½e eða=dÞf 1 ðR;V 1 Þ r eða=dÞf 1 ðR;V 1 Þ 1
O kP 1 k1
the linear problem Z
ða=dÞf 2 ðR;V 1 Þ ða=dÞf 2 ðR;V 1 Þ ða=dÞf 1 ðR;V 1 Þ
r½e c þ ½e e
r ½drc acrf 2 ðR; V 1 Þ þ cðk þ aV 1 Þ ¼ lc in O, O
198 S.M. Flaxman, Y. Lou / Journal of Theoretical Biology 256 (2009) 187–200
This completes the proof. & As cv o0 in O, Z1 o0 for small positive a and k close to k0 .
If j1 0, then jv c0 and it satisfies
Proof of part (iii), Theorem 1. We first show that for small a40
and k close to k0 , if system (A.21) has an eigenvalue l with non- dv Djv þ ½RðxÞ 2V 1 bP 1 jv ¼ Z1 jv in O; rjv njqO ¼ 0.
positive real part, then j2 0. We argue by contradiction: if not, Since P 1 ! 0, V 1! y as ða; kÞ ! ð0; k0 Þ, and the smallest eigen-
Corollary A.7, Eq. (A.9) and Lemma A.19 imply that for small value of the operator dv D þ ðR þ 2yÞ with the zero Neumann
positive a and k close to k0 , lða; kÞ40, which is a contradiction.
boundary condition is positive, we see that Z1 o0. Hence, all
Hence, it suffices to consider the reduced linear eigenvalue
eigenvalues of (A.21)–(A.22) must have positive real part, i.e.,
problem (i.e., (A.21) with j2 ¼ 0)
ðP 1 ; 0; V 1 Þ is linearly stable.
8
> r ½drj1 aj1 rf 1 ðR; V 1 Þ að1 t1 ÞP1 rjv þ j1 ðk þ aV 1 Þ For the instability of ð0; P2 ; V 2 Þ, the corresponding linear
<
þaP 1 jv ¼ lj1 ; eigenvalue problem is
>
: d Dj þ j ½RðxÞ 2V bP bV j ¼ lj 8
v v v 1 1 1 1 v >
> r ½drj1 aj1 rf 1 ðR; V 2 Þ þ j1 ðk þ aV 2 Þ ¼ lj1 ;
>
>
(A.27) < r ½drj aj rf 2 ðR; V Þ að1 t2 ÞP rj þ j ðk þ aV Þ
2 2 2 2 v 2 2
>
> þaP 2 jv ¼ lj2 ;
in O and corresponding no-flux boundary conditions as in (A.22). >
>
: dv Dj þ j ½RðxÞ 2V bP bV j bV j ¼ lj
For g 2 ð0; 1Þ, set v v 2 2 2 1 2 2 v
(A.30)
X 1 ¼ fðj1 ; jv Þ 2 C 2;g ðŌÞ C 2;g ðŌÞ : ½drj1 aj1 rf 1 ðR; V 1 Þ n
in O with corresponding no-flux boundary conditions. We shall
¼ rjv n ¼ 0 on qOg.
show that (A.30) has a negative eigenvalue.
Define operators T a;k ; T 0;k0 : X 1 ! Y by Let l:¼lða; kÞ denote the principal eigenvalue of the linear
! problem
j2
T a;k
jv r ½dj1 aj1 rf 1 ðR; V 2 Þ þ j1 ðk þ aV 2 Þ ¼ lj1 in O,
!
r ½drj1 aj1 rf 1 ðR; V 1 Þ að1 t1 ÞP 1 rjv þ ðk þ aV 1 Þj1 þ aP1 jv ½drj1 aj1 rf 1 ðR; V 2 Þ njqO ¼ 0
¼
dv Djv þ ½RðxÞ 2V 1 bP1 jv bV 1 j1
with kj1 k1 ¼ 1. Then, the following limit holds (the proof is
(A.28) identical to that of Lemma A.19 by switching f 1 with f 2 and V 1
with V 2 ):
and
Z
! ! lða; kÞ t2 t1
j1 dDj1 þ ðk0 þ ayÞj1 lim ¼ R c0 rðR yÞ rc0 . (A.31)
T 0;k0 ¼ . (A.29) ða;kÞ!ð0;k0 Þ a 2
O c0 O
jv dv Djv þ ½RðxÞ 2yjv byj1
Since t2 ot1 , we see that lða; kÞo0 for all small positive a and k
By Lemma A.18, P 1 ! 0, V 1 ! y in C 1 ðŌÞ as ða; kÞ ! ð0; k0 Þ. Hence, close to k0 .
