Coexistence phenomena and global bifurcation structure

in a chemostat-like model with species-dependent

diffusion rates
François Castella, Sten Madec

To cite this version:

François Castella, Sten Madec. Coexistence phenomena and global bifurcation structure in a
chemostat-like model with species-dependent diffusion rates. Journal of Mathematical Biology,
Springer Verlag (Germany), 2014, 68 (1-2), pp.377-415. �10.1007/s00285-012-0633-7�. �hal-00777025�

HAL Id: hal-00777025

https://hal.archives-ouvertes.fr/hal-00777025
Submitted on 17 Jan 2013

HAL is a multi-disciplinary open access L’archive ouverte pluridisciplinaire HAL, est

archive for the deposit and dissemination of sci- destinée au dépôt et à la diffusion de documents
entific research documents, whether they are pub- scientifiques de niveau recherche, publiés ou non,
lished or not. The documents may come from émanant des établissements d’enseignement et de
teaching and research institutions in France or recherche français ou étrangers, des laboratoires
abroad, or from public or private research centers. publics ou privés.
Biological Mathematics manuscript No.
(will be inserted by the editor)

Coexistence phenomena and global bifurcation structure

in a chemostat-like model with species-dependent diffusion
rates

François Castella · Sten Madec

the date of receipt and acceptance should be inserted later

Abstract We study the competition of two species for a single resource in a chemostat.
In the simplest space-homogeneous situation, it is known that only one species survives,
namely the best competitor. In order to exhibit coexistence phenomena, where the two
competitors are able to survive, we consider a space dependent situation: we assume
that the two species and the resource follow a diffusion process in space, on top of
the competition process. Besides, and in order to consider the most general case, we
assume each population is associated with a distinct diffusion constant. This is a key
difficulty in our analysis: the specific (and classical) case where all diffusion constants
are equal, leads to a particular conservation law, which in turn allows to eliminate
the resource in the equations, a fact that considerably simplifies the analysis and the
qualitative phenomena.
Using the global bifurcation theory, we prove that the underlying 2-species, sta-
tionary, diffusive, chemostat-like model, does possess coexistence solutions, where both
species survive. On top of that, we identify the domain, in the space of the identified
bifurcation parameters, for which the system does have coexistence solutions.
Keywords Global bifurcation · Elliptic systems · Heterogeneous environment ·
Coexistence · Chemostat
Mathematics Subject Classification (2000) 35Q92 · 35K58 · 92D25 · 92D30

1 Introduction

The present paper is devoted to the study of coexistence solutions in some chemostat-
like systems, where various species compete for a single resource. The starting point

Sten Madec
Institut de Mathematiques de Bordeaux Universite Victor Segalen Bordeaux 2, 3ter place de
la victoire, 33000 Bordeaux Cedex, France
E-mail: sten.madec@univ-rennes1.fr
François Castella
Université de Rennes 1, UMR CNRS 6625 Irmar, Campus de Beaulieu, 35042 Rennes cedex,
France
E-mail: francois.castella@univ-rennes1.fr
2

of our analysis is the fact that in the simplest models, i.e. in the space-homogeneous
situation, only one species survives, namely the best competitor. Therefore, and in
order to observe situations where all species are able to survive, we readily consider
the space-inhomogeneous situation, where the various species and the single resource
follow a diffusion process in space. Technically speaking, and in order to tackle the most
general situation, we assume that each population possesses its own distinct diffusion
coefficient. This is a major difficulty and originality in the present text, as we discuss
later in this introduction.
The main result of this paper is that the underlying 2-species chemostat-like model,
does possess coexistence solutions, i.e. solutions where all species survive. Besides,
we are able to identify a domain in the space of the relevant parameters, for which
coexistence holds.
Our construction relies on global bifurcations in elliptic systems. Although we con-
jecture that our analysis may be generalized to the case of N competing species for
any N ≥ 2, our results can only be proved in the case N = 2 for the time being.
Let us come to technical statements.
We study the nonnegative steady-state solutions of the reaction-diffusion system

 ∂t R = a0 ∆R − F1 (x, R)U − F2 (x, R)V − m0 (x)R + I,
∂t U = a1 ∆U + (F1 (x, R) − m1 (x)) U, (x ∈ Ω, t > 0),

∂t V = a2 ∆V + (F2 (x, R) − m2 (x)) V,
where Ω is a bounded region in Rn with smooth boundary. The above system is
supplemented with Neumann1 boundary conditions
∂n R(t, x) = ∂n U (t, x) = ∂n V (t, x) = 0 (x ∈ ∂Ω, t > 0),
where ∂n is the normal derivative on the boundary ∂Ω.
The above system describes a situation where two species with density U = U (t, x)
and V = V (t, x) respectively, compete for the same resource with density R = R(t, x),
through the nonlinear terms Fi (x, R)U and Fi (x, R)V (i = 1, 2). Besides, the space
dependent resource R, as well as the two species U , V , follow a diffusion process in
space, with the distinct diffusion constants a0 > 0, a1 > 0, a2 > 0 respectively2 . The
space dependent functions mi (x) > 0 on Ω (i = 0, 1, 2), are death rates, while the
space dependent functions Fi (x, R) = Fi (x, R(t, x)) ≥ 0 are the consumption rates.
The given, time-independent function I = I(x) ≥ 0 is the nutrient input. All these
data are assumed smooth.
In order to implement a bifurcation method, we normalize the consumption rates
as follows. We readily choose given, smooth, functions f1 = f1 (x, R), f2 = f2 (x, R),
and introduce two bifurcation parameters c1 > 0 and c2 > 0, which somehow measure
the strength of the interaction between the species and the resource, through
F1 (x, R) ≡ c1 f1 (x, R), F2 (x, R) ≡ c2 f2 (x, R). (1.1)
1 Robin boundary conditions, of the form a ∂ R + b (x)R = g(x), a ∂ U + b (x)U =
0 n 0 1 n 1
a2 ∂n V + b1 (x)V = 0 on ∂Ω, with g(x) ≥ 0 and bi (x) ≥ 0 (i = 0, 1, 2), would do as well, as we
discuss later in this text.
2 Our analysis is valid when the various constant coefficients diffusion operators a ∆ become
i
div ai (x)∇ for some smooth, space-dependent coefficients ai (x) > 0 on Ω, provided all coef-
ficients ai (x) are proportional, i.e. ai (x) = λi a0 (x) (i = 1, 2) for some constants λ1 > 0 and
λ2 > 0. This easy extension is discussed later in the text. Needless to say, in that case, Robin
boundary conditions become a0 (x)∂n R + b0 (x)R = g(x) on ∂Ω, and so on, with g(x) ≥ 0 and
bi (x) ≥ 0 on ∂Ω (i = 0, 1, 2).
 3

Note that, since we are only interested in nonnegative solutions (R, U, V ), the only
important data is the value of fi (x, R) for R ≥ 0: as shown by our analysis, any
smooth extension of fi (x, R) may be retained for values R ≤ 0, provided fi (x, R) ≤ 0
whenever R ≤ 0.
With the above notations, in this paper we look for stationary solutions U = U (x),
V = V (x), R = R(x) to the above system, namely3

 (m0 (x) − a0 ∆)R + c1 f1 (x, R) U + c2 f2 (x, R) V = I(x),

 (m1 (x) − a1 ∆)U − c1 f1 (x, R) U = 0, (x ∈ Ω)
(m2 (x) − a2 ∆)V − c2 f2 (x, R) V = 0, (1.2)



∂n R = ∂n U = ∂n V = 0 (x ∈ ∂Ω).

More precisely, our goal is to exhibit coexistence solutions in (1.2), i.e. solutions R, U ,
V for which R > 0, U > 0, V > 0. Our approach relies on a global bifurcation method,
where c1 and c2 are used as bifurcation parameters. In that respect, we also aim at
identifying a domain in the (c1 , c2 )-plane for which coexistence holds.
Let us come to some bibliographical comments.
Bifurcation methods have been used in many texts concerning interacting species
(competition models, predator-prey systems), see [20, 21, 23, 22] and more recently in
the study of some age structured models, see [8, 9]. In that respect, we wish to stress
that the chemostat involves a fairly specific mathematical structure, a fact that plays
a crucial role below: the nonlinear coupling in (1.2), say, only involves terms of the
form fi (x, R) U or fi (x, R) V ; in other words the two species U and V in (1.2) are
only coupled through the resource R. This observation holds in any chemostat model
and allows, in some situations, to reduce the original model to a standard competition
system by eliminating the equation on the resource, see [13, 12, 11, 3, 18, 19, 10].
Steady states of unstirred chemostats have been first studied by Waltman et al.
in [11]. The authors consider two species evolving in the one-dimensional situation
Ω = [0, 1]. A generalisation in the case of two species evolving in a higher dimensional
domain Ω is studied by Wu [18] and Wu and Nie [19]. Using the index in a positive
cone (see [24]), Zheng et al. [15, 14] show coexistence results in systems with various
trophic levels. In all these texts, the heterogeneity in space, that is crucial to recover
coexistence phenomena, is introduced by imposing a gradient of the resource, which in
turn is obtained through the boundary condition, of Robin type. All other coefficients
are space independent. In the present text at variance, we allow the reaction terms
(and other less crucial coefficients) to actually depend on space.
A key point is the following. In all the above works, the authors assume that the
competing species, and the resource, have the same diffusion rate and the same death
rate. This assumption provides a specific conservation law, that links the resource and
the competing species. In our case it reads (taking a0 = a1 = a2 = a and m0 (x) =
m1 (x) = m2 (x) = m(x))

m(x)(R + U + V ) − a∆(R + U + V ) = I(x). (1.3)

Relation (1.3) allows to eliminate the resource R from the equations, and to write a
reduced system whose semi-trivial solutions satisfy a simple, scalar, elliptic equation.
3 Recall that Robin boundary conditions are covered by our analysis, as well as variable

coefficients diffusion operators div ai (x)∇, provided ai (x) = λi a0 (x) (i = 1, 2), see footnotes 1
and 2.
4

Semi-trivial solutions are those corresponding to either (U > 0, V = 0) or to (U =

0, V > 0). They correspond to the case where one and only one species survives. Once
the semi-trivial solutions are constructed, global bifurcation techniques can be applied
to obtain true coexistence solutions, i.e. solutions of the form (U > 0, V > 0), from
the semi-trivial ones,.
When the conservation law (1.3), is not aivalable, very few is known. Some pertur-
bation results are available. In [13], the authors use a perturbation method to extend
the above mentioned result when the equation (1.3) is nearly verified. Baxley and
Robinson [16] study a very general system in the case of N competing species, and
they establish a result close to the bifurcation point.
In this paper, we propose a global method using the more general conservation
equation

(m0 (x) − a0 ∆)R + (m1 (x) − a1 ∆)U + (m2 (x) − a2 ∆)V = I. (1.4)

Eliminating the unknow R in (1.4) leads to nonlocal semi-trivial problems. We are able
to study these semi-trivial problems by using a lower-upper solutions technique in the
so-obtained scalar, nonlocal, elliptic equations. In an independent step, a specific use of
global bifurcation techniques then allows to construct true coexistence solutions (U >
0, V > 0), starting from the semi-trivial solutions (U > 0, V = 0) or (U = 0, V > 0).
This is a key step of our approach. We wish to stress that the lower-upper solutions
part of our analysis requires (see Assumption 2 below) the crucial hypothesis4

mi (x) m (x)
∀x ∈ Ω, ≤ 0 (i = 1, 2). (1.5)
ai a0
It means that the ratio between death rate and diffusion rate should be larger for
the resource than for the competing species, or, in other words, that the two species
should diffuse relatively faster than the resource. Since spatial heterogeneity, and the
associated diffusion processes, are the key to obtaining systems which allow coexistence,
this assumption is quite natural: diffusion of the competing species helps obtaining
coexistence situations. To be complete, let us mention that in the case when Robin
boundary conditions are retained, another crucial assumption appears, namely5

bi (x) b (x)
∀x ∈ ∂Ω, ≤ 0 (i = 1, 2). (1.6)
ai a0
Assumption (1.6) is similar to (1.5) in spirit, in that a stronger ratio between the
escape rate and the diffusion rate is required for the resource R at the boundary, in
comparison with the analogous ratio for populations U and V .
The organization of the paper is as follows. In section 2 we present the notations and
recall some technical results used in the paper. We also state our main results, namely
Theorems 2.14 and 2.16. In section 3, we construct the above mentioned semi-trivial
solutions. Under assumption 2, the lower-upper solutions method, in conjunction with
bifurcation arguments, allows to prove existence, uniqueness, and non-degeneracy of
the semi-trivial solutions. Section 4 is the main step of our study, in that we prove the
4 In the case when the diffusion operators a ∆ become div a (x)∇ with a (x) = λ a (x)
i i i i 0
(i = 1, 2), the condition below becomes mi (x)/ai (x) ≤ m0 (x)/a0 (x) for x ∈ Ω (i = 1, 2).
5 This assumption obviously becomes b (x)/a (x) ≤ b (x)/a (x) for x ∈ ∂Ω (i = 1, 2), when
i i 0 0
the ai ’s depend on x.
 5

existence of solutions (R, U, V ) to (1.2) that satisfy R > 0, U > 0, V > 0. A global
bifurcation theorem is used to construct these coexistence solutions, by joining the two
families of semi-trivial solutions. Our construction leads to define a domain Θ ⊂ R2+ in
the space of bifurcation parameters (c1 , c2 ), called the coexistence domain. This domain
is such that whenever (c1 , c2 ) ∈ Θ, a coexistence solution is at hand. In section 5, we
state some consequences of our analysis, which provide an ecological point of view.
Section 6 concludes this paper.

2 Preliminaries and statement of our results

2.1 Generalities

For i = 0, 1, 2, the constants ai are supposed positive, and the fonctions mi (x) and
I(x) are assumed smooth, with mi (x) > 0 on Ω and I(x) ≥ 0 and I(x) 6≡ 0 on Ω.
Taking a given α ∈ (0, 1) whose value is irrelevant, we define the spaces6

X = {u ∈ C 2+α (Ω), ∂n u = 0 on ∂Ω}

∗
X+ = {u ∈ X, ∀x ∈ Ω, u(x) ≥ 0}, X+ = {u ∈ X+ , ∀x ∈ Ω, u(x) > 0}. (2.1)
3
In the sequel, a solution to (1.2) is a triple (R, U, V ) ∈ X+ that satisfies (1.2). A
∗ ∗ ∗
coexistence solution is a solution that lies in X+ × X+ × X+ . For i = 0, 1, 2, we note

Ai := mi (x) − ai ∆. (2.2)

It is well known that, for all α ∈ (0, 1), we have

Ai : {w ∈ C 2+α (Ω), ∂n w = 0 on ∂Ω} −→ C α (Ω) is one-to-one.

In order to keep simple notations, the above operator will always be denoted by the
same symbol Ai for any choice of α. In the similar spirit we note

Ki := A−1
i . (2.3)

For each i = 0, 1, 2, the operator Ki is compact when seen as (more precisely : when
extended to) an operator from C 1 (Ω) to C 1 (Ω) and from L2 (Ω) to L2 (Ω). Note that
each operator Ki maps X to X compactly as well. Recall that the strong maximum
principle for elliptic operators with Neumann (or Robin) boundary conditions reads,
whenever u ∈ X,

 Ai u ≥ 0
∂n u ≥ 0 =⇒ min u(x) = m > 0. (2.4)
 x∈Ω
u 6≡ 0
The strong maximum principle also implies the following uniqueness

Ai u = 0
=⇒ u ≡ 0. (2.5)
∂n u = 0
We last recall the following standard Lemmas
6 with the obvious adaptation if Robin boundary conditions and/or variable coefficients a ’s
i
are retained: to each operator div ai (x) ∇−mi (x) with boundary condition ai (x)∂n ·+bi (x)· = 0
is associated the space Xi = {u ∈ C 2+α (Ω), ai (x)∂n u + bi (x)u = 0 on ∂Ω}, and the triple
(R, U, V ) then is to be exhibited in X0,+ × X1,+ × X2,+ .
6

Lemma 2.1 Take m(x) ∈ C α (Ω) and q(x) ∈ C α (Ω). Assume m(x) > 0 for all x ∈ Ω.
Take a ∈ R∗+ . Then the eigenvalue problem

(m(x) − a∆)φ + q(x)φ = λφ on Ω, ∂n φ = 0 on ∂Ω

has an infinite sequence of eigenvalues

λ1 (q) < λ2 (q) ≤ . . .

R R
a(∇φ)2 + (m + q)φ2
Moreover, λ1 (q) = min R is a simple eigenvalue and
φ∈H 1 (Ω), φ6=0 φ2
the corresponding eigenfunction does not change sign on Ω. The quantity λ1 (q) is the
only eigenvalue whose associated eigenfunction does not change sign on Ω. Finally
λ1 (q) depends continuously on q and, if q1 ≤ q2 with q1 6= q2 , then λ1 (q1 ) < λ1 (q2 ).