T a;k ! T 0;k0 uniformly in operator norm as ða; kÞ ! ð0; k0 Þ. It is We further show that lða; kÞ is also an eigenvalue for (A.30) for
easy to check that the kernel of T 0;k0 is spanned by ðc0 ; cv Þ, and small positive a and k close to k0 . For g 2 ð0; 1Þ, set
zero is a K-simple eigenvalue of T 0;k0 (where the operator K is the
X 2 ¼ fðj2 ; jv Þ 2 C 2;g ðŌÞ C 2;g ðŌÞ : ½drj2 aj2 rf 2 ðR; V 2 Þ n
canonical injection from X 1 into Y). Hence, for ða; kÞ close to ð0; k0 Þ,
¼ rjv n ¼ 0 on qOg.
there is a unique K-simple eigenvalue Z1 :¼Z1 ða; kÞ of T a;k with
limða;kÞ!ð0;k0 Þ Z1 ¼ 0. To establish the stability of ðP 1 ; 0; V 1 Þ, it Define operators La;k ; L0;k0 : X 2 ! Y by
suffices to show that ReðZ1 Þo0 for ða; kÞ close to ð0; k0 Þ. Let !
j2
ðj1 ; jv Þ be an eigenfunction of Z1 , i.e., ðj1 ; jv Þ satisfy (A.27) with La;k
jv
l ¼ Z1 . r ½drj2 aj2 rf 2 ðR; V 2 Þ að1 t2 ÞP2 rjv þ ðk þ aV 2 þ lÞj2 þ aP2 jv
!
If j1 c0, after scaling we may assume that kj1 k1 ¼ 1 and j1 is ¼
dv Djv þ ½RðxÞ 2V 2
bP 2 þ ljv
bV 2 j2
,
positive somewhere in O. Since P1 ! 0, V 1 ! y and Z1 ! 0, we
(A.32)
can show that j1 ! c0 and jv ! cv in C 1 ðŌÞ as ða; kÞ ! ð0; k0 Þ,
where cv is the unique solution of (A.19). Multiplying the where l ¼ lða; kÞ, and
equation of j1 by eða=dÞf 1 ðR;V 1 Þ P 1 and the equation of P 1 by ! !
j2 dDj2 þ ðk0 þ ayÞj2
eða=dÞf 1 ðR;V 1 Þ j1 , subtracting and integrating in O, similar as in the L0;k0
jv ¼ dv Djv þ ½RðxÞ 2yjv byj2 . (A.33)
proof of Lemma A.19, after integration by parts and rearrangement
we have Similarly as in Lemma A.18, we can show that as ða; kÞ ! ð0; k0 Þ,
Z
Z
P2 ! 0, V 2 ! y in C 1 ðŌÞ. Also note that l ! 0 by (A.31). Hence,
Z1 eða=dÞf 1 ðR;V 1 Þ P1 j1 ¼ a ðP1 Þ2 jv eða=dÞf 1 ðR;V 1 Þ La;k ! L0;k0 uniformly in operator norm as ða; kÞ ! ð0; k0 Þ. It is easy
O O
Z to check that the kernel of L0;k0 is spanned by ðc0 ; cv Þ, and zero is a
þ að1 t1 Þ P 1 rðeða=dÞf 1 ðR;V 1 Þ P 1 Þ rjv .
O K-simple eigenvalue of L0;k0 (where the operator K is the canonical
injection from X 2 into Y). Hence, for ða; kÞ close to ð0; k0 Þ, there
Dividing the above equation by kP 1 k21, since P1 ! 0, is a unique K-simple eigenvalue Z2 ¼ Z2 ða; kÞ of La;k with
P 1 =kP 1 k1 ! c0 , V 1 ! y, j1 ! c0 and jv ! cv in C 1 ðŌÞ limða;kÞ!ð0;k0 Þ Z2 ¼ 0.
ARTICLE IN PRESS
S.M. Flaxman, Y. Lou / Journal of Theoretical Biology 256 (2009) 187–200 199
Claim. For ða; kÞ close to ð0; k0 Þ, Z2 o0. Let ðj2 ; jv Þ denote an Cosner, C., Lou, Y., 2003. Does movement toward better environments always
eigenfunction of Z2 . Multiplying the equation of j2 by benefit a population? J. Math. Anal. Appl. 277, 489–503.
Crandall, M.G., Rabinowitz, P.H., 1971. Bifurcation from simple eigenvalues. J.