Lemma 2.2 Take q(x) ∈ C α (Ω), a ∈ R∗+ such that q(x) > 0 for any x ∈ Ω. Then the
eigenvalue problem
(m(x) − a∆)φ = µq(x)φ, ∂n φ = 0,

has an infinite sequence of eigenvalues

0 < µ1 (q) < µ2 (q) ≤ . . .

R R
a(∇φ)2 + mφ2
Moreover, µ1 (q) = min R is a simple eigenvalue and the
φ∈H 1 (Ω), φ6=0 qφ2
corresponding eigenfunction does not change sign on Ω. The quantity µ1 (q) is the only
eigenvalue whose associated eigenfunction does not change sign on Ω. Moreover, µ1 (q)
depends continuously on q and, if q1 ≤ q2 with q1 6= q2 , then µ1 (q1 ) < µ1 (q2 ).

2.2 Lower- and upper-solutions

In order to make use of a lower-upper solution technique later in this text, we readily
introduce the following assumption

Assumption 1 For i = 1, 2, we assume fi (x, R) ∈ C 1 (Ω × R), with f1 (x, R) ≤ 0

whenever R ≤ 07 . Besides, we assume that for any x ∈ Ω, we have

∂fi
∀R > 0, fi (x, R) > 0, and (x, R) > 0.
∂R

In other words, the consumption rate is supposed to be non-negative and increasing

function of the resource. We also introduce the following crucial one-sided condition8

Assumption 2 For i = 1, 2 and x ∈ Ω, we have

mi (x)/ai ≤ m0 (x)/a0 .
7 Recall that we are only interested in situations with R ≥ 0, hence the way we extend f
i
for negative values of R is irrelevant.
8 See footnote 4 in the case of variable coefficients diffusion operators.
 7

As we show now, this condition provides a monotonicity property that plays a key
rôle in our analysis. Whenever w ∈ X+ , define Ri (w) ∈ X as the unique solution in X
to

A0 Ri (w) + Ai w = I. (2.6)

The operator w 7→ Ri (w) is introduced for the following reason. The one-species prob-
lem (corresponding to semi-trivial solutions (U > 0, V = 0) say), reads

A0 R + c1 f1 (x, R)U = I, A1 U − c1 f1 (x, R)U = 0. (2.7)

This in turn is equivalent to

R = R1 (U ), A1 U − c1 f1 (x, R1 (U )) U = 0, (2.8)

and R1 (U ) may be seen as the resource at hand in the presence of the population U .
In any circumstance, the one-species problem leads to considering the above nonlinear
and nonlocal elliptic problem, with nonlinearity w 7→ f1 (R1 (w)) w.
Now, an easy computation provides the alternative formula9 .
ai 1

Ri (w) = K0 (I) − K0 A 0 w + K0 (ai m0 (x) − a0 mi (x))w . (2.9)
a0 a0
A key point is the fact that the nonlocal term K0 (ai m0 − a0 mi )w above satisfies

K0 (ai m0 − a0 mi )w ≥ 0 whenever w ≥ 0, (2.10)

as an obvious consequence of Assumption 2 together with the maximum principle.

Another remark is in order. In the case of Neumann boundary conditions, we have the
obvious relation K0 A0 w = w. The reader’s attention is drawn to the fact that in the
case of Robin boundary condition, we have K0 A0 w 6= w in general. Note however that
the following holds. Provided we assume bi /ai ≤ b0 /a0 (i = 1, 2) – see equation (1.6)
and footnote 5 – we have

K0 A0 w ≤ w whenever w ≥ 0. (2.11)

This comes from the maximum principle together with the fact that, when w ≥ 0, the
function v = K0 A0 w satisfies A0 (v − w) = 0 with the boundary condition (a0 ∂n +
b0 )(v − w) = +(a0 b1 − b0 a1 )w/a1 ≤ 0.
We readily show that Assumption 2 implies the following one-sided Lipschitz con-
dition for the nonlinearity w 7→ f1 (R1 (w)) w in (2.8).

Lemma 2.3 Suppose Assumption 2 is true. Let M be a positive constant and take
i = 1, 2. Then, there exists γ = γi (M ) > 0 such that

w1 (x) fi (x, Ri (w1 ))(x) − w2 (x) fi (x, Ri (w2 ))(x) ≥ −γ(w1 (x) − w2 (x))

whenever w1 , w2 ∈ X satisfy 0 ≤ w2 ≤ w1 ≤ M .

Remark 2.4 The point is, the above estimate is pointwise in x, though it involves the
nonlocal operator Ri . ⊓
⊔
9 When the diffusion operators become div a (x)∇ with a (x) = λ a (x), see footnotes 2
i i i 0

and 4, the formula below becomes Ri (w) = K0 (I) − λi K0 A0 w + K0 (λi m0 (x) − m1 (x))w ,
with λi m0 (x) − m1 (x) ≥ 0 for all x, and our analysis is unchanged.
8

Remark 2.5 If all diffusion operators are the same, as in the previously quoted papers,
namely if Ai ≡ A0 (i = 1, 2), then the nonlocal terms of the form K0 (ai m0 − a0 mi )w
vanish in the course of the analysis. In that particular case, the method we develop
coincides with that of [18]. The nonlocal terms constitute the main difficulty we treat.
⊓
⊔

Admitting Lemma 2.3 is proved for the moment, we readily state that this result
allows us to apply a lower-upper solution method in the nonlocal elliptic system

Ai w − ci fi (x, Ri (w)(x)) w = 0, (2.12)

where w ∈ X is the unknown. Indeed, using Lemma 2.3, the following definition and
Theorem are standard (see [5]).

Definition 2.6 (lower- and

upper-solutions) An upper-solution to equation (2.12)
is a function w ∈ C 2+α Ω verifying10

Ai w(x) − ci fi (x, Ri (w)(x)) w(x) ≥ 0 for all x ∈ Ω, and ∂n W ≥ 0 on ∂Ω.

A lower-solution is defined in the similar way with reversed inequalities.

Theorem 2.7 (lower-upper solutions method – See [5])

Assume there exists a lower resp. upper solution W 1 resp. W 2 to equation (2.12),
which satisfies 0 ≤ W 1 ≤ W 2 .
Then, equation (2.12) admits a pair (W − , W + ) of solutions, with W 1 ≤ W − ≤
W + ≤ W 2.
If W 1 and W 2 are not solutions to (2.12), we have W 1 < W − ≤ W + < W 2 on
Ω.
The pair (W − , W + ) is maximal in the sense that each solution W to (2.12) which
satisfies W ∈ [W 1 , W 2 ] necessarily verifies W ∈ [W − , W + ] as well.

Remark 2.8 Stricto sensu the above Theorem is not to be found in [5]. Smoller requires
the nonlinear term be Lipschitz in w, a property that we do not have at hand in
the present case. It is standard to observe that the key of the proof, which relies on
an iteration of the maximum principle, is the following. When writing the equation
Ai w = ci fi (x, Ri (w)) w =: Gi (x, w), the point is to find a (large) K > 0 and a
(large) M > 0 such that whenever 0 ≤ W1 (x) ≤ W2 (x) ≤ M for all x, we have
G(x, W1 )(x) + KW1 (x) ≤ G(x, W2 )(x) + KW2 (x) for all x as well. The one-sided
Lipschitz estimate of Lemma 2.3 is enough in that respect.
Note that Pao [7, 6] establishes variants of the above techniques for systems, in
the case where the nonlinear terms, which are vector-valued, satisfy so-called quasi-
monotonicity properties. ⊓ ⊔

There remains to prove Lemma 2.3.

Proof of Lemma 2.3

Firstly, when w ∈ X satisfies 0 ≤ w ≤ M , the maximum principle provides in (2.9)

M
kRi (w)kL∞ ≤ kK0 (I)kL∞ + M kK0 (m0 )kL∞ + kK0 (ai m0 − a0 mi )kL∞ =: M∞ .
a0
10 With the obvious extension in the case of Robin boundary conditions.
 9

The assumed smoothness of fi ensures that fi is globally Lipschitz on Ω ×[−M∞ , M∞ ].

We call Ci the Lipstchitz constant associated with fi .
Next, whenever 0 ≤ w2 ≤ w1 ≤ M , with wi ∈ X (i = 1, 2), we have

ai 1

Ri (w1 ) − Ri (w2 ) = − K0 A0 (w1 − w2 ) + K0 (ai m0 (x) − a0 mi (x)) (w1 − w2 )
a0 a0
a
≥ − i K0 A0 (w1 − w2 )
a0
a
≥ − i (w1 − w2 ).
a0

where the first lower bound uses Assumption 2 while the second uses the observation
(2.10). Hence, writing

w1 (x)fi (x, Ri (w1 )(x)) − w2 (x)fi (x, Ri (w2 )(x))

= fi (x, Ri (w1 )(x)) (w1 − w2 )(x) + w2 (x) fi (x, Ri (w1 )(x)) − fi (x, Ri (w2 )(x))


≥ w2 (x) fi (x, Ri (w1 )(x)) − fi (Ri (x, w2 )(x)) ,

we distinguish two cases. If x is such that Ri (w1 )(x) ≥ Ri (w2 )(x), then fi being an
increasing function of R, we recover

w1 (x)fi (x, Ri (w1 )(x)) − w2 (x)fi (x, Ri (w2 )(x)) ≥ 0.

In the opposite case we have

w1 (x)fi (x, Ri (w1 )(x)) − w2 (x)fi (x, Ri (w2 )(x))

≥ w2 (x) fi (x, Ri (w1 )(x)) − fi (Ri (x, w2 )(x))
≥ +Ci w2 (x) (Ri (w1 )(x) − Ri (w2 )(x))
a a
≥ −Ci i w2 (x) (w1 (x) − w2 (x)) ≥ −Ci i M (w1 (x) − w2 (x)).
a0 a0

The proposition is proved. ⊓

⊔

2.3 Bifurcation methods

We state the two bifurcation theorems we use in the sequel; for equations of the form,

T (c, W ) = W,

where c ∈ R is the bifurcation parameter, W ∈ Y is the seeked solution, and Y is

a Banach space, while T (c, W ) ∈ C 0 (R × Y ; Y ) is a given, continuous map. In the
following we assume that T is twice Fréchet-differentiable in (c, W ), and we denote by
Dc resp. DW the Fréchet derivatives of T with respect to c resp. W .
We start with the local bifurcation theorem of Crandall-Rabinowitz [1].

Theorem 2.9 (Local bifurcation from a simple eigenvalue – see [1])

With the above notation, we assume that

∀c ∈ R, T (c, 0) = 0.
10

We also assume that for some value c0 ∈ R, the following holds:




 dim Ker(L(c0 ) = 1, where we note L(c) = Id − DW T (c, 0),





 codim Im(L(c0 )) = 1,



 and, whenever W0 satisfies Ker L(c0 ) = span(W0 ), we have




 Dc L(c0 ) · W0 ∈/ Im(L0 ).

Then, there exists ε > 0 and a map (c(s), X(s)) ∈ C 0 ((−ε, ε); R × Y ), with c(0) = c0 ,
X(0) = 0, such that close to (c0 , 0) in R × Y , the only nontrivial solution to T (c, W ) =
W is given by

T (c, W ) = W

⇐⇒ ∃s ∈ (−ε, ε) such that (c, W ) = c(s), sW0 + s X(s) .
(c, W ) 6= (c, 0)
We complete the picture by stating a global version of the theorem. Some additional
assumptions are required. We need the following compactness assumption
T : R × Y → Y is a compact operator, and ,
∀(c, W ), T (c, W ) = DW T (c, 0) · W + R(c, W ), (2.13)
where DW T (c, 0) is a linear compact operator.
In other words we assume that the linearized part of equation T (c, W ) = W , close to
the trivial solution W = 0, is always a compact perturbation of the identity.
Now, for those values of c such that the trivial solution W = 0 is an isolated
solution to T (c, W ) = W , i.e. typically whenever DW T (c, 0) does not admit 1 as an
eigenvalue, one may define the index of the solution W = 0, as the Leray-Schauder
degree deg(Id − T (c, ·), B, 0) (here B ⊂ Y is a ball centred at 0 such that W = 0 is
the only solution to T (c, W ) = W in B). In other words, the index of the considered
solution W = 0 is
i(T (c, ·), 0) := deg(Id − T (c, ·), B, 0). (2.14)
It has the value
i(T (c, ·), 0) = deg(Id − DW T (c, 0), B, 0) = (−1)p , (2.15)
where p is the sum of the algebraic multiplicities of all (real) eigenvalues of DW T (c, 0)
that are greater than 1.
The following theorem holds true
Theorem 2.10 (Global bifurcation from a simple eigenvalue – see [20, 18])
Under the assumptions and notation of Theorem 2.9, we suppose that T is a compact
operator such that DW T (c, 0) is linear compact for any c, as in (2.13).
We also assume11 that for some ε > 0, the index i(T (c, ·), 0) is constant on (c0 −
ε, c0 ) and on (c0 , c0 + ε), and that whenever c0 − ε < α < c0 < β < c0 + ε we have
i(T (α, ·), 0) 6= i(T (β, ·), 0).
11 This second assumption is not needed when D T (c, 0) does not depend on c. In our case –
c
see below – we shall apply this Theorem for T ’s of the form T (c, W ) = A + cB(W ) where A is
a constant and B a compact operator independent of c. This is due to our choice of bifurcation
parameters: they are only involved in the two terms c1 f1 (R)U and c2 f2 (R)V in (1.2), terms
which are proportional with c1 resp. c2 . We nevertheless describe our bifurcation method in
the present more general form, in order to keep a procedure that applies as well in the case
of a nonlinear dependence on the bifurcation parameters, as would be the case when choosing
(c1 , c2 )-dependent consumption rates for instance.
 11

Then, there exists a continuum12 C of nontrivial solutions to T (c, W ) = W in

R × Y such that one of the following alternatives holds:
(i) The closure C joins the trivial solution (c0 , 0) to another trivial solution (b
c, 0), for
some bc ∈ R, b
c 6= c0 , where Id − DW T (b
c, 0) is not invertible.
(ii) The closure C joins (c0 , 0) to ∞ in R × Y .

2.4 Statement of our results

Our whole construction relies on a recursive procedure. We construct coexistence so-

lutions to (1.2) (we do not rewrite the boundary conditions),

 A0 R + c1 f1 (x, R)U + c2 f2 (x, R)V = I,
A1 U − c1 f1 (x, R)U = 0, (2.16)

A2 V − c2 f2 (x, R)V = 0,

by starting from the 0-species problem (namely trivial solutions corresponding to R >
0, U = 0, V = 0),. Then we construct 1-species, or semi-trivial, solutions (corresponding
to R > 0, and either (U > 0, V = 0) or (U = 0, V > 0)), by using lower-upper solutions
techniques. This step is complemented with the use of bifurcations from the 0-species
problem, to prove the non-degeneracy of the so-obtained semi-trivial solutions, and to
compute the index of these solutions. This step is crucial, and makes a strong use of our
Assumption 2. It is the most difficult and technical part of our analysis. Armed with
these results, we then use bifurcations again to construct true coexistence solutions
R > 0, U > 0, V > 0. This last step uses all informations gathered on the semi-trivial
solutions.

We start with the 0-species problem.

Theorem 2.11 (Trivial solution)

∗
(i) The following equation has a unique solution S ∈ X+ ,

A0 S = I. (2.17)

3
(ii) If (R, U, V ) ∈ X+ is a solution to (1.2) with U 6≡ 0 or V 6≡ 0, then13 0 < R < S.
(iii) Let w ∈ X+ . The equation

A0 R + ci fi (x, R) w = I

(i) (i) (i)

has a unique solution14 Rw ∈ X+
∗
. It satisfies 0 < Rw ≤ S. The map w 7→ Rw is
decreasing from X+ to X+ .

12 We call a continuum of solutions a connected familly of solutions (c, W ) ∈ R × Y .

13 Recall that throughout this text the notation R < S means S − R ∈ X+ ∗ , or, in other

words, that for any x ∈ Ω we have R(x) < S(x)  

14 Note that R(i) 6= R (w), see (2.9), unless we have A w = c f (i)
w i i i i x, Rw w.
12

We postpone the (easy) proof of this statement.

We next focus our attention on semi-trivial solutions to (1.2).
If V ≡ 0 (the case U ≡ 0 is similar), system (1.2) reduces to (we do not rewrite the
boundary conditions)

A0 R + c1 f1 (x, R)U = I,
(2.18)
A1 U − c1 f1 (x, R)U = 0.
We define the operator
     
K0 (I − c1 f1 (x, R)U ) S −K0 (f1 (x, R)U )
T1 (c1 , R, U ) = = + c1 . (2.19)
K1 (c1 f1 (x, R)U ) 0 K1 (f1 (x, R)U )

Clearly, T1 : R × X 2 → X 2 is continuous and compact, any fixed point (R, U ) ∈ X 2

of T1 (c1 , ·, ·), i.e. such that T1 (c1 , R, U ) = t (R, U ), is clearly a solution to (2.18), and
the trivial solution is T1 (c1 , S, 0) = t (c1 , S, 0).

The following theorem describes two solution branches to (2.18). It is proved in

section 4, using a global bifurcation technique with c1 used as the bifurcation param-
eter.

Theorem 2.12 (Semi-trivial solutions)

Under Assumptions 1 and 2, the following holds.

(i) There exists c01 > 0 such that:

• if c1 ≤ c01 , then (S, 0) is the only solution to (2.18) in X+
2
,
0 ∗ 2
• if c1 > c1 , the system (2.18) has a unique solution in (X+ ) , noted

∗
(Ru (c1 ), U ∗ (c1 )).

(ii) Whenever c1 > c01 , the solution (Ru ∗

(c1 ), U ∗ (c1 )) ∈ (X+
∗ 2
) is non-degenerate15
16 ∗ ∗ 17
and i(T1 (c1 , ·), (Ru (c1 ), U (c1 ))) = 1
∗ ∗
(iii) The map Ru : c1 7→ Ru (c1 ) is decreasing, and belongs to C 1 ((c01 , +∞), X+∗
).
Moreover, the following two limits hold uniformly on Ω, namely,

∗ ∗
Ru (c1 ) −→ S, and Ru (c1 ) −→ 0,
c1 →c01 c1 →+∞

(iv) The map U ∗ : c1 7→ U ∗ (c1 ) is increasing, and belongs to C 1 ((c01 , +∞), X+

∗
).
Moreover, the following two limits hold uniformly on Ω, namely,

U ∗ (c1 ) −→ 0, and U ∗ (c1 ) −→ U∞ ,

c1 →c01 c1 →+∞

∗
where U∞ ∈ X+ is the unique solution to A1 U∞ = I.

15


In other words, Ker Id − D(R,U ) T1 (c1 , Ru ∗ (c ), U ∗ (c )) = {0}.
1 1
16 Our proof not only provides that the index of this solution (R∗ (c ), U ∗ (c )) is one, but
u 1 1
also that all eigenvalues of Id − D(R,U ) T1 (c1 , Ru ∗ (c ), U ∗ (c )) are less than one whenever
1 1
c1 > c01 is close to c01 . This implies that the so-obtained solution is stable, i.e. the associated
time-dependent parabolic problem admits (Ru ∗ (c ), U ∗ (c )) as a locally stable steady state.
1 1
17 This apparently technical statement is the key to constructing true coexistence solutions

and obtaining Theorem 2.14 below.

 13

Remark 2.13 In fact, the mere existence of semi-trivial solutions may be obtained us-
ing a simple global bifurcation argument, without making use of our Assumption 2.
Assumption 2 is required at variance to obtain uniqueness of these solutions. This as-
sumption also plays a key rôle to establish non-degeneracy, and to compute the value
of the index. ⊓
⊔

Naturally, the similar results hold in the case U ≡ 0 and V > 0. This provides a
critical value c02 , and a solution branch (Rv∗ (c2 ), V ∗ (c2 )) ∈ (X+ ∗ 2
) whenever c2 > c02 ,
which satisfies the properties similar to the ones listed before. The natural semi-trivial
∗
solutions to (1.2) are (R, U, V ) = (Ru (c1 ), U ∗ (c1 ), 0) (with c1 > c01 ), and (R, U, V ) =
(Rv (c2 ), 0, V (c2 )) (with c2 > c2 ). We define the following two subsets of R2+ × X+
∗ ∗ 0 3
,
namely
n o
∗
Cu = (c1 , c2 , Ru (c1 ), U ∗ (c1 ), 0) ; c1 > c01 ,
n o
Cv = (c1 , c2 , Rv∗ (c2 ), 0, V ∗ (c2 )) ; c2 > c02 . (2.20)

With this notation at hand, the following Theorem is the main result of the present
paper. It establishes that coexistence solutions to (1.2) may be defined using bifurca-
tions from the two sets Cu and Cv . The proof is provided in section 5.2. Figure 2.1
illustrates the situation.

Theorem 2.14 (Coexistence solutions)

Under Assumptions 1 and 2, the following holds.
(i) (Bifurcations from Cu to Cv ). Let c1 > c01 be fixed.
There exist c∗2 = c∗2 (c1 ) > c02 and c∗∗ ∗∗ 0
2 = c2 (c1 ) > c2 , and there is a continuum

∗ 3
of positive solutions to (1.2), noted (c1 , c2 , R, U, V ) ∈ (c01 , +∞) × (c02 , +∞) × X+ ,
∗ ∗ ∗
whose closure joins the semi-trivial (c1 , c2 , Ru (c1 ), U (c1 ), 0) ∈ Cu to the semi-trivial
(c1 , c∗∗ ∗ ∗∗ ∗ ∗∗
2 , Rv (c2 ), 0, V (c2 )) ∈ Cv .
In particular, noting c2 (c1 ) = min({c∗2 , c∗∗ ∗ ∗∗
2 }) ≤ max({c2 , c2 }) = c2 (c1 ), we have

∗
3
∀c2 ∈ c2 (c1 ), c2 (c1 ) , ∃(R, U, V ) ∈ X+ coexistence solution to (1.2).

(ii) (Bifurcations from Cv to Cu ). Let c2 > c02 be fixed.

There exist c∗1 = c∗1 (c2 ) > c01 and c∗∗ ∗∗ 0
1 = c1 (c2 ) > c1 , and there is a continuum

∗ 3
of positive solutions to (1.2), noted (c1 , c2 , R, U, V ) ∈ (c1 , +∞) × (c02 , +∞) × X+
0
,
whose closure joins the semi-trivial (c∗1 , c2 , Rv∗ (c2 ), 0, V ∗ (c2 )) ∈ Cv to the semi-trivial
(c∗∗ ∗ ∗∗ ∗ ∗∗
1 , c2 , Ru (c1 ), U (c1 ), 0) ∈ Cu .
In particular, noting c1 (c2 ) = min({c∗1 , c∗∗ ∗ ∗∗
1 }) ≤ max({c1 , c1 }) = c1 (c2 ), we have

∗
3
∀c1 ∈ c1 (c2 ), c1 (c2 ) , ∃(R, U, V ) ∈ X+ coexistence solution to (1.2).

Remark 2.15 Note that the situation where
c∗2 (c1 ) = c∗∗ 2 (c1 ), say, may very well hap-
pen. In that case the interval c1 (c2 ), c1 (c2 ) is void. Hence, as we can see, the second
statement in part (i) of the Theorem is a weak byproduct of the first one, which ex-
hibits at variance an actual branch of coexistence solutions. We refer to the conjecture
stated in paragraph 6 below for a discussion of this point. ⊓ ⊔
14

V ∗ (c∗∗
2 )

∗ c∗∗ c2
c02 c2 2

U ∗ (c1 )

Fig. 2.1 Coexistence solution in the space R × X+ × X+ .

The parameter c1 is fixed here, with c1 > c01 .
Dashed lines in the (U, c2 )-plane represent (the projection of) few semi-trivial solutions
(R, U, 0) in this plane: due to their very definition, these solutions do not depend on c2 .
The particular semi-trivial solution associated with U ∗ (c1 ) – see Theorem 2.12 – is repre-
sented by a full line. The full curve in the (c2 , V )-plane represents the (projection of the)
family of semi-trivial solutions (R, 0, V ∗ (c2 )). Finally, the bold curve joining the two planes
(c2 , U ) and (c2 , V ) represents the (projection of the) coexistence solutions (c1 , c2 , R, U, V ) ∈
(c01 , ∞) × c2 (c1 ), c2 (c1 ) × X+ ∗ 3 obtained in part (i) of the Theorem. In the present figure

we have assumed c2 (c1 ) < c2 (c1 ).

∗ ∗∗

With the use of the above Theorem, one may define a coexistence domain Θ, as

Θ = (c1 , c2 ) ∈ (c01 , +∞) × (c02 , +∞), s.t.



c1 ∈ c1 (c2 ), c1 (c2 ) and c2 ∈ c2 (c1 ), c2 (c1 ) . (2.21)

It corresponds to values of the parameters (c1 , c2 ) for which a coexistence solution may
be exhibited (a subset of the set of all values (c1 , c2 ) such that a coexistence solution
may be exhibited – see paragraph 6 on that point).
The following Theorem is proved in section 5.3. It explores the structure of Θ.

Theorem 2.16 (Coexistence domain)

Under Assumption 1 and 2, and with the notation of Theorem 2.14, the following holds.
(i) Whenever c1 > c01 , the quantity c∗∗
2 (c1 ) is characterised by the relation

c∗1 (c∗∗
2 (c1 )) = c1 ,

and similarly when indices 1 and 2 are reversed.

(ii) The two maps

c∗1 (c2 ) : (c02 , +∞) −→ (c01 , +∞), and c∗2 (c1 ) : (c01 , +∞) −→ (c02 , +∞)
 15

V
V ∗ (c2 )

c2

1)
2 (c
c∗ Θ−
c2
)
∗ (c 2
c1

U ∗ (α)
Θ+ U ∗ (β)

U ∗ (γ)
c02
U
c01 α β γ c1

(a) Coexistence domain (b) Bifurcation curves

Fig. 2.2 Coexistence domain and bifurcation solutions.

Figure (a) shows a possible coexistence domain Θ. The full curve represents (c1 , c∗2 (c1 )) and the
dashed one represents (c∗1 (c2 ), c2 ). For any t > c01 , the line c1 = t intersects these two curves
at (t, c∗2 (t)) resp. (t, c∗∗
2 (t)), as implied by the very definition of the two quantities c2 (c1 ) and
∗

c∗∗
2 (c 1 ).
Figure (b) represents some bifurcating solutions corresponding to three values α, β and γ of
the parameter c1 > c01 . The retained values are here assumed to satisfy c∗2 (α) < c∗∗ 2 (α), resp.
c∗2 (β) = c∗∗ 0
2 (β), resp. c2 (γ) > c2 (γ). For each c1 > c1 , there is a coexistence solution joining
∗ ∗∗

(R, U ∗ (c1 ), 0) and (R, 0, V ∗ (c1 )).

are continuous and increasing. Moreover, for {i, j} = {1, 2}, we have

lim c∗j (ci ) = c0j , and lim c∗j (ci ) = +∞.

ci →c0i ci →+∞

(iii) With the notation (2.21), whenever (c1 , c2 ) ∈ Θ, system (1.2) has a coexistence


∗ 3
solution (R, U, V ) ∈ X+ , and we have

Θ = Θ− ∪ Θ+ , with Θ− = {(c1 , c2 ), c1 < c∗1 (c2 ) and c2 < c∗2 (c1 )},
and Θ+ = {(c1 , c2 ), c1 > c∗1 (c2 ) and c2 > c∗2 (c1 )}.

The next sections are devoted to the proof of Theorem 2.11 (trivial solutions),
Theorem 2.12 (semi-trivial solutions), as well as Theorems 2.14 and 2.16 (coexistence
solutions and coexistence domain).

3 Zero species: trivial solutions – Proof of Theorem 2.11

We prove here the various statements of Theorem 2.11. Recall that the problem with
zero species reads, shortly, A0 R = I.
Point (i). Existence and uniqueness of S is clear.
3
Point (ii). Let (R, U, V ) ∈ X+ be a solution to (1.2) with U ≥ 0 and V ≥ 0. We have
A0 R = I − c1 f1 (x, R)U − c2 f2 (x, R)V ≤ I. Hence A0 R ≤ I with A0 R 6≡ I whenever
16

U 6≡ 0 or V 6≡ 0. The strong maximum principle provides 0 < R < S, with R < S

whenever U 6≡ 0 or V 6≡ 0.
∗
Point (iii). Take w ∈ X+ . Due to Assumption 1, for ε > 0 small enough, S resp. ε are
upper resp. lower solutions to

A0 R + ci fi (x, R) w = I. (3.1)

As a consequence, there exists a pair (R− , R+ ) ∈ X 2 of maximal solutions to (3.1),

with 0 < R− ≤ R+ < S. Let us show that R− ≡ R+ . We have A0 (R+ − R− ) +
ci (fi (x, R+ )−fi (x, R− )) w = 0. Integrating over Ω and taking the boundary conditions
into account18 , we obtain
Z h i
m0 (R+ − R− ) + ci (fi (x, R+ ) − fi (x, R− )) w dx = 0.
Ω

Since R 7→ fi (x, R) is an increasing function of R for any value of x, we recover

(i)
R− = R+ . Existence and uniqueness of Rw in the Theorem follows.  
(i) (i)
Lastly, take 0 < w1 < w2 , with w1 , w2 ∈ X. We have A0 Rw2 + ci fi x, Rw2 w1 ≤
(i)
I. Hence, Rw2 ∈ X is a lower-solution to A0 R + ci fi (x, R)w1 = I. As a consequence,
(i)
ew1 ∈ X to A0 R + ci fi (x, R)w1 = I, which satisfies
there exists an actual solution R
(i) ew(i) e(i) (i)
R w2 < R 1 < S. Uniqueness then provides Rw1 = Rw1 . We recover the necessary
(i) (i)
relation Rw2 < Rw1 . This ends the proof.

4 One species: semi-trivial solutions – Proof of Theorem 2.12

In this section, we study the one species problem (2.18), corresponding to the semi-
∗ ∗
trivial solution (R, U, 0) ∈ X+ × X+ × X+ to (1.2). Recall that the one species problem
reads

A0 R + c1 f1 (x, R)U = I,
A1 U − c1 f1 (x, R)U = 0.

4.1 General facts about the one species problem

Lemma 4.1 Let c1 > 0 be fixed. There exists M0 > 0 such that each solution (R, U ) ∈
∗ 2
(X+ ) to (2.18) verifies
0 ≤ U ≤ M0 .
∗ 2
Proof of Lemma 4.1. Let (R, U ) ∈ (X+ ) be a solution to (2.18). Summing the equa-
tions on R and U provides, as already noted, A0 R + A1 U = I. As a consequence, for
some α > 0 small enough we have

(α − ∆) (a0 R + a1 U ) ≤ I.
Z
18 b0 (x) R+ − R− ≥ 0, and the conclu-
 
Robin boundary conditions would add a term
∂Ω
sion would remain unchanged.
 17

1
The strong maximum principle19 then provides 0 ≤ a0 R + a1 U ≤ kIkL∞ . In the case
α
of variable coefficients ai (x) with ai (x) = λi a0 (x), see footnote 2, the argument is the
same, due to the bound (α − div a0 (x)∇) (R + λ1 U ) ≤ I = (m0 (x) − div a0 (x)∇) R
+ (m1 (x) − λ1 div a0 (x)∇) U. ⊓ ⊔

Lemma 4.2 The eigenvalue problem A1 φ − µf1 (x, S)φ = 0 with φ ∈ X has a prin-
cipal eigenvalue c01 > 0 and a corresponding eigenfunction φ0 ∈ X+
∗
, unique up to a
multiplicative constant. We have

A1 φ0 − c01 f1 (x, S)φ0 = 0, (4.1)

Z  Z 

with c01 given by c01 = min a1 ∇φ2 + m1 φ2 f1 (x, S)φ2 .
φ∈H 1 (Ω), φ6=0

Proof of Lemma 4.2. This is a direct application of Lemma 2.2. ⊓

⊔
∗ 2
Proposition 4.3 Let c1 > 0 be fixed. Suppose there exists (R, U ) ∈ (X+ ) solution to
0
(2.18). Then we necessarily have c1 > c1 .

Proof of Proposition 4.3. The function U > 0 verifies A1 U −c1 f1 (R)U = 0. Multiplying
by φ0 , defined in Lemma 4.2, and integrating over Ω leads to
Z Z Z
0= A1 U φ0 − c 1 f1 (x, R)U φ0 = U φ0 (c01 f1 (x, S) − c1 f1 (x, R)).
Ω Ω Ω

Since Proposition 2.11 ensures R < S hence f1 (x, R) < f1 (x, S), we recover the neces-
sary condition c1 > c01 . ⊓
⊔

4.2 Existence, uniqueness, and some properties of solutions to the one species problem

The main result of this paragraph is the

Proposition 4.4 Suppose Assumptions 1 and 2 are verified. Assume c1 > c01 .
∗ 2 ∗
Then, system (2.18) has a unique solution in (X+ ) , denoted by (Ru (c1 ), U ∗ (c1 )).

∗ 2
Proof of Proposition 4.4. Take a solution (R, U ) ∈ (X+ ) to (2.18). Defining, as in
(2.9), the quantity R1 (U ) ∈ X by the relation A0 R1 (U ) + A1 U = I we recover the
necessary condition R = R1 (U ), and system (2.18) can be rewritten (with ∂n U = 0 on
∂Ω),
A1 U − c1 f1 (x, R1 (U ))U = 0, (4.2)
Let φ0 > 0 be the eigenfunction defined in Lemma 4.2, which satisfies A1 φ0 −
c01 f1 (x, S)φ0 = 0. We claim that for ε > 0 small enough and M > 0 large enough,
the pair (εφ0 , M ) is a pair of lower-upper solutions to (4.2). Indeed, on the one hand,
choosing M > 0 large enough leads to R1 (M ) < 0 (since A0 R1 (M ) = I − A1 M =
I − m1 (x)M ). Therefore, we obtain

A1 M − c1 M f1 (x, R1 (M )) ≥ m1 M ≥ 0,
19 with the obvious adaptation in the case of Robin boundary conditions.
18

with ∂n M = 0 on ∂Ω, and M is an upper-solution to (4.2). On the other hand, taking

ε > 0 small enough leads to
 
A1 (εφ0 ) − c1 f1 (x, R1 (εφ0 )) · (εφ0 ) = εφ0 c01 f1 (x, S) − c1 f1 (x, R1 (εφ0 )) ,

with A0 R1 (εφ0 ) + εc01 f1 (x, S)φ0 = I.

It is clear that lim kR1 (εφ0 ) − Sk∞ = 0. Therefore, we recover

ε→0
 
A1 (εφ0 ) − c1 f1 (x, R1 (εφ0 )) · (εφ0 ) = ε c01 − c1 φ0 f1 (x, S) + oε→0 (1) ≤ 0,

with ∂n (εφ0 ) = 0 on ∂Ω. Therefore εφ0 is a lower solution to (4.2) for ε small enough.
These considerations allow us to conclude (see Theorem 2.7) that there exists a
pair (U − , U + ) of maximal solutions to (4.2), satisfying εφ0 < U − ≤ U + < M , and
for any solution U ∈ [εφ0 , M ] to (4.2) we necessarily have U − ≤ U ≤ U + . Besides,
∗
Lemma 4.1 ensures one can choose M ≥ M0 such that any solution U ∈ X+ to (4.2)
anyhow satisfies 0 ≤ U ≤ M . Remembering that 0 is a lower-solution, we thus obtain
∗
that every solution U ∈ X+ necessarily verifies 0 ≤ U ≤ U + as well.
Let us show that U = U . We first observe that the relation 0 ≤ U ≤ U + implies
+

0 ≤ R1 (U + ) ≤ R1 (U ).

(1)
This is due to Theorem 2.11, together with the fact that R1 (U ) = RU and R1 (U + ) =
(1)
RU + in the present case (for U and U + solve the auxiliary equation A1 U = c1 f1 (. . .)U
and similarly for U + ). We deduce f1 (x, R1 (U + )) ≤ f1 (x, R1 (U )). On the other hand,
the obvious integration by parts, together with the definition of U and U + , provide
Z  h i  Z  
0= [A1 U ] U + − A1 U + U = c1 U U + f1 (x, R1 (U )) − f1 (x, R1 (U + )) .
Ω Ω

Therefore we obtain f1 (x, R1 (U )) = f1 (x, R1 (U + )), hence R1 (U ) = R1 (U + ). Eventu-

ally we deduce, using the equations satisfied by U and U + again, the relation U = U + .
The same proof works in the case of Robin boundary conditions. ⊓ ⊔

With the above Proposition at hand, we complete the picture by stating some
∗
properties of the pair (Ru (c1 ), U ∗ (c1 )). We begin with the asymptotic behaviour as
c1 → ∞.

Proposition 4.5 With the notation of Proposition 4.4, we have

∗

lim kRu (c1 )k∞ + kU ∗ (c1 ) − U∞ k∞ = 0.
c1 →+∞

where U∞ is the unique solution to A1 U = I in X+ .

Proof of Proposition 4.5.

Firstly, the function U∞ is an upper-solution to A1 U − c1 f1 (x, R1 (U )) U = 0 in
∗
X+ . Indeed, we clearly have, using the definition of U∞ and R1 (. . .), the relation
R1 (U∞ ) = 0, from which it follows that A1 U∞ − c1 f1 (x, R1 (U∞ )) U∞ = I ≥ 0.
 19

On the other hand, take an ε > 0 fixed. For c1 large enough, the function (1−ε)U∞
∗
is a lower-solution

to A1 U − c1 f1 (x, R1 (U )) U = 0 in X+ . Indeed, we have R1 (1 −
ε)U∞ = εK0 (I) = εS > 0 on Ω, so that

(1 − ε)A1 U∞ − c1 f1 (x, R1 ((1 − ε)U∞ )) (1 − ε)U∞

= (1 − ε) [I − c1 f1 (x, εS) U∞ ] < 0,

on Ω, whenever c1 is chosen large enough.

The maximum principle, as stated in Theorem 2.7, establishes that there is a
maximal pair (U − , U + ) of solutions to A1 U − c1 f1 (x, R1 (U )) U = 0, satisfying 0 <
(1 − ε)U∞ ≤ U − ≤ U + ≤ U∞ . Lastly, we readily know that U ∗ (c1 ) is the unique
positive solution of A1 U − c1 f1 (x, R1 (U )) U = 0 so that U − = U + = U ∗ (c1 ).
In particular, we recover (1 − ε)U∞ ≤ U ∗ (c1 ) ≤ U∞ . This shows lim kU ∗ (c1 ) −
c1 →+∞
U∞ k∞ = 0.
∗ ∗
Next, we observe that Ru (c1 ) satisfies A0 Ru (c1 ) + A1 U ∗ (c1 ) = I = A1 U∞ , so that
formula (2.9) provides

∗ a1 1  
Ru (c1 ) = − K0 A0 (U ∗ (c1 ) − U∞ ) + K0 (a1 m0 − a0 m1 )(U ∗ (c1 ) − U∞ ) .
a0 a0

(with the similar formula if the coefficients ai become space-dependent, with a1 (x) =
λ1 a0 (x) and a2 (x) = λ2 a0 (x) – see footnotes 2, 4 and 9). Using the fact that U ∗ (c1 ) ≤
∗
U∞ , Assumption 2, and, more precisely, relations (2.10) and (2.11), give 0 ≤ Ru (c1 ) ≤
a1 ∗ ∗
(U∞ − U (c1 )). Using the established limiting behaviour of U (c1 ) we deduce
a0
∗
lim kRu (c1 )k∞ = 0. ⊓ ⊔
c1 →+∞

The next result is a monotonicity property.

Proposition 4.6 With the notation of Proposition 4.4 the map c1 7→ U ∗ (c1 ) is in-
creasing from (c01 , +∞) to X+
∗ ∗
, while the map c1 7→ Ru (c1 ) is decreasing from (c01 , +∞)
∗
to X+ .

Proof of Proposition 4.6.

Take b2 > b1 > c01 . For i = 1, 2 the function U ∗ (bi ) is the only solution in X+ to


A1 U ∗ (bi ) − bi f1 x, R1 (U ∗ (bi )) U ∗ (bi ) = 0.

We observe that



A1 U ∗ (b1 ) − b2 f1 x, R1 (U ∗ (b1 )) U ∗ (b1 ) = (b1 − b2 )f1 x, R1 (U ∗ (b1 )) U ∗ (b1 ) < 0,

hence U ∗ (b1 ) is a lower-solution to A1 U − b2 f1 (x, R1 (U )) U = 0 in X+ . On the other

hand, we have already established that U∞ > U ∗ (b1 ) is an upper-solution as well.
Hence the maximum principle, as stated in Theorem 2.7, allows to conclude that there
exists a solution U e (b2 ) to A1 U − b2 f1 (x, R1 (U )) U = 0 in X which satisfies U (b1 ) <
e (b2 ) ≤ U∞ . The uniqueness we proved in Proposition 4.4 then provides U
U e (b2 ) =
U (b2 ). Therefore we have U ∗ (b1 ) < U ∗ (b2 ).
(1)
From this we deduce, using the already observed fact that R1 (U ∗ (bi )) ≡ RU ∗ (b ) ,
i
(by definition of the various objects), and using Theorem 2.11 part (iii), the relation
∗ ∗
Ru (b1 ) > Ru (b2 ). This ends the proof. ⊓ ⊔
20

4.3 Non-degeneracy and index of the semi-trivial solutions

The previous paragraph, and more precisely Proposition 4.4 shows that two families
of solutions to the one-species problem (2.18) coexist whenever c1 > c01 , namely the
∗
trivial (c1 , S, 0) and the semi-trivial (c1 , Ru (c1 ), U ∗ (c1 )). As an immediate consequence,
0 2
it appears that (c1 , S, 0) ∈ R ×(X+ ) is a bifurcation point for system (2.18). Note that
∗
the bifurcation solution (c1 , Ru (c1 ), U ∗ (c1 )) is readily constructed for all values c1 >
c01 , without using the Crandall-Rabinowitz theorem, so that it is not even clear that
∗
the branch (c1 , Ru (c1 ), U ∗ (c1 )) actually coincides with a bifurcation in the Crandall-
Rabinowitz sense (for instance, the limit as c1 → c01 of (Ru ∗
(c1 ), U ∗ (c1 )) may well differ
from (S, 0) at this stage).
In this section, we show essentially two results. On the one hand we show that
the Crandall-Rabinowitz theorem applies, and uniqueness allows to conclude that the
∗
already constructed semi-trivial solution (c1 , Ru (c1 ), U ∗ (c1 )) coincides with the one
obtained by bifurcation. On the other hand, and as a consequence, we deduce various
properties such as the non-degeneracy of the semi-trivial branch, or we compute the
index of this branch. This part of the analysis prepares for the next section where we
construct coexistence solutions to the full 2-species problem.
We begin with the
Proposition 4.7 (Local bifurcations in the one-species problem (2.18))
∗
With the above notation, let φ0 ∈ X+ and c01 > 0 be as in Lemma 4.2. Define
0 ∗
ρ0 = c1 K0 (f1 (S)φ0 ) ∈ X+ . On the other hand, recall from (2.19) the definition

T1 (c1 , R, U ) = t (K0 (I − c1 f1 (x, R) U ) , K1 (c1 f1 (x, R)U )) .

Then, the following holds

(i) The point (c01 , S, 0) is a bifurcation point for T1 , in that Theorem 2.9 applies.
b(s)) ∈ C 0 ((−ε, ε); R×X 2 ),
In particular, there exists ε > 0, and a map (c1 (s), rb(s), u
0 20
with c1 (0) = c1 , rb(0) = ub(0) = 0, such that the branch
n o
b(s)) ∈ c01 × (X+
(c1 (s), S − s(ρ0 + rb(s)), s(φ0 + u ∗ 2
) ; 0<s<ε

is a familly of positive solutions to (2.18). We set R(s) = S − s(ρ0 + rb(s)) and U (s) =
s(φ0 + u b(s)).
Moreover, each solution (c1 , R, U ) ∈ R ×(X+ )2 to (2.18) near (c01 , 0, 0) is either the
trivial solution (c1 , S, 0), or it coincides with (c1 (s), R(s), U (s)) for some s ∈ (−ε, ε).
In particular, for any c1 > c01 , close to c01 , there exists s > 0, such that
∗
(c1 , Ru (c1 ), U ∗ (c1 )) = (c1 (s), R(s), U (s)).

(ii) If s > 0 is small enough, we have

i T1 (c1 (s), ·), (R(s), U (s)) = 1.


Hence for c1 > c01 close to c01 we have21 i T1 (c1 , ·), (Ru
∗
(c1 ), U ∗ (c1 )) = 1.
20 Only positive values of the parameter s are retained. This is due to the fact that we only

keep track of positive solutions to system (2.18)

21 Our proof also shows that all eigenvalues of Id−D
(R,U ) T1 (c1 , Ru (c1 ), U (c1 )) are less than
∗ ∗
0 0
one when c1 > c1 is close to c1 , hence the corresponding solution (Ru (c1 ), U ∗ (c1 )) is stable
∗

for the time-dependent parabolic problem associated with the present stationary problem.
 21

∗
Remark 4.8 Point (i) establishes that the branch (c1 , Ru (c1 ), U ∗ (c1 )) constructed so
far coincides at least locally with the bifurcation branch (c1 (s), R(s), U (s)).
Point (ii) plays a crucial rôle later in the analysis, when exhibiting coexistence solu-
tions to the full 2-species system. We stress the fact that the computation of the above
index uses tools from bifurcation theory, hence relies on the identification between the
∗
bifurcation branch (c1 (s), R(s), U (s)) and the branch (c1 , Ru (c1 ), U ∗ (c1 )). ⊓
⊔

Proof of Proposition 4.7 - Point (i).

System(2.18) is equivalent to T1 (c1 , R, U ) = t (R, U ), where the (compact, con-
tinuous and twice Fréchet differentiable) operator T1 : R × X 2 → X 2 X, defined in
(2.19), is 22 T1 (c1 , R, U ) = t (S, 0) + c1 t (−K0 (f1 (R)U ), K1 (f1 (R)U )) . We have, for
any c1 ∈ R, the relation T1 (c1 , S, 0) = t (S, 0), which defines the trivial solution to
T1 (c1 , R, U ) = t (R, U ). Lastly, we define

L1 (c1 ) = Id − D(R,U ) T1 (c1 , S, 0). (4.3)

With these notations at hand, we show that the Crandall-Rabinowitz Theorem 2.9
applies at the bifurcation point (c01 , S, 0).
Firstly, let (ρ, φ) ∈ Ker(L1 (c01 )). We have

ρ + c01 K0 (f1 (S)φ) = 0 and φ − c01 K1 (f1 (S)φ) = 0.

If φ ≡ 0, then ρ ≡ 0. Hence φ 6≡ 0 verifies A1 φ − c01 f1 (S)φ = 0. By Lemma

4.2 we recover, up to a multiplicative constant, the two relations φ = φ
0 > 0 and
ρ = −c01 K0 (f1 (S)φ0 ) := −ρ0 < 0. This establishes dim Ker(L1 (c01 )) = 1 and
Ker(L1 (c01 )) = span(−ρ0 , φ0 ). Next, since L1 (c01 ) is a
compact perturbation of the
identity, its Fredholm index is 0 and codim Im(L1 (c01 )) = 1. Lastly, there remains to


/ Im L1 (c01 ) . A direct computation shows
prove the relation Dc1 L1 (c01 ) · (−ρ0 , φ0 ) ∈

Dc1 L1 (c01 ) · (−ρ0 , φ0 ) = t (+K0 (f1 (S)φ0 ), −K1 (f1 (S)φ0 )).


Arguing by contradiction, if Dc L1 (c01 ) · (−ρ0 , φ0 ) ∈ Im L1 (c01 ) , there exists φ and ρ
in X such that
 
(+K0 (f1 (S)φ0 ) , −K1 (f1 (S)φ0 )) = ρ + c01 K0 (f1 (S)φ) , φ − c01 K1 (f1 (S)φ) .

Applying A1 to the second equation, multiplying by Zφ0 , integrating23 over Ω, and

using the fact that A1 φ0 − c01 f1 (S)φ0 = 0, leads to f1 (S)φ20 = 0 so φ0 = 0, a
Ω
contradiction.
We have established that Theorem 2.9 may be applied in the present situation,
which guarantees the existence of the branch (c1 (s), R(s), U (s)). The uniqueness part
∗
of Proposition 4.4 ensures the identification of this branch with (c1 , Ru (c1 ), U ∗ (c1 ))
0
whenever s > 0 and c1 > c1 . ⊓⊔

22 Here and below we abuse notation by writing f1 (R) instead of f1 (x, R) and so on.
23 Robin boundary conditions lead to the same calculation.
22

Proof of Proposition 4.7 - Point (ii)

This proof is more delicate and uses more information from local bifurcation theory.
Since the local bifurcation Theorem of Crandall and Rabinowitz applies, it is known
(see [4, 20], see also [5] p. 179 for more details) that there exists two maps with C 1
smoothness,


s 7→ (µ(s), w(s)) = µ(s), ρ(s), φ(s)) ∈ R × X × X,
with µ(0) = 0, w(0) = (−ρ0 , φ0 ),


c1 7→ (γ(c1 ), w0 (c1 )) = γ(c1 ), ρ0 (c1 ), φ0 (c1 ) ∈ R × X × X,
with γ(c01 ) = 0, w0 (c01 ) = (−ρ0 , φ0 )

defined in the neighbourhood of s = 0, resp. c1 = c01 , such that along the trivial solution
(c1 , S, 0), we have
 
Id − D(R,U ) T1 (c1 , S, 0) · w0 (c1 ) = γ(c1 ) w0 (c1 ),

while along the semi-trivial solution (c1 (s), R(s), U (s)) we have
 
Id − D(R,U ) T1 (c1 (s), R(s), U (s)) · w(s) = µ(s)w(s),


In order to prove that i T1 (c1 (s), ·), (R(s), U (s)) = 1 for small values of s > 0,
we now show that D(R,U ) T1 (c1 (s), R(s), U (s)) has no eigenvalue greater than one (see
equation (2.15)), i.e. all eigenvalues of Id − D(R,U ) T1 (c1 (s), R(s), U (s)) are positive.
Since µ(s) is the smallest eigenvalue of Id − D(R,U ) T1 (c1 (s), R(s), U (s)) (thanks to
Lemma 2.1, and using the value of the above operator together with the fact that the
components of w(0) = (−ρ0 , φ0 ) are uniformly negative resp. positive on Ω, so that
the same property holds for the components of w(s) = (−ρ(s), φ(s)), at least for small
values of s), we therefore need to show µ(s) > 0 for small values of s > 0.
To do so we use the following known fact from local bifurcation theory, (see [5] p.
179), namely

µ(s) = −sγ ′ (c01 )c′1 (s) + o(s) as s → 0. (4.4)

This is the key piece of information here. There remains to study the signs of the various
terms on the right-hand-side of (4.4). Concerning c′1 (s), if s > 0 is small enough, we
have, by definition of c1 (s), the relation



b(s)) = s c1 (s) f1 S − s(ρ0 + rb(s))
s A1 (φ0 + u b(s) .
φ0 + u

d
Dividing by s and computing ds |s=0 , gives

b′ (0) = c01 f1 (S)u

A1 u b′ (0) + c′1 (0)f1 (S)φ0 − c01 DR f1 (S)φ0 ρ0 .

Multiplying by φ0 and integrating

Z over Ω, then provides,
Z using the fact that A1 φ0 −
c01 f1 (S)φ0 = 0, the relation c′1 (0) f1 (S)φ20 = c01 DR f1 (S)φ20 ρ0 . We recover c′1 (0) >
Ω Ω
0, hence c′1 (s) > 0 for small values s > 0. Concerning γ ′ (c01 ), we start from the relation,
valid whenever c1 is close to c01 ,

A1 φ0 (c1 ) − c1 f1 (S)φ0 (c1 ) = γ(c1 )A1 φ0 (c1 ).

 23

Applying d
dc1 |c1 =c01 , multiplying by φ0 = φ0 (c01 ), using γ(c01 ) = 0, and integrating over
24
Ω leads to Z Z
− f1 (S) φ20 = +γ ′ (c01 ) f1 (S)φ20 .
Ω Ω

Hence γ ′ (c01 ) < 0. Eventually we have established
that µ(s) > 0 whenever s > 0 is
small. This provides i T1 (c1 (s), ·), (R(s), U (s)) = 1 whenever s > 0 is small.
The proof is complete. ⊓ ⊔

The following is an obvious consequence of Theorem 4.7.

Proposition 4.9 With the notation of Proposition 4.4, we have

lim kR∗ (c1 ) − Sk∞ + kU ∗ (c1 )k∞ = 0.

c1 →c01

Proof of Proposition 4.9. Using Theorem 4.7, together with the uniqueness statement
of Theorem 4.4, we have (R(s), U (s)) = (R∗ (c1 (s)), U ∗ (c1 (s))). Since lim c1 (s) = c01 ,
s→0
the result follows from the continuity of s 7→ (R(s), U (s)). ⊓
⊔

The next Proposition is independent from the previous considerations. It states

that each semi-trivial solution (R∗ (c1 ), U ∗ (c1 )) is non-degenerate.

Proposition 4.10 With the notation of Proposition 4.4, for each c1 > c01 , we have
∗

Ker Id − D(R,U ) T1 (c1 , Ru (c1 ), U ∗ (c1 )) = {0}.

Proof of Propostion 4.10.

The proof is by contradiction.
Take c1 > c01 and assume 0 is an eigenvalue of Id − D(R,U ) T1 (c1 , Ru
∗
(c1 ), U ∗ (c1 )).
We define, for each u ∈ X, the following auxiliary operator, acting on X. Taking a
large, fixed number K > 0 , we introduce

u ∈ X 7→ H(u) := (A1 + K)−1 [f1 (R1 (U )) U + KU ] ∈ X (4.5)

where A0 R1 (u) + A1 u = I as usual (see (2.9)). Up to the introduction of the terms

involving K, the function H is essentially the second component of T1 , evaluated at
(R1 (U ), U ). From the definition of H, the following equivalence is clear whenever U ∈
∗
X+ , namely

T1 (R, U ) = t (R, U ) ⇔ [R = R1 (U ) and H(U ) = U.] (4.6)

Hence we readily have H (U ∗ (c1 )) = U ∗ (c1 ), and the equivalence (4.6) also implies,
∗
since 0 is an eigenvalue of Id − D(R,U ) T1 (c1 , Ru (c1 ), U ∗ (c1 )), that 1 is an eigenvalue
∗
of Du H(U (c1 )) as well.
We claim that the operator H is nondecreasing, i.e. whenever U and V belong to
X, we have

U ≥ V ≥ 0 =⇒ H(U ) ≥ H(V ). (4.7)

This property is actually the reason for our introduction of the parameter K. It comes
from the fact that, according to Lemma 2.3, from U ≥ V ≥ 0, we deduce f1 (R1 (U )) U −
24 The computation is the same in the case of Robin boundary conditions.
24

f1 (R1 (V )) V ≥ −γ(U − V ) hence f1 (R1 (U )) U − f1 (R1 (V )) V + K(U − V ) ≥ (K −

γ)(U − V ) ≥ 0, and the maximum principle allows to conclude.
Our second claim is

Du H(U ∗ (c1 )) · U ∗ (c1 ) = kU ∗ (c1 ), where k < 1. (4.8)

(Note that k is a function in X). This is the key ingredient. It comes from the following
computation. We have

d 

Du H U ∗ (c1 ) · U ∗ (c1 ) = H (1 + t)U ∗ (c1 )
dt t=0
d  


= t=0
(A1 + K)−1 f1 R1 (1 + t)U ∗ (c1 ) + K (1 + t) U ∗ (c1 )
dt



= (A1 + K)−1 f1 R1 U ∗ (c1 ) + K U ∗ (c1 )
 


d 

+ (A1 + K)−1 DR f1 R1 U ∗ (c1 ) U ∗ (c1 ) t=0
R1 (1 + t)U ∗ (c1 ) .
dt

On the other hand, we have





(A1 + K)−1 f1 R1 U ∗ (c1 ) + K U ∗ (c1 ) = H U ∗ (c1 ) ,

while
d 
d  
R1 (1 + t)U ∗ (c1 ) = t=0 K0 I − (1 + t) A1 U ∗ (c1 )
dt t=0 dt
d 

= − t=0 (1 + t)K0 c1 f1 R1 (U ∗ (c1 )) U ∗ (c1 )
dt 

= −K0 c1 f1 R1 (U ∗ (c1 )) U ∗ (c1 ) = R1 (U ∗ (c1 )) − S < 0.

Eventually we have established

Du H U ∗ (c1 ) · U ∗ (c1 ) = U ∗ (c1 ) + R1 (U ∗ (c1 )) − S =: k U ∗ (c1 ),

with k < 1 as claimed. This proves relation (4.8).

Our third claim is a consequence of the previous one. It somehow asserts that the
two functions (1 + ε)U ∗ (c1 ) and (1 − ε)U ∗ (c1 ) are lower-upper solutions to H(u) = u
in a strong sense. Namely, taking a (fixed) parameter µ > 0 such that

k + µ < 1.

We define

Hµ (u) := H(u) + µ(u − U ∗ (c1 )). (4.9)

We claim that whenever ε > 0 is small enough, we have

Hµ (1 − ε)U ∗ (c1 ) ≥ (1 − ε)U ∗ (c1 ), Hµ (1 + ε)U ∗ (c1 ) ≤ (1 + ε)U ∗ (c1 ). (4.10)

This comes from the following expansion

Hµ (1 + ε)U ∗ (c1 ) = Hµ U ∗ (c1 ) + εDu Hµ U ∗ (c1 ) · U ∗ (c1 ) + O(ε2 )
= U ∗ (c1 ) + ε (k + µ)U ∗ (c1 ) + O(ε2 )
≤ (1 + ε)U ∗ (c1 ),
 25

provided ε is small enough. We have used relation (4.8) together with the fact that
U ∗ (c1 ) > 0.
Gathering all the above claims, let us now show that Du H (U ∗ (c1 )) cannot have 1
as an eigenvalue. Take φ ∈ X (φ 6≡ 0) such that
Du H(U ∗ (c1 )) · φ = φ.
Up to rescaling φ, we may assume that
−U ∗ (c1 ) ≤ φ ≤ U ∗ (c1 ).
For technical reasons that become clear later, we may rescale φ again, so as to ensure
that there is a point x0 ∈ Ω such that
(1 + µ)φ(x0 ) > U ∗ (c1 )(x0 ),
where µ > 0 is as before. The idea is to compute Hµ (U ∗ (c1 ) + εφ) in two different
ways, to obtain the desired contradiction.
On the one hand we have, from the relation (1 − ε)U ∗ (c1 ) ≤ U ∗ (c1 ) + εφ ≤
(1 + ε)U ∗ (c1 ), and using (4.10), the bounds



Hµ U ∗ (c1 ) + εφ ≤ Hµ (1 + ε)U ∗ (c1 ) ≤ (1 + ε)U ∗ (c1 ),
as well as Hµ (U ∗ (c1 ) + εφ) ≥ Hµ ((1 − ε)U ∗ (c1 )) ≥ (1 − ε)U ∗ (c1 ). On the other hand,
we may expand (the expansion holds in X)




Hµ U ∗ (c1 ) + εφ = Hµ U ∗ (c1 ) + εDu Hµ U ∗ (c1 ) · φ + O(ε2 )
= U ∗ (c1 ) + (1 + µ)εφ + O(ε2 )
= (1 + ε) U ∗ (c1 ) + ε((1 + µ)φ − U ∗ (c1 )) + O(ε2 ).
Hence, at the point x0 , we have Hµ (U ∗ (c1 ) + εφ) (x0 ) > (1 + ε) U ∗ (c1 )(0 ), provided ε
is small enough, which contradicts the fact that Hµ (U ∗ (c1 ) + εφ) ≤ (1 + ε)U ∗ (c1 ).
To summarize, the whole idea of our contradiction argument is that on the one
hand (1 + ε)U ∗ (c1 ) satisfies H(U ) < U in a strict fashion (as a consequence of (4.8)),
while the upper-lower solution technique, together with the fact that φ is associated
with the eigenvalue 1 of the linear part of H, imply that when perturbing U ∗ (c1 ) in the
direction φ, the function H must at the same time be almost constant in that direction
and it should decay in a strict fashion as well. ⊓ ⊔
∗
As an immediate consequence of the non-degeneracy of the solution (Ru (c1 ), U ∗ (c1 )),
together with the implicit function theorem, we deduce the
∗
Proposition 4.11 The map c1 7→ (Ru (c1 ), U ∗ (c1 )) is continuously differentiable from
(c01 , +∞) to X+
∗ ∗
× X+ .
Proof of Proposition 4.11.
∗
The pair (Ru (c1 ), U ∗ (c1 )) is defined by the equation
∗
T1 (c1 , Ru (c1 ), U ∗ (c1 )) = t (Ru
∗
(c1 ), U ∗ (c1 ).
∗
On the other hand, we have just proved that Id−D(R,U ) T1 (c1 , Ru (c1 ), U ∗ (c1 )) does not
admit 0 as an eigenvalue, while it is clear from the definition of T1 that the linearized
operator D(R,U ) T1 (c1 , R, U ) is compact for any value of (c1 , R, U ) ∈ R × X 2 . As a
∗
consequence, we have that Id − D(R,U ) T1 (c1 , Ru (c1 ), U ∗ (c1 )) is invertible, and the
local inversion Theorem applies. ⊓ ⊔
26

A key consequence is the following

Proposition 4.12 For any c1 > c01 , we have

∗
i(T1 (c1 , ·), (Ru (c1 ), U ∗ (c1 ))) = 1.

Proof of Proposition 4.12.

∗
Since Id − D(R,U ) T1 (c1 , Ru (c1 ), U ∗ (c1 )) is invertible, and a compact perturbation
of the identity, we have

∗
i(T1 (c1 , ·), (Ru (c1 ), U ∗ (c1 ))) = (−1)p(c1 ) ,

∗
where p(c1 ), is the number of eigenvalues of D(R,U ) T1 (c1 , Ru (c1 ), U ∗ (c1 )) that are
+ − 0
greater than 1. Now, take any c1 > c1 > c1 . By uniqueness of the solution to
∗ 2
T1 (c1 , R, U ) = (R, U ) in (X+ ) , for any c1 > c01 , we can choose a neighbourhood U of
the set {(c1 , Ru (c1 ), U (c1 )) ; c1 ∈ (c−
∗ ∗ + ∗ 2
1 , c1 )} in R × (X+ ) such that, if c1 ∈ (c1 , c1 ),
− +

no solution solution to T1 (c1 , R, U ) = (R, U ) exists on ∂U . The homotopy conserva-

tion (see e.g. [4]) shows that i(T1 (c1 , ·), (Ru ∗
(c1 ), U ∗ (c1 ))) is constant on (c− +
1 , c1 ). This
+ − 0 ∗ ∗
argument is valid for each c1 > c1 > c1 hence i(T1 (c1 , ·), (R (c1 ), U (c1 ))) is con-
stant on (c01 , +∞). We conclude using Proposition 4.7, part (ii), according to which
∗
i(T1 (c1 , ·), (Ru (c1 ), U ∗ (c1 ))) = 1 whenever c1 > c01 is close to c01 . This ends the proof.
⊓
⊔

5 Coexistence solutions

We now show the main result of this paper, namely we exhibit coexistence solutions to
∗ 3
the full 2-species system (1.2), i.e. solutions (R, U, V ) to (1.2) that lie in (X+ ) . Recall
that the system with 2 species reads, shortly,

 A0 R + c1 f1 (x, R)U + c2 f2 (x, R)V = I,
A1 U − c1 f1 (x, R)U = 0, (5.1)

A2 V − c2 f2 (x, R)V = 0,

5.1 Preliminary results

The following fact summarizes the work we have performed at this stage.

3
Proposition 5.1 The system (1.2) has the trivial solution (S, 0, 0) ∈ X+ . Besides,
0 ∗ ∗ 3
(i) if c1 > c1 , system (1.2) has the semi-trivial solution (Ru (c1 ), U (c1 ), 0) ∈ X+ .
0 ∗ ∗ 3
(ii) if c2 > c2 , system (1.2) has the semi-trivial solution (Rv (c2 ), 0, V (c2 )) ∈ X+ .
We denote these two families by

∗
Cu = {(c1 , c2 , Ru (c1 ), U ∗ (c1 ), 0), (c1 , c2 ) ∈ (c01 , +∞) × (c02 , +∞)},
Cv = {(c1 , c2 , Rv∗ (c2 ), 0, V ∗ (c2 )), (c1 , c2 ) ∈ (c01 , +∞) × (c02 , +∞)}.

Our first result in the direction of obtaining coexistence solutions to (5.1) is the
 27

Proposition 5.2 Let (c1 , c2 ) ∈ R2 . Assume that (R, U, V ) ∈ (X+ ∗ 3

) is a coexistence
solution to (5.1).
Then, the following holds:
(i) We necessarily have c1 > c01 and c2 > c02 .
∗
(ii) With the above notation, the function R − Ru (c1 ) (resp. R −Rv∗ (c2 )) either changes
sign on Ω, or it vanishes identically.
(iii) We have 0 < U < U ∗ (c1 ) and 0 < V < V ∗ (c2 ) (on Ω).

Proof of Proposition 5.2.

∗ 3
Let (R, U, V ) ∈ (X+ ) be a coexistence solution to (1.2).
Point (i).
By Theorem 2.11 we have R < S. Hence, as in the proof of Proposition 4.3, we
deduce that ci > c0i for i = 1, 2.
Point (ii).
We have A1 U − c1 f1 (R)U = A1 U ∗ (c1 ) − c1 f1 (Ru
∗
(c1 ))U ∗ (c1 ) = 0 with U > 0 and
∗
U (c1 ) > 0. Hence, by Lemma 2.1, we recover
∗
λ1 (A1 − c1 f1 (R)) = λ1 (A1 − c1 f1 (Ru (c1 ))) = 0.

Point (ii) therefore comes as a direct consequence of the fact that R 7→ f1 (R) in-
creases with R, from which it is deduced that R 7→ λ1 (A1 − c1 f1 (R)) decreases with
∗
R (Lemma 2.1). The function R − Ru (c1 ) cannot have constant sign on Ω, unless it
vanishes identically.
Point (iii).
We use a lower-upper solution method. Whenever u and v belong to X, denote by
R(u, v) the only solution in X to A0 R + A1 u + A2 v = I. With this notation at hand,
the function u = U is seen to satisfy the following, nonlinear, nonlocal, elliptic problem

A1 u − c1 f1 (R(u, V )) u = 0. (5.2)

We first claim that U is the only positive solution to (5.2). To prove this, we observe
that whenever M > 0 is large enough, the constant function u = M is an upper-solution
to (5.2). Indeed, it is clear that R(M, V ) ≥ 0 when M is large (for A0 (R(M, V )) ≤ 0
under these circumstances), from which it follows A1 M −c1 (f1 (R(M, V ))M ≥ m1 M ≥
0. The constant function u = 0 being clearly a lower-solution to (5.2), it follows that
there exist a maximal solution 0 ≤ U + ≤ M such that any solution u to (5.2) such
that 0 ≤ u ≤ M also satisfies 0 ≤ u ≤ U + . In particular, taking M > U , we deduce
0 ≤ U ≤ U +.
To prove that U = U + , we define for convenience R+ = R(U + , V ) and R =
R(U, V ). We clearly have25
Z   Z h i
0= A1 U + · U − A1 U · U + = c 1 f1 (R+ ) − f1 (R) U + U,
Ω Ω

which proves U + = U provided we establish R+ ≤ R. On the other hand, the function

r = R satisfies
A0 r + c1 f1 (r)U = I − c2 f2 (r)V, (5.3)
25 with the obvious adaptation in the case of Robin boundary conditions
28

while the function r = R+ satisfies

A0 r + c1 f1 (r)U + = I − c2 f2 (r)V.

Since U ≤ U + , we see that R+ is a lower-solution to (5.3). This implies, similarly to

the proof of the Theorem 2.11, that R+ ≤ R. Hence U + = U and U is the only positive
solution to (5.2).
Let s ∈ (0, 1), we now claim U ∗ (c1 ) resp. sU are (strict) upper resp. lower solutions to
(5.2). Indeed, on the one hand, we have


A0 R(U ∗ (c1 ), V ) − R(U ∗ (c1 ), 0) = −A2 V = −c2 f2 (R(U, V )) < 0,

so that R(U ∗ (c1 ), V ) < R(U ∗ (c1 ), 0). We deduce

A1 U ∗ (c1 ) − c1 f1 (R(U ∗ (c1 ), V )) U ∗ (c1 )

 
= c1 f1 (R(U ∗ (c1 ), 0)) − f1 (R(U ∗ (c1 ), V )) U ∗ (c1 ) > 0.

On the other hand, we have

A0 (R(sU, V ) − R(U, V )) = (1 − s)c1 f1 (R(U, V )) > 0,

so that R(sU, V ) > R(U, V ). We deduce

A1 (sU ) − c1 f1 (R(sU, V )) sU = c1 [f1 (R(U, V )) − f1 (R(sU, V ))] sU < 0.

Now, since inf Ω U ∗ (c1 ) > 0, one can choose s ∈ (0, 1) small enough such that sU <
e to (5.2) such that sU < U
U ∗ (c1 ) and it follows that there exists a solution U e < U ∗ (c1 )
(the inequalities being strict because sU and U ∗ (c1 ) are not true solution). Uniqueness
of the positive solution yields U e = U hence U < U ∗ (c1 ).
The same proof shows that V < V ∗ (c2 ). ⊓ ⊔

To conclude this section, we also state the following two Lemmas.

Lemma 5.3 Let c1 > c01 .

∗
Then the eigenvalue problem A2 ψ − µf2 (Ru (c1 ))ψ = 0 has a principal eigen-
∗
value c2 (c1 ) >
R 0 and a corresponding eigenfunction ψ ∗ (c1 ) > 0. We have c∗2 (c1 ) =
2 2
a2 ∇φ + m2 φ
min RΩ ∗ (c ))φ2
. In particular, there holds c∗2 (c1 ) > c02 .
1
φ∈H (Ω), φ6=0 f (Ru
Ω 2 1

Proof of Lemma 5.3.

We only need to prove the inequality c∗2 (c1 ) > c02 , which comes from the formulae
giving c∗2 (c1 ) resp. c02 , in conjunction with the maximum principle. ⊓⊔

Lemma 5.4 Let c1 > c01 be fixed.

Then, there exists cmax
2 = cmax
2 (c1 ) > c02 such that, if (R, U, V ) ∈ (X+
∗ 3
) is a
max
solution of (1.2), we necessarily have c2 < c2 .

Proof of Lemma 5.4. Let c1 > c01 be given fixed. We suppose by contradiction that
there exists a sequence of solutions (ck2 , Rk , Uk , Vk ) ∈ (c02 , +∞)×(X+
∗ 3
) with ck2 → +∞.
As in the proof of Lemma 4.1, from the relation A0 Rk + A1 Uk + A2 Vk = I we
deduce that for some α > 0 we have (α − ∆)(a0 Rk + a1 Uk + a2 Vk ) ≤ I (with the
obvious adaptation in the case of variable coefficients ai = ai (x), see the proof of
Lemma 4.1)), hence 0 ≤ a0 Rk + a1 Uk + a2 Vk ≤ M for some M ≥ 0 independent of k.
 29

We deduce that all functions Rk , Uk , and Vk are bounded in L∞ , uniformly in k. In

turn we recover that A0 Rk , A1 Uk , and A2 Vk are uniformly bounded in L∞ as well,
and a bootstrap argument shows that Rk , Uk , and Vk are uniformly bounded in some
C 2+β space (β > 0), hence converge towards some R∞ , U∞ , V∞ in X+ , say.
Vk
We claim that R∞ = 0. Indeed, we define the function vk := ck kV k
verifies
2 k ∞
∗
A2 vk = f2 (Rk )vk . It follows that vk converges in X+ to some nonnegative function
v∞ verifiyng A2 v∞ = f2 (R∞ )v∞ . If R∞ 6= 0 then v∞ > 0 which contradicts the fact
that kvk k∞ = c1k → 0. We recover Rk → 0 in X.
2
Now, the fact that Uk > 0 provides λ1 (A1 − c1 f1 (Rk )) = 0. We deduce 0 =
λ1 (A1 − c1 f1 (Rk )) → λ1 (A1 ) as k → ∞. The known fact λ1 (A1 ) > 0 provides the
contradiction. ⊓ ⊔

5.2 Proof of Theorem 2.14

Let us now come to the construction of coexistence solutions.

For a given value of c1 > c01 , we introduce the (compact, continuous, twice Fréchet
differentiable) operator T2 : (c02 , ∞) × X 3 → X 3 as
 
K0 (I − c1 f1 (R)U − c2 f2 (R)V )
T2 (c2 , R, U, V ) =  c1 K1 (f1 (R)U ) 
c2 K2 (f2 (R)V )
     
S −K0 (f1 (R)U ) −K0 (f2 (R)V )
=  0  + c1  K1 (f1 (R)U )  + c2  0 . (5.4)
0 0 K2 (f2 (R)V )

Clearly (c2 , R, U, V ) ∈ (c02 , ∞) × (X+

∗ 3
) is a coexistence solution if and only if

T2 (c2 , R, U, V ) = t (R, U, V ).
∗
We readily know that the semi-trivial solution (c2 , Ru (c1 ), U ∗ (c1 ), 0) satisfies
∗
T2 (c2 , Ru (c1 ), U ∗ (c1 ), 0) = t (Ru
∗
(c1 ), U ∗ (c1 ), 0),

for any value of c2 . We now construct coexistence solutions using bifurcations from
∗
the (family of) point(s) (c∗2 (c1 ), Ru (c1 ), U ∗ (c1 ), 0), where c∗2 (c1 ) > c02 is provided by
Lemma 5.3.

Proposition 5.5 Take c1 > c01 . Let c∗2 = c∗2 (c1 ) > c02 be the eigenvalue defined in
Lemma 5.3 and ψ ∗ = ψ ∗ (c1 ) ∈ X+ ∗
be the associated eigenfunction.
∗ ∗ ∗
Then (c2 (c1 ), Ru (c1 ), U (c1 ), 0) is a bifurcation point for T2 , in that the local bi-
furcation Theorem 2.9 applies.
In particular, there exists ρ∗ = ρ∗ (c1 ) ∈ X and ∗ ∗

φ = φ (c1 ) ∈ X, there exists
1 3
ε > 0, there exists a map (re, u e, ve) ∈ C (−ε, ε), X verifying re(0) = u e(0) = ve(0) = 0,


together with a map c2 ∈ C 1 (−ε, ε), R+ verifying c2 (0) = c∗2 (c1 ), such that the
following holds. The branch
n  o
e
c2 (s), R(s), e (s), Ve (s) ; 0 < s < ε
U
30

is a family of positive solutions to (5.1), where we set

e = Ru∗ (c1 ) + s ρ∗ (c1 ) + re(s) ,
R(s) e (s) = U ∗ (c1 ) + s φ∗ (c1 ) + u
U e(s) ,


Ve (s) = s ψ ∗ (c1 ) + ve(s) .

Moreover, any solution (c2 , R, U, V ) ∈ R × X 3 to (5.1) near the bifurcation point

(c∗2 (c1 ), Ru
∗
(c1 ), U ∗ (c1 ), 0) is either the semi-trivial solution (c2 , Ru
∗
(c1 ), U ∗ (c1 ), 0), or
it coincides for some s ∈ (−ε, ε) with (c2 (s), R(s), e Ue (s), Ve (s)).

Proof of Proposition 5.5..

Recall that the value of c1 > c01 is fixed. We set
∗
L2 (c2 ) = Id − D(R,U,V ) T2 (c2 , Ru (c1 ), U ∗ (c1 ), 0). (5.5)
Using again the operator T1 of the one species problem, see (2.19), we have, whenever
(ρ, φ, ψ) ∈ X 3 , the relation
L2 (c∗2 (c1 )) · t (ρ, φ, ψ) = t (ρ, φ, ψ)
    ∗

∗ ∗ ρ −K0 (f2 (Ru (c1 ))ψ)
−  D(R,U ) T1 (c1 , Ru (c1 ), U (c1 )) · φ ∗
 − c2 (c1 )  0  . (5.6)
∗
0 K2 (f2 (Ru (c1 ))ψ)
Take now (ρ, φ, ψ) ∈ Ker (L2 (c∗2 (c1 ))). We have
∗



Id − D(R,U ) T1 (c1 , Ru ∗
(c1 ), U ∗ (c1 )) · t (ρ, φ) = t c∗2 (c1 )K0 f2 (Ru (c1 ))ψ , 0 , (5.7)
ψ − c∗2 (c1 ) K2 (f2 (Ru
∗
(c1 ))ψ) = 0. (5.8)
Equation (5.8) on ψ, and the definition of c∗2 (c1 ),
implies that ψ = ψ (c1 ) > 0 up to ∗

a multiplicative constant. Equation (5.7) on (ρ, φ), together with the already proved
∗
invertibility of Id − D(R,U ) T1 (c1 , Ru (c1 ), U ∗ (c1 )) (see Proposition 4.10), then provides
∗ ∗
(ρ, φ) = (ρ (c1 ), φ (c1 )), where we have set
t
(ρ∗ (c1 ), φ∗ (c1 )) := (5.9)
∗
−1 t


Id − D(R,U ) T1 (c1 , Ru (c1 ), U ∗ (c1 )) c∗2 (c1 )K0 ∗
f2 (Ru (c1 ))ψ ∗ (c1 ) , 0 .
Hence Ker (L2 (c∗2 (c1 ))) = span(ρ∗ (c1 ), φ∗ (c1 ), ψ ∗ (c1 )) and dim (Ker (L2 (c∗2 (c1 )))) = 1.
The Fredholm alternative also provides codim (Im (L2 (c∗2 (c1 )))) = 1.
There remains to show that


Dc2 L2 (c∗2 (c1 )) · t (ρ∗ (c1 ), φ∗ (c1 ), ψ ∗ (c1 )) ∈
/ Im L2 (c∗2 (c1 )) . (5.10)
We clearly have
Dc2 L2 (c∗2 (c1 )) · t (ρ∗ (c1 ), φ∗ (c1 ), ψ ∗ (c1 )) =
∗
t
(−K0 (f2 (Ru (c1 ))ψ ∗ (c1 )), 0, −K2 (f2 (Ru
∗
(c1 ))ψ ∗ (c1 ))).
If relation (5.10) is false, we can find ψ1 such that
∗
−K2 (f2 (Ru ∗
(c1 ))ψ ∗ (c1 )) = ψ1 − c∗2 (c1 ) K2 (f2 (Ru (c1 ))ψ1 ).
Z
∗
2
As in the proof of Proposition 4.7, we get f2 (Ru (c1 )) ψ ∗ (c1 ) = 0, which contra-
∗
dicts ψ (c1 ) > 0.
Eventually we have proved that the local bifurcation Theorem 2.9 applies, and the
Proposition follows. ⊓
⊔
 31

Next, a global argument provides the

Proposition 5.6 Let c1 > c01 be fixed. Then, equation (5.1) admits a continuum of
nontrivial solutions

C0 = {(c2 , R, U, V ))} ⊂ R × X 2 × (X \ {0}) ,

whose closure C0 joins the bifurcation point (c∗2 (c1 ), Ru ∗

(c1 ), U ∗ (c1 ), 0) either to ∞ in
3 ∗
R × X , or to a point (cb2 , Ru (c1 ), U (c1 ), 0) in R × X , where cb2 6= c∗2 (c1 ) is such that
∗ 3
∗
Id − D(R,U,V ) T2 (cb2 , Ru (c1 ), U ∗ (c1 ), 0) is not invertible.

Proof of Proposition 5.6. We apply the global bifurcation Theorem 2.10. It suffices to
∗
show that i(T2 (c2 , ·), (Ru (c1 ), U ∗ (c1 ), 0)) actually changes sign when crossing the value
∗
c2 = c2 (c1 ).
∗
Let µ > 1 be an eigenvalue of D(R,U,V ) T2 (c2 , Ru (c1 ), U ∗ (c1 ), 0) = Id − L2 (c2 ).
There exists (ρ, φ, ψ) 6≡ (0, 0, 0) such that (Id − L2 (c2 )) t (ρ, φ, ψ) = µ t (ρ, φ, ψ).
If ψ = 0, we recover, using relation (5.6), that
∗

Id − D(R,U ) T1 (c1 , Ru (c1 ), U ∗ (c1 )) t
(ρ, φ) = µ t (ρ, φ).
∗


Hence µ > 1 is an eigenvalue of Id − D(R,U ) T1 (c1 , Ru (c1 ), U ∗ (c1 )) . We know from
Proposition 4.12 that such µ’s are in even number.
∗
If ψ 6= 0, we recover using relation (5.6), that A2 ψ − c2 f2 (Ru (c1 ))ψ = (1 − µ)A2 ψ,
which means,
c
A2 ψ − 2 f2 (Ru ∗
(c1 ))ψ = 0. (5.11)
µ
Thanks to Lemma 5.3, it becomes clear that the above problem has no nontrivial
solution ψ 6≡ 0 whenever c2 ≤ c∗2 (c1 ), while it has exactly one nontrivial solution (up
to a multiplicative constant), namely ψ ∗ (c1 ), whenever c2 > c∗2 (c1 ) is close enough to
c∗2 (c1 ). This establishes
∗
i(T2 (c2 , ·), (Ru (c1 ), U ∗ (c1 ), 0)) = 1, if c2 < c∗2 (c1 ),
∗
i(T2 (c2 , ·), (Ru (c1 ), U ∗ (c1 ), 0)) = −1, if c2 > c∗2 (c1 ).

The theorem 2.10 is proved. ⊓

⊔

At this stage we have exhibited the continuum of nontrivial solutions C0 ⊂ R ×

X 2 × (X \ {0}). We need to select positive solutions (i.e. coexistence solutions) out of
C0 .
Close to the bifurcation point (c∗2 (c1 ), Ru
∗
(cn ∗
1 ), U (c1 ), 0), the only solutions
 that be-
o
∗ 3
long to (X+ ) necessarily belong to the branch e
c2 (s), R(s), e (s), Ve (s) ; 0 < s < ε ,
U
as stated in Proposition 5.5. To transform this construction into a global one, we now
define

C0+ is the closure of the maximal connected componant of

n  o
C0 \ e
c2 (s), R(s), e (s), Ve (s) ; −ε < s < 0 .
U (5.12)

The question we need to address now is whether C0+ \ {(c∗2 (c1 ), Ru

∗
(c1 ), U ∗ (c1 ), 0)} ⊂
∗ 3 ∗ 3
R × (X+ ) . The following proposition states that this set cannot remain in R × (X+ )
globally.
32

Proposition 5.7 We have

C0+ \ {(c∗2 (c1 ), Ru

∗
(c1 ), U ∗ (c1 ), 0)} 6⊂ R × (X+
∗ 3
) .