P 2 eða=dÞf 2 ðR;V 2 Þ and the equation of P 2 by j2 eða=dÞf 2 ðR;V 2 Þ, subtracting Functional Anal. 8, 321–340.
and integrating the resulting equation in O, using Green’s formula Cressman, R., Křivan, V., 2006. Migration dynamics for the ideal free distribution.
and the no-flux boundary conditions of P 2 and j2 , we have Am. Nat. 168, 384–397.
Cressman, R., Křivan, V., Garay, J., 2004. Ideal free distributions, evolutionary
Z Z games, and population dynamics in multiple-species environments. Am. Nat.
164, 473–489.
Z2 j2 P2 eða=dÞf 2 ðR;V 2 Þ pa ðP 2 Þ2 jv eða=dÞf 2 ðR;V 2 Þ
Farnsworth, K.D., Beecham, J.A., 1997. Beyond the ideal free distribution: more
O O
Z general models of predator distribution. J. Theor. Biol. 187, 389–396.
þ að1 t2 Þ P2 rðP 2 eða=dÞf 2 ðR;V 2 Þ Þ rjv , Flaxman, S.M., deRoos, C.A., 2007. Different modes of resource variation provide a
O critical test of ideal free distribution models. Behav. Ecol. Sociobiol. 61,
(A.34) 877–886.
Flaxman, S.M., Reeve, H.K., 2006. Putting competition strategies into ideal free
R ða=dÞf 2 ðR;V 2 Þ distribution models: habitat selection as a tug of war. J. Theor. Biol. 243,
where we dropped the term l O j 2 P2 e
as l is negative
587–593.
for ða; kÞ close to ð0; k0 Þ. Similar as in Lemma A.18, we can further Fox, L.R., Eisenbach, J., 1992. Contrary choices: possible exploitation of enemy-free
show that P2 =kP2 k1 ! c0 , and under suitable scalings, j2 ! c0 space by herbivorous insects in cultivated vs. wild crucifers. Oecologia 89,
and jv ! cv in C 1 ðŌÞ as ða; kÞ close to ð0; k0 Þ. Dividing (A.34) by 574–579.
Fretwell, S.D., Lucas, H.L., 1969. On territorial behavior and other factors
kP 2 k21 , we have influencing habitat selection in birds. Acta Biotheor. 19, 16–36.
R Garcia-Domingo, J.L., Saldana, J., 2007. Food-web complexity emerging from
Z2 a Oc20 cv ecological dynamics on adaptive networks. J. Theor. Biol. 247, 819–826.
lim sup p R 2
o0, Gilbarg, D., Trudinger, N., 1983. Elliptic Partial Differential Equations of Second
ða;kÞ!ð0;k0 Þ kP 2 k1 O c0 Order, second ed. Springer, Berlin.
Godfray, H.C.J., 1994. Parasitoids: Behavioral and Evolutionary Ecology. Princeton
where the last inequality follows from cv o0 in O. This establishes University Press, Princeton.
our assertion that Z2 o0 for ða; kÞ ! ð0; k0 Þ. In particular, since La;k Grindrod, P., 1988. Models of individual aggregation or clustering in single and
has no other eigenvalues close to zero except Z2 for ða; kÞ close to multiple-species communities. J. Math. Biol. 26, 651–660.
Hammond, J.I., Luttbeg, B., Sih, A., 2007. Predator and prey space use: dragonflies
ð0; k0 Þ, this implies that the operator La;k is invertible. Hence, and tadpoles in an interactive game. Ecology 88, 1525–1535.
! ! Heithaus, M.R., 2001. Habitat selection by predators and prey in communities with
j2 0 asymmetrical intraguild predation. Oikos 92, 542–554.
La;k ¼ (A.35) Houston, A.I., 2008. Matching and ideal free distributions, Oikos 117 (7), 978–983.
jv bV 2 j1
Hugie, D.M., Dill, L.M., 1994. Fish and game—a game-theoretic approach to habitat
selection by predators and prey. J. Fish Biol. 45, 151–169.
is uniquely solvable for ða; kÞ close to ð0; k0 Þ, i.e., lða; kÞ is an Hugie, D.M., Grand, T.C., 1998. Movement between patches, unequal competitors
eigenvalue for (A.30) with ðj1 ; j2 ; jv Þ as its eigenfunction. This and the ideal free distribution. Evol. Ecol. 12, 1–19.
Hugie, D.M., Grand, T.C., 2003. Movement between habitats by unequal
establishes the instability of ð0; P 2 ; V 2 Þ for small a and k close
competitors: effects of finite population size on ideal free distributions. Evol.
to k0 . & Ecol. Res. 5, 131–153.
Iwasa, Y., 1982. Vertical migration of zooplankton—a game between predator and
prey. Am. Nat. 120, 171–180.
References
Jackson, A.L., Humphries, S., Ruxton, G.D., 2004a. Resolving the departures of
observed results from the ideal free distribution with simple random
Abrahams, M.V., 1986. Patch choice under perceptual constraints—a cause for movements. J. Anim. Ecol. 73, 612–622.
departures from an ideal free distribution. Behav. Ecol. Sociobiol. 19, 409–415. Jackson, A.L., Ranta, E., Lundberg, P., Kaitala, V., Ruxton, G.D., 2004b. Consumer–
Abrams, P.A., 2007. Habitat choice in predator–prey systems: spatial instability due resource matching in a food chain when both predators and prey are free to
to interacting adaptive movements. Am. Nat. 169, 581–594. move. Oikos 106, 445–450.
Abrams, P.A., Cressman, R., Křivan, V., 2007. The role of behavioral dynamics in Kacelnik, A., Krebs, J.R., Bernstein, C., 1992. The ideal free distribution and
determining the patch distributions of interacting species. Am. Nat. 169, predator–prey populations. Trends Ecol. Evol. 7, 50–55.
505–518. Kareiva, P., Odell, G., 1987. Swarms of predators exhibit ‘‘preytaxis’’ if individual
Alonzo, S.H., 2002. State-dependent habitat selection games between predators predators use area-restricted search. Am. Nat. 130, 233–270.
and prey: the importance of behavioural interactions and expected lifetime Kennedy, M., Gray, R.D., 1993. Can ecological theory predict the distribution of
reproductive success. Evol. Ecol. Res. 4, 759–778. foraging animals? a critical analysis of experiments on the ideal free
Belgacem, F., Cosner, C., 1995. The effects of dispersal along environmental distribution. Oikos 68, 158–166.
gradients on the dynamics of populations in heterogeneous environments. Kimbrell, T., Holt, R.D., 2004. On the interplay of predator switching and prey
Can. Appl. Math. Q. 3, 379–397. evasion in determining the stability of predator–prey dynamics. Israel J. Zool.
Bernstein, C., Kacelnik, A., Krebs, J.R., 1988. Individual decisions and the 50, 187–205.
distribution of predators in a patchy environment. J. Anim. Ecol. 57, 1007–1026. Kimbrell, T., Holt, R.D., 2005. Individual behaviour, space and predator evolution
Bernstein, C., Kacelnik, A., Krebs, J.R., 1991. Individual decisions and the promote persistence in a two-patch system with predator switching. Evol. Ecol.
distribution of predators in a patchy environment. 2. The influence of travel Res. 7, 53–71.
costs and structure of the environment. J. Anim. Ecol. 60, 205–225. Kondoh, M., 2006. Does foraging adaptation create the positive complexity–st-
Bolker, B., Holyoak, M., Křivan, V., Rowe, L., Schmitz, O., 2003. Connecting ability relationship in realistic food-web structure? J. Theor. Biol. 238,
theoretical and empirical studies of trait-mediated interactions. Ecology 84, 646–651.
1101–1114. Kondoh, M., 2007. Anti-predator defence and the complexity–stability relationship
Bouskila, A., 2001. A habitat selection game of interactions between rodents and of food webs. Proc. R. Soc. B—Biol. Sci. 274, 1617–1624.
their predators. Ann. Zool. Fenn. 38, 55–70. Kshatriya, M., Cosner, C., 2002. A continuum formulation of the ideal free
Cantrell, R.S., Cosner, C., 1999. A comparison of foraging strategies in a patchy distribution and its implications for population dynamics. Theor. Popul. Biol.
environment. Math. Biosci. 160, 25–46. 61, 277–284.
Cantrell, R.S., Cosner, C., 2003. Spatial Ecology via Reaction-Diffusion Equations. Křivan, V., 1997. Dynamic ideal free distribution: effects of optimal patch choice on
Series in Mathematical and Computational Biology. Wiley, Chichester, UK. predator–prey dynamics. Am. Nat. 149, 164–178.
Cantrell, R.S., Cosner, C., Lou, Y., 2006. Movement toward better environments and Lima, S.L., 2002. Putting predators back into behavioral predator–prey interactions.
the evolution of rapid diffusion. Math. Biosci. 204, 199–214. Trends Ecol. Evol. 17, 70–75.
Cantrell, R.S., Cosner, C., Lou, Y., 2007. Advection mediated coexistence of Lou, Y., Ni, W.M., 1996. Diffusion, self-diffusion and cross-diffusion. J. Differential
competing species. Proc. R. Soc. Edinburgh 137A, 497–518. Equations 131, 79–131.