Proof of Proposition 5.7.

We argue by contradiction. Assume that C0+ \ {(c∗2 (c1 ), Ru
∗
(c1 ), U ∗ (c1 ), 0)} ⊂ R ×
(X+ ) . The key point is, according to Rabinowitz [2], the subset C0+ of C0 satisfies
∗ 3

an alternative similar to the one satisfied by C0 , namely, one of the three following
situations occur:

(i) The set C0+ joins (c∗2 (c1 ), Ru ∗

(c1 ), U ∗ (c1 ), 0) to (b ∗
c2 , Ru (c1 ), U ∗ (c1 ), 0) where Id −
∗
D(R,U,V ) T2 (b
c2 , Ru (c1 ), U ∗ (c1 ), 0) is not invertible and b c2 6= c∗2 .
(ii) The set C0+ joins (c∗2 (c1 ), Ru ∗
(c1 ), U ∗ (c1 ), 0) to ∞ in R × X 3 .
(iii) There exists (c2 , R, U, V ) in C0+ , such that, writing

∗
(c2 , R, U, V ) = (c2 , Ru (c1 ) + r, U ∗ (c1 ) + u, v),

∗
with (r, u, v) 6= (0, 0, 0), the symmetric point (c2 , Ru (c1 )) − r, U ∗ (c1 ) − u, −v)
+
belongs to C0 as well.

In the present contradiction argument, case (iii) cannot occur, nor can case (i)
occur. On top of that, take a point (c2 , R, U, V ) ∈ C0+ . Lemma 5.4 asserts that we
necessarily have c02 < c2 < cmax
2 (c1 ). Hence c2 remains in a fixed bounded subset of
R. Besides, the proof of Lemma 5.4 also asserts that (R, U, V ) necessarily belong to a
fixed compact subset of X 3 . Hence situation (ii) cannot occur.
This ends the proof. ⊓⊔

The above proposition asserts that C0+ necessarily leaves the positive cone. The
following Lemma provides information on the points where C0+ leaves the positive cone.

Lemma 5.8 Take c1 > c01 . Let (c2 , R, U, V ) ∈ R × (X+ )3 be the limit, in R × X 3 , of
∗ 3
a sequence of positive solutions (ck2 , Rk , Uk , Vk ) ∈ R × (X+ ) to (5.1). Then, we have

λ1 (A1 − c1 f1 (R)) = λ1 (A2 − c2 f2 (R)) = 0.

Proof of Lemma 5.8.

For all k ≥ 0, the function ψk = Uk kUk k−1X > 0 verifies A1 ψk − c1 f1 (Rk )ψk = 0.
Passing to the strong limit and using elliptic regularization provides a ψ ≥ 0, limit of
the ψk ’s, with kψkX = 1 and A1 ψ − c1 f1 (R)ψ = 0. Hence, Lemma 2.2 provides ψ > 0
and λ1 (A1 − c1 f1 (R)) = 0. The proof for λ1 (A2 − c2 f2 (R)) is similar. ⊓
⊔

The maximum principle now implies the following proposition.

Proposition 5.9 Take c1 > c01 .

Then, there exists c∗∗ 0
2 (c1 ) > c2 , such that

∗
C0+ joins (c∗2 (c1 ), Ru (c1 ), U ∗ (c1 ), 0)
to (c∗∗ ∗ ∗∗ ∗ ∗∗
2 (c1 ), Rv (c2 (c1 )), 0, V (c2 (c1 ))).
 33

Proof of Proposition 5.9.

Define for convenience C˙0+ := C0+ \ {(c∗2 (c1 ), Ru ∗
(c1 ), U ∗ (c1 ), 0)}.
In the neighbourhood of (c2 (c1 ), Ru (c1 ), U (c1 ), 0), we anyhow have C˙0+ ⊂ R ×
∗ ∗ ∗
∗ 3
(X+ ) .
On the other hand, by Proposition 5.7, there exists (cb2 , R, b Ub , Vb ) in the set C˙+ ∩

0
∗ 3
R × ∂(X+ ) , which is the limit of a sequence of solutions (ck2 , Rk , Uk , Vk ) lying in


C˙0+ ∩ R × (X+ ∗ 3
) . In particular, (R, b Ub , Vb ) ∈ (X+ )3 satisfies (5.1), hence for some
b b b
x ∈ Ω, we have R(x)U (x)V (x) = 0.
The maximum principle and the Hopf Lemma then assert that R b (resp. Ub , resp.
b
V ) cannot reach its minimal value 0 in Ω unless it is constant. In the case when
b ≡ 0, we recover I = 0, which is impossible. It follows that either U
R b ≡ 0 or Vb ≡ 0.
If Ub = Vb ≡ 0, then (cb2 , U b , Vb , R)
b is the trivial solution. By Lemma 5.8, this implies
that c1 is an eigenvalue of A1 φ − c1 f1 (S)φ = 0, hence that c1 = c01 . This contradicts
c1 > c01 . Now, suppose Vb ≡ 0 and U b > 0. Uniqueness of the semi-trivial solution
provides U b = U ∗ (c1 ) and R b = Ru∗ (c1 ). By Lemma 5.8, there exists ψ > 0 satisfying
∗
A2 ψ − cb2 f2 (Ru (c1 ))ψ = 0. Lemma 5.3 then provides the necessary relation cb2 = c∗2 (c1 )
which is again a contradiction. Eventually, the only possibility is Vb > 0, R b > 0 and
b b b b
U ≡ 0. Hence (c1 , cb2 , U , V , R) ∈ Cv . ⊓ ⊔
Theorem 2.14 is now a combinaison of Propositions 5.6 and 5.9.

5.3 Coexistence domain : proof of Theorem 2.16

Theorem 2.14 states that two families of coexistence solutions may be obtained, namely
the first one is constructed by freezing c1 > c01 and seeing c2 as a bifurcation parameter
to bifurcate from the semi-trivial (c∗2 (c1 ), Ru∗
(c1 ), U ∗ (c1 ), 0) where c∗2 (c1 ) > c02 , while
the second one is constructed by freezing c2 > c02 and seeing c1 as a bifurcation param-
eter to bifurcate from the semi-trivial (c∗1 (c2 ), Rv∗ (c1 ), 0, V ∗ (c2 )) where c∗1 (c2 ) > c01 .
This construction leads to defining the quantities c∗∗ 0 ∗∗ 0
2 (c1 ) > c2 and c1 (c2 ) > c1 . Note
∗∗ ∗ ∗∗ ∗ ∗∗ ∗
that the three situations c2 (c1 ) > c2 (c1 ), c2 (c1 ) < c2 (c1 ), c2 (c1 ) = c2 (c1 ) may very
well occur, and similarly for c∗∗ ∗
1 (c2 ) and c1 (c2 ).
Let us now exhibit some properties of the c∗i (cj )’s and c∗∗ i (cj )’s.
Lemma 5.10 For each c1 > c01 and c2 > c02 , we define
µ(c1 , c2 ) := λ1 (A1 − c1 f1 (Rv∗ (c2 ))), ∗
ν(c1 , c2 ) := λ1 (A2 − c2 f2 (Ru (c1 ))).
We have the relation (where sgn(s) = +1 if s > 0, = −1 if s < 0 and = 0 if s = 0)



sgn (µ(c1 , c2 )) = sgn c∗1 (c2 ) − c1 = −sgn c∗∗
2 (c1 ) − c2 ,



sgn (ν(c1 , c2 )) = sgn c∗2 (c1 ) − c2 = −sgn c∗∗
1 (c2 ) − c1 .

Proof of Lemma 5.10.

We show the result for µ. The proof for ν is similar.
Take c2 > c02 . The definition of c∗1 (c2 ) readily provides µ(c∗1 (c2 ), c2 ) = λ1 (A1 −
c1 (c2 )f1 (Rv∗ (c2 ))) = 0. On the other hand, Lemma 2.2 states that the map c1 7→
∗

µ(c1 , c2 ) is increasing. This shows that sgn (µ(c1 , c2 )) = sgn (c∗1 (c2 ) − c1 ).
Take c1 > c01 . The construction of the point (c∗∗ ∗ ∗∗ ∗ ∗∗
2 (c1 ), Rv (c2 (c1 )), 0, V (c2 (c1 ))),
together with Lemma 5.8, provide µ(c1 , c∗∗ 2 (c 1 )) = λ 1 (A 1 − c f
1 1 (R ∗ ∗∗
(c
v 2 (c 1 )))) = 0.
By Theorem 2.12, the map c2 7→ Rv∗ (c2 ) is decreasing, hence by Lemma 2.1, the map
c2 7→ µ(c1 , c2 ) is increasing. This shows sgn (µ(c1 , c2 )) = −sgn (c∗∗ 2 (c1 ) − c2 ). ⊓ ⊔
34

With this Lemma at hand, we may now prove the

Proposition 5.11 Let be {i, j} = {1, 2}. For all cj > c0j , the scalar c∗∗
i (cj ) is charac-
terized by 26
c∗i (c∗∗
j (ci )) = ci .

Proof of Proposition 5.11.

Take c1 > c01 . Set d2 = c∗∗ 2 (c1 ), a quantity that is characterized by the fact that
λ1 (A1 − c1 f1 (Rv∗ (d2 )) = 0, according to the previous Lemma. Now the quantity d1 :=
c∗1 (d2 ) is in turn characterized by the relation λ1 (A1 − d1 f1 (Rv∗ (d2 )) = 0, and the
previous relation shows d1 = c1 . This establishes c∗1 (c∗∗ 2 (c1 )) = c1 . The proof of the
relation c∗2 (c∗∗
1 (c2 )) = c2 is the same. ⊓ ⊔

The following Proposition is another consequence of the above Lemma.

Proposition 5.12 (i) The function c1 7→ c∗2 (c1 ) is continuous and increasing from
(c01 , +∞) to (c02 , +∞). The similar statement holds for c2 7→ c∗1 (c2 ).
(ii) We have lim c∗2 (c1 ) = +∞ and lim c∗1 (c2 ) = +∞.
c1 →∞ c2 →∞
(iii) We have lim c∗2 (c1 ) = c02 and lim c∗1 (c2 ) = c01 .
c1 →c01 c2 →c02

Proof of Proposition 5.12.

We prove only the properties concerning the map c2 7→ c∗1 (c2 ).
Take c2 > c02 . We have λ1 (A1 −c∗1 (c2 )f1 (Rv∗ (c2 ))) = 0. On the other hand, Theorem
2.12 asserts that the function c2 7→ Rv∗ (c2 ) is continuous and decreasing. Hence, from
Lemma 2.1 we deduce that c1 7→ c∗2 (c1 ) is continuous and increasing.
Now, we have that Rv∗ (c2 ) tends uniformly to 0 when c2 → ∞. If c∗1 (c2 ) remains
bounded as c2 → ∞, then λ1 (A1 − c∗1 (c2 )f1 (Rv∗ (c2 ))) = 0 → λ1 (A1 ) > 0 as c2 → ∞,
which is impossible. Therefore, we necessarily have c∗1 (c2 ) → ∞ as c2 → ∞.
Similarly, as Rv∗ (c2 ) tends uniformly to S as c2 → c02 , Lemma 5.3 provides the
relation limc2 →c0 c∗1 (c2 ) = c01 . ⊓
⊔
2

At this level of the analysis, one may define the three open sets

Θ+ = {c1 , c2 ) ∈ (c01 , +∞) × (c02 , +∞), c∗i (cj ) < ci < c∗∗
i (cj ), i 6= j},

Θ− = {c1 , c2 ) ∈ (c01 , +∞) × (c02 , +∞), c∗∗ ∗

i (cj ) < ci < ci (cj ), i 6= j}.
Θ = Θ− ∪ Θ+ .

It is clear that whenever (c1 , c2 ) ∈ Θ, a coexistence solution may be exhibited to (5.1).

Note however that these sets may be void. Note as well that our construction anyhow
exhibits coexistence solutions for some values of (c1 , c2 ), obtained by fixing c1 and
letting c2 vary, say: in that respect the set Θ may not exhaust all values of (c1 , c2 ) for
which a coexistence solution may be exhibited.
In any circumstance, the above results show that these sets have the simpler value

Θ+ = {c1 , c2 ) ∈ (c01 , +∞) × (c02 , +∞), c1 > c∗2 (c1 ), c2 > c∗1 (c2 )},
Θ− = {c1 , c2 ) ∈ (c01 , +∞) × (c02 , +∞), c1 < c∗2 (c1 ), c2 < c∗1 (c2 )}.

26 That is, c∗∗

i = (ci )
∗ −1 .
 35

For later convenience we also define

f+ = {c1 , c2 ) ∈ (c01 , +∞) × (c02 , +∞), c∗i (cj ) ≤ ci ≤ c∗∗

Θ i (cj ), i 6= j},
g 0 0 ∗∗ ∗
Θ − = {c1 , c2 ) ∈ (c1 , +∞) × (c2 , +∞), ci (cj ) ≤ ci ≤ ci (cj ), i 6= j}.
e=Θ
Θ g f+ .
−∪Θ

6 Interpretations, and ecological aspects

6.1 A conjecture

Conjecture
(i) If (c1 , c2 ) ∈ e then there cannot exist (R, U, V ) ∈ (X+
/ Θ, ∗ 3
) solution to (5.1).
(ii) We have Θ− = ∅, or, in other words, c∗i (cj ) ≤ c∗∗ i (c j whenever i 6= j.
)

This conjecture is motivated by our numerical simulations. It states that the set
e actually characterizes those values of (c1 , c2 ) for which a coexistence solution may
Θ
be exhibited. It also states that species i survives if and only if ci ≤ c∗i (cj ). In other
words, species i survives if and only if λ1 (Ai − ci fi (R∗ (cj )) ≥ 0.

6.2 Two ecological properties

Lemma 5.3 readily provides the following result.

Proposition 6.1 (dependence of the coexistence solutions on the diffusion

rates).
Take a0 and a1 in (0, +∞), and consider the system (5.1) as a function of the
diffusion rate a2 .
Then, the map a2 7→ c∗2 (c1 )(a2 ) is increasing.
∗
Moreover, if x 7→ m2 (x) − f2 (x, Ru (c1 )(x)) is not a constant function , then a2 7→
∗
c2 (c1 )(a2 ) is strictly increasing.

Provided the above conjecture holds, this assertion implies that as the diffusion
rate of a given species increases, its ability to survive decreases.

Proposition 6.2 (rôle of the heterogeneity).

∗
(i) If (c1 , c2 ) ∈ Θ, we necessarily have that Ru (c1 ) − Rv∗ (c2 ) is neither positive nor
negative.
∗
(ii) If Ru (c1 ) = Rv∗ (c2 ), then, for all i, j = 1, 2, i 6= j, we have c∗i (cj ) = c∗∗ i (cj ) = ci
and
∗
{(c1 , c2 , Ru , (1 − t)U ∗ , tV ∗ , t ∈ [0, 1]} ∈ {c1 , c2 } × X+ 3
.
∗
is a familly of solutions joining (c1 , c2 , Ru (c1 ), U ∗ (c1 ), 0) to (c1 , c2 , Rv∗ (c2 ), 0, V ∗ (c2 )).

In other words, the coexistence domain Θ is embedded in the set of the (c1 , c2 )
∗
such that Ru (c1 ) − Rv∗ (c2 ) is neither positive nor negative. This point highlights the
importance of the spatial heterogeneity in the coexistence process. This point in further
discussed in the next subsection.
36

Proof of Proposition 6.2.

If (c1 , c2 ) ∈ Θ, then µ(c1 , c2 )ν(c1 , c2 ) > 0. On the other hand, we know that
∗
λ1 (A1 − c1 f1 (Ru (c1 ))) = 0 and λ1 (A2 − c2 f2 (Rv∗ (c2 ))) = 0. Hence, if Ru ∗
(c1 ) ≥
∗ ∗
Rv (c2 ), then µ(c1 , c2 ) = λ1 (A1 − c1 f1 (Rv (c2 ))) > 0. Therefore, ν(c1 , c2 ) = λ1 (A2 −
∗
c2 f2 (Ru (c1 ))) > 0 as well. Hence λ1 (A2 − c2 f2 (Rv∗ (c2 ))) > 0, which is impossible. The
same arguments shows that Rv∗ (c2 ) ≥ Ru ∗
(c1 ) is impossible.
∗ ∗
Now, if Ru (c1 ) = Rv (c2 ) := R, then µ(c1 , c2 ) = ν(c1 , c2 ) = 0, hence ci = c∗i (cj ) =
c∗∗
i (cj ). One gets A0 R + c1 f1 (R)U = A0 R + c2 f2 (R)V thus, for all t ∈ [0, 1], we have
A0 R + (1 − t)c1 f1 (R)U ∗ + tc1 f1 (R)V ∗ = I. Since A1 U ∗ = c1 f1 (R)U ∗ and A2 V ∗ =
c2 f2 (R)V ∗ , we see that {(c1 , c2 , Ru
∗
, (1 − t)U ∗ , tV ∗ , t ∈ [0, 1]} is a family of solutions.
⊓
⊔

6.3 Two degenerate cases

In the homogeneous case where the functions I(x), fi (x), mi (x), ai (x) do not depend on
∗
x, and when Neumann boundary conditions are retained, we have that Ru (c1 )(x) and
∗
Rv (c2 )(x) are constant functions. Hence, by Proposition 6.2, the coexistence is possible
∗ ∗
only if Ru (c2 ) = Ru (c1 ), which induces a degenerate solution. Moreover, the fact that
∗ ∗
Ru (c1 ) and Rv (c2 ) decrease imply that meas{(c1 , c2 ) ∈ (c01 , +∞) × (c02 , +∞), Ru ∗
(c2 ) =
∗
Ru (c1 )} = 0. In that degenerate case we have the

Proposition 6.3 (The homogeneous case). Assume the problem is homogeneous,

i.e. I, mi and fi do not depend on x. Assume Neumann boundary conditions are re-
tained. 
Then we have c0i = inf ci > 0, fi−1 (mi /ci ) exists and is smaller than S := mI0 ,
∗
and the only semi-trivial solutions are (Ru (c1 ), U ∗ (c1 ), 0) and (Rv∗ (c2 ), 0, V ∗ (c2 )), where
∗
Ru (c1 ) = f1−1 (m1 /c1 ), Rv∗ (c2 ) = f2−1 (m2 /c2 ),
U ∗ (c1 ) = (I − m1 Ru
∗
(c1 ))/m0 , V ∗ (c2 ) = (I − m2 Rv∗ (c2 ))/m0 .
∗
In particular, either Ru (c1 ) > Rv∗ (c2 ), or Ru
∗
(c1 ) < Rv∗ (c2 ), or Ru
∗
(c1 ) = Rv∗ (c2 ).
In any circumstance, we have

Θ = ∅,
n o
e = (c1 , c2 ) ∈ (c01 , +∞) × (c02 , +∞) s.t. R1∗ (c1 ) = R2∗ (c2 ) < S
Θ
     
m1 m2
= (c1 , c2 ) ∈ (c01 , +∞) × (c02 , +∞) s.t. f1−1 = f2−1 <S .
c1 c2

e there exists a family of solutions {(R1∗ (c1 ), tU ∗ (c1 ), (1 −

Moreover, for all (c1 , c2 ) ∈ Θ,
t)V ∗ (c2 )), t ∈ [0, 1]}.

Another critical case appears when the two species possess heterogeneous but pro-
portional diffusion rates, mortality rate, and consumption rate, namely

Proposition 6.4 (Case of similar species). Suppose that f1 = f2 and A2 = αA1

for some constant α ∈ R∗+ .
Then, for all (c1 , c2 ) ∈ Θ, we have the four relations
∗
Ru (c1 ) = Rv∗ (c2 /α), Θ = ∅, Θ
e = {(c1 , αc1 ), c1 ≥ c01 }, c∗2 (c1 ) = αc1 .
 37

∗ 3
Moreover, the system has a coexistence solution (R, U, V ) ∈ (X+ ) if and only if
e ∗ ∗ ∗
(c1 , c2 ) ∈ Θ. In that case {(Ru , tU , (1 − t)U ), t ∈ (0, 1)} is a family of solutions
∗
and each coexistence solution satisfies (R, U, V ) ∈ {(Ru , tU ∗ , (1 − t)U ∗ ), t ∈ (0, 1)}.
∗
Proof of Proposition 6.4. The system defining (Ru (c1 ), U ∗ (c1 )) is
A1 U ∗ (c1 ) − c1 f1 (Ru
∗
(c1 ))U ∗ (c1 ) = 0, ∗
A0 Ru (c1 ) + A1 U ∗ (c1 ) = 0,
while the system defining (Rv∗ (c2 ), V ∗ (c2 )) is in the present case
αA1 V ∗ (c2 ) − c2 f1 (Rv∗ (c2 ))V ∗ (c1 ) = 0, A0 Rv∗ (c2 ) + αA1 V ∗ (c2 ) = 0.
1

c

The uniqueness result of Proposition 4.4 provides V ∗ (c2 ) = U ∗ 2 , and Rv∗ (c2 ) =
  α α
∗ c2
Ru . Now, since c∗2 (c1 ) is defined as the unique value of the parameter c2 such that
α ∗ ∗
λ1 (A2 − c2 f2 (Ru (c1 ))) = 0, i.e. λ1 (A1 − (c2 /α)f1 (Ru (c1 ))) = 0, it comes c∗2 (c1 ) = αc1 .
∗
This together with the analogous relation for c1 (c2 ) provides
Θ = ∅, e = {(c1 , αc1 ) ; c1 > c01 }.
Θ
Take now (c1 , c2 ) such that (R, U, V ) is an associated coexistence solution. We have
c
λ1 (A1 − 2 f1 (R)) = λ1 (A1 − c1 f1 (R)) = 0,
α
and monotone dependence of the above λ1 ’s with the parameters c1 and c2 implies
c2 = αc1 . Besides, summing the last two equations of (1.2) leads to

A0 R + c1 f1 (R)(U + αV ) = I,
(A1 − c1 f1 (R))(U + αV ) = 0,
so that uniqueness provides U +αV = U ∗ (c1 ), and R = Ru ∗
(c1 ). On top of that, coming
back to the equations satisfied by U resp. V , it appears that there exists (t, y) ∈ R2+
such that U = tU ∗ (c1 ), and αV = yU ∗ (c1 ). Gathering the relations then provides the
necessary equation t + y = 1. This ends the proof. ⊓ ⊔

6.4 Coexistence domain when diffusion rates tend to ∞

For a small value of ε > 0, we consider the system

 a

 (m0 − 0 ∆)R + c1 f1 (R)U + c2 f2 (R)V = I,

 ε

 a1
(m1 − ∆)U − c1 f1 (R)U = 0, (6.1)
 ε


 (m − a2 ∆)U − c f (R)V = 0,

 2 ε
2 2

with Neumann boundary condition27 It can be shown (see [25]), using the central
manifold theorem, that the solution to (6.1)converges to the solution of the so-called
aggregated system 
m f0 r + c1 fe1 (r)u + c2 fe2 (r)v = Ie
(mf − c1 fe1 (r))u = 0 (6.2)
 1
(mf2 − c2 fe2 (r))v = 0
27 This is the only place in this text where Neumann – and not Robin – boundary conditions

are required
38

R R R
where m 1
fi = |Ω| Ω
mi (x)dx, fei (r) = |Ω|
1
f (x, r), Ie = |Ω|
Ω i
1
Ω
I(x)dx, and the un-
known r, u, v now are scalars (independent of x). System (6.2) is a homogeneous
chemostat system. Therefore, and as is easily seen on the equations, in the generic case
there are no positive solution to (6.2). As it is proved in [25], it turns out that for
ε > 0 small enough, the original system (6.1) has no positive solution in the generic
case neither.
This result allows to describe the behavior of the coexistence domain Θ when the
diffusion rates tend to +∞. Remark in passing that, if Assumption 2 is true for a given
ε > 0, then it remains true for each ε > 0. In this case, Theorem 2.16 shows that there
exists Θε ⊂ R2+ such that, for each (c1 , c2 ) ∈ Θε , the system (6.1) admits a coexistence
solution. The boundaries of Θ fε are given by the curves

{(c1 , c∗,ε 0,ε

2 (c1 )) ; c1 > c1 } and {(c∗,ε 0,ε
1 (c2 ), c2 ; c2 > c2 }.

Define for convenience the quantity ri∗ (ci ) as

−1
ri∗ (ci ) = fei (mi /ci ) if this is well-defined, and ri∗ (ci ) = +∞ otherwise.

We have the
Proposition 6.5 Denote Θ∞ = {(c1 , c2 ) s.t. r1∗ (c1 ) = r2∗ (c2 ) < +∞}.
/ Θ∞ , there exists ε0 > 0 such that ∀ε ∈ (0, ε0 ), we have
Then, for each (c1 , c2 ) ∈
ε
(c1 , c2 ) ∈
/Θ .

/ Θ∞ , then r1∗ (c1 ) 6= r2∗ (c2 ). Therefore, the system

Proof of Proposition 6.5. If (c1 , c2 ) ∈
(6.2) has no positive solution and this implies ([25]) that there exists ε0 > 0 such that,
for all ε ∈ (0, ε0 ), the system (6.2) has no positive solution. Hence, (c1 , c2 ) ∈ / Θε if ε is
small enough. ⊓ ⊔

In this sense, the coexistence domain Θε tends to the curve Θ∞ when ε → 0. As

the diffusion rates increase, the aggregation phenomena leads to system that is close
to homogeneous in space, and the coexistence domain shrinks to a curve.

7 Conclusion and perspectives

This study examines a model where two species compete for a single resource, in a spa-
tially heterogeneous domain. Our system differs from the classical unstirred chemostat
system [13, 12, 11, 3, 18] in that the the reaction terms do depend on space, and, more
importantly, we allow the diffusion rates to depend on the species under consideration.
This point leads to a new mathematical difficulty. Namely, the conservation law which
links the resource R with the two species U and V , noted A0 R + A1 U + A2 V = I
in the core of the paper, becomes a nonlocal equation (as compared to the previously
quoted papers where the analogous equation is local). We circumvent this difficulty by
introducing Assumption 2 (supplemented with Assumption (1.6) in the case of Robin
boundary conditions).
We show that coexistence occurs when the consumption parameters (c1 , c2 ) lie in
a subdomain Θ ⊂ R2+ . In addition, we study the set Θ by using a characterisation of
Θ that relies on the two functions c∗1 (c2 ) and c∗2 (c1 ) defined in the text.
Several direction may extend this study. Firstly, our numerical observations indicate
that the coexistence solution are non-degenerate, except in the particular case when the
 39

two functions c∗1 (.) and c∗2 (.) coincide. When the coexistence solution is non-degenerate,
it turns out that our construction can be extended to three species, and by iteration,
to N species for any value of N . It would therefore be a key step to actually prove that
the coexistence solutions necessarily are non-degenerate, unless c∗1 (.) and c∗2 (.) coincide
Note in passing that Propositions 6.3 and 6.4 give two examples of situations where
c∗1 (.) and c∗2 (.) coincide, and a complete description of the coexistence phenomena is
provided in these situations.
Secondly, we defined Θ as the union of two subdomain Θ− and Θ+ . If (c1 , c2 ) ∈ Θ−
then c∗i > c∗∗i and the bifurcation occurs ”to the left” (see Figure 2.2). We conjecture
that Θ− = ∅ in any case. In fact, to rephrase our conjecture, if (c1 , c2 ) ∈ Θ− , then
both species are ”not invasive” in the sense that

λ1 (A1 − c1 f1 (Rv∗ (c2 )) < 0, and λ1 (A2 − c2 f2 (Ru

∗
(c1 )) < 0.

Note that Waltmann et al. [11] formulate a similar conjecture. Namely, they conjecture
that a necessary condition for two species to coexist is that both species are ”invasive”
in the sense that λ1 (A1 − c1 f1 (Rv∗ (c2 )) ≥ 0 and λ1 (A2 − c2 f2 (Ru
∗
(c1 )) ≥ 0. Lastly,
note that if (c1 , c2 ) ∈ Θ− , then the index of both semi-trivial solutions is equal to
1. To rephrase the above considerations, Waltmann et al. in [11] conjecture that a
necessary condition for two species to coexist is that both sem-trivial solution are
unstable (for the time-dependent problem), which in our case, means that the index
of the two semi-trivial solutions is equal to −1. Note that even if the latter result is
proved, it is not clear that the coexistence solution itself is stable. Indeed, Hofbauer
and So [17] show that there exists gradostats (that is, similar models with a discrete
spatial structuration) for which an unstable coexistence solution may be exhibited. A
more precise description of Θ would be a first step to understand the situation.
Thirdly, we conjecture that if (c1 , c2 ) ∈ e then no coexistence solution can be
/ Θ,
found. Would this result be proved, we could use Θ as a geometrical indicator of the
possibility of coexistence in a given system. Numerical investigations on the relation
between Θ, spatial heterogeneity, and the biodiversity, will be published soon.
Finally, our proof uses basically Assumption 2, an assumption that allows us to
extend the analysis of the (known) case where all diffusion operators coincide. It is
to be noted, however, that our approach proves the existence of semi-trivial solutions
without using Assumption 2. This assumption is only needed to obtain uniqueness and
non-degeneracy of the so-obtained semi-trivial solutions. A natural question is: can one
extend our construction to situations where Assumption 2 is not verified?

References

1. M. G. Crandall and P. H. Rabinowitz, Bifurcation from simple eigenvalue, J. Func. Anal.,

8 (1971), 321-340.
2. P. H. Rabinowitz, Some global results for nonlinear eigenvalue problems, J. Func. Anal.,
7 (1971), 487-513.
3. H. L. Smith and P. Waltman, The theory of the Chemostat, Cambridge university press,
1995.
4. A. Ambrosetti and A. Malchiodi, Nonlinear Analysis and Semilinear elliptic problems,
Cambrige university press, 2007.
5. J. Smoller, Shock Waves and Reaction-diffusion Equations, Springer-Verlag, 1993.
6. C.V. Pao, Quasisolutions and global attractor of reaction-diffusion systems, Nonlinear
Ana., Theory, Methods and Applications, 26, 1996, 1889-1903.
40

7. C.V. Pao, On nonlinear reaction-diffusion equations, J. Math. Analysis applic, 87, 1982,
165-198.
8. C. Walker, Coexistence Steady States in a Predator-Prey Model, Arch. Math., 95, 2010,
87-99.
9. C. Walker, Global Bifurcation of Positive Equilibria in Nonlinear Population Models, J.
Differential Equations, 248, 2010, 1756-1776.
10. Z. Zhang, Coexistence and Stability of Solutions for a Class of Reaction-Diffusion Systems,
E. J . Diff. Eq., 137, 2005, 1-16.
11. S. B. Hsu and P. Waltman, On a system of reaction-diffusion equations arising from
competition in an unstirred chemostat., SIAM J. Appl. Math, 53, 1993, 1026-1044.
12. S.B. Hsu, H. Smith and P. Waltman, Dynamic of competition in the unstirred chemostat.,
Can. Appl. Math. Quart., 2, 1994, 461-483.
13. J. L. Dung, H. L. Smith and P. Waltman,Growth in the unstirred chemostat with different
diffusion rates, Fields institute communications, 21, 1999, 131-142.
14. S. Zheng and J. Liu, Coexistence solutions for a reaction-diffusion system of un-sirred
chemostat model, Applied Math. and Comp., 145, 2003, 579-590.
15. S. Zheng and H. Guo and J. Liu, A food chain model for two resources in unstirred
chemostat, Applied Math. Comp., 206, 2008, 389-402.
16. J. V. Baxley and S. B. Robinson, Coexistence in the unstirred chemosat,
Applied Math. And Comput., 39, 1998, 41-65.
17. J. Hofbauer and J. W. H. So, Competition in the gradostat: The global stability problem,
Nonlinear analysis, 22, 1994, 1017-1033.
18. J. H. Wu, Global bifurcation of coexistence state for the competition model in the chemo-
stat, Nonlinear Analysis, 39, 2000, 817-835.
19. H. Nie and J. Wu: Uniqueness and stability for coexistence solutions of the unstirred
chemostat model, Applicable Analysis 89, 1151–1159 (2010)
20. J. Blat and K. J. Brown, Bifurcation of steady-state solutions in predator-prey and com-
petition systems, Proc. Roy. Soc. Edin.A, 97, 1984, 21-34.
21. J. Blat and K. J. Brown, Global bifurcation of positive solutions in some systems of elliptic
equations, SIAM J. Math. Anal., 17, 1986, 1339-1353.
22. Y. Du end K. J. Brown,Bifurcation and Monotonicity in Competition Reaction-Diffusion
Systems, Nonlinear Ana. Th. Meth. & Appl., 23 No. 1, 1994, 1-13.
23. E. D. Conway, Diffusion and The Predator-Prey Interaction: Steady States with Flux at
the Boundaries, Contemporary. Math., 17, 1983, 215-234.
24. E. Dancer, On positive solutions of some partial differential equations, Trans. Amer. Math.
Soc., 284, 1984,
25. S. Madec and F. Castella, Global behavior of N competiting species with strong diffusion:
diffusion leads to exclusion, to appear