Casten, R.G., Holland, C., 1978. Instability result for reaction–diffusion equations Matano, H., 1979. Asymptotic behavior and stability of solutions of semilinear
with Neumann boundary conditions. J. Differential Equations 27, 266–273. diffusion equations. Publ. Res. Inst. Math. Sci. 15, 401–454.
Chen, X., Lou, Y., 2008. Principal eigenvalue and eigenfunctions of an elliptic May, R.M., 1972. Will a large complex system be stable? Nature 238, 413–414.
operator with large advection and its application to a competition model. May, R.M., 1973. Stability and Complexity in Model Ecosystems. Princeton
Indiana Univ. Math. J. 57, 627–658. University Press, Princeton, NJ.
Chen, X., Hambrock, R., Lou, Y., 2008. Evolution of conditional dispersal: a Milinski, M., 1994. Ideal free theory predicts more than only matching—a critique
reaction–diffusion–advection model. J. Math. Biol. 57, 361–386. of Kennedy and Gray’s review. Oikos 71, 163–166.
Cosner, C., 2005. A dynamic model for the ideal-free distribution as a partial Murray, J.D., 2002. Mathematical Biology. 1. An Introduction, third ed. Springer,
differential equation. Theor. Popul. Biol. 67, 101–108. New York.
ARTICLE IN PRESS
200 S.M. Flaxman, Y. Lou / Journal of Theoretical Biology 256 (2009) 187–200
Murray, J.D., 2003. Mathematical Biology. 2. Spatial Models and Biomedical Schreiber, S.J., Vejdani, M., 2006. Handling time promotes the coevolution of
Applications, third ed. Springer, New York. aggregation in predator–prey systems. Proc. R. Soc. Biol. Sci. 273, 185–191.
Okubo, A., 1980. Diffusion and Ecological Problems: Mathematical Models. Schreiber, S.J., Fox, L.R., Getz, W.M., 2000. Coevolution of contrary choices in
Springer, Berlin. host–parasitoid systems. Am. Nat. 155, 637–648.
Okubo, A., Levin, S.A. (Eds.), 2001. Diffusion and Ecological Problems: Modern Schwinning, S., Rosenzweig, M.L., 1990. Periodic oscillations in an ideal-free
Perspectives. Interdisciplinary Applied Mathematics, vol. 14, second ed. predator–prey distribution. Oikos 59, 85–91.
Springer, Berlin. Sih, A., 1998. Game theory and predator–prey response races. In: Dugatkin, L.A.,
Protter, M.H., Weinberger, H.F., 1984. Maximum Principles in Differential Reeve, H.K. (Eds.), Game Theory and Animal Behavior. Oxford University Press,
Equations, second ed. Springer, Berlin. New York, pp. 221–238.
Rosenheim, J.A., 2004. Top predators constrain the habitat selection games played Sih, A., 2005. Predator–prey space use as an emergent outcome of a behavioral
by intermediate predators and their prey. Israel J. Zool. 50, 129–138. response race. In: Barbosa, P., Castellanos, I. (Eds.), Ecology of Predator–Prey
Rowell, J.T., 2007. The limitation of species range: a consequence of searching along Interactions. Oxford University Press, London, pp. 240–255.
resource gradients, preprint. Tregenza, T., 1995. Building on the ideal free distribution. Adv. Ecol. Res. 26,
Ruxton, G.D., Humphries, S., 1999. Multiple ideal free distributions of unequal 253–307.
competitors. Evol. Ecol. Res. 1, 635–640. Uchida, S., Drossel, B., 2007. Relation between complexity and stability in food
Safran, R.J., 2004. Adaptive site selection rules and variation in group size of barn webs with adaptive behavior. J. Theor. Biol. 247, 713–722.
swallows: individual decisions predict population patterns. Am. Nat. 164, van Baalen, M., Sabelis, M.W., 1993. Coevolution of patch selection strategies of
121–131. predator and prey and the consequences for ecological stability. Am. Nat. 142,
Safran, R.J., Doerr, V.A.J., Sherman, P.W., Doerr, E.D., Flaxman, S.M., Winkler, D.W., 646–670.
2007. Group breeding in vertebrates: linking individual- and population-level van Baalen, M., Sabelis, M.W., 1999. Nonequilibrium population dynamics of ‘‘ideal
approaches. Evol. Ecol. Res. 7, 1163–1185. and free’’ prey and predators. Am. Nat. 154, 69–88.
Note: This service is not intended for secure transactions such as banking, social media, email, or purchasing. Use at your own risk. We assume no liability whatsoever for broken pages.
Alternative Proxies